Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesWebinar 3 ndash Safety Data Workshop
May 23 2014
Welcome and Objective
bull Thank you for participating
bull Todayrsquos Objectivesndash Provide in-depth
information for IAP applications
ndash Allow questions to research data experts
ndash Answer questions on application process which opens May 27
Agenda
Noon (ET) Welcome Introduction Purpose Rudy Malfabon and Sandra Larson
1210 SHRP2 Safety DataQampA
Neil Pedersen
1255 Data AccessQampA
Miguel Perez VTTI
140 Lessons from SHRPs Safety Data ResearcherQampA
Shauna Hallmark CTRE
225The Application Process
QampA Pam Hutton
245 General Questions from the Audience Pam Hutton
330 Closing Comments and Wrap Up Rudy Malfabon
A New Tool for Safety
bull We now can see inside the vehicle and on the road to understand what precedes crashes
bull New understanding of crash factors are possible
5272014
Objectives of Implementation
bull Demonstrate use of the SHRP2 Safety Data
bull Increase statesrsquo understanding of its potential
bull Identify countermeasures
bull Reduce crashes
Possible Research Topics
SHRP2 Safety Task Force is particularly interested in proposals addressing one of five research topics
bull Driver Speed How do drivers change or adjust speed as the navigate various roadway and environmental conditions
bull Roadway Features and Driver Performance How do certain roadway features influence driver performance or behavior
bull Preceding Contributory Events What driver characteristics behavior or performance precede crashes and near-crash events
Possible Research Topics
continuedhellip
bull Vulnerable Road Users How do drivers interact with vulnerable road users
bull Intersections How is navigation performance at either rural or urban intersections influenced by roadway elements driver characteristics or driver behaviors
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Welcome and Objective
bull Thank you for participating
bull Todayrsquos Objectivesndash Provide in-depth
information for IAP applications
ndash Allow questions to research data experts
ndash Answer questions on application process which opens May 27
Agenda
Noon (ET) Welcome Introduction Purpose Rudy Malfabon and Sandra Larson
1210 SHRP2 Safety DataQampA
Neil Pedersen
1255 Data AccessQampA
Miguel Perez VTTI
140 Lessons from SHRPs Safety Data ResearcherQampA
Shauna Hallmark CTRE
225The Application Process
QampA Pam Hutton
245 General Questions from the Audience Pam Hutton
330 Closing Comments and Wrap Up Rudy Malfabon
A New Tool for Safety
bull We now can see inside the vehicle and on the road to understand what precedes crashes
bull New understanding of crash factors are possible
5272014
Objectives of Implementation
bull Demonstrate use of the SHRP2 Safety Data
bull Increase statesrsquo understanding of its potential
bull Identify countermeasures
bull Reduce crashes
Possible Research Topics
SHRP2 Safety Task Force is particularly interested in proposals addressing one of five research topics
bull Driver Speed How do drivers change or adjust speed as the navigate various roadway and environmental conditions
bull Roadway Features and Driver Performance How do certain roadway features influence driver performance or behavior
bull Preceding Contributory Events What driver characteristics behavior or performance precede crashes and near-crash events
Possible Research Topics
continuedhellip
bull Vulnerable Road Users How do drivers interact with vulnerable road users
bull Intersections How is navigation performance at either rural or urban intersections influenced by roadway elements driver characteristics or driver behaviors
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Agenda
Noon (ET) Welcome Introduction Purpose Rudy Malfabon and Sandra Larson
1210 SHRP2 Safety DataQampA
Neil Pedersen
1255 Data AccessQampA
Miguel Perez VTTI
140 Lessons from SHRPs Safety Data ResearcherQampA
Shauna Hallmark CTRE
225The Application Process
QampA Pam Hutton
245 General Questions from the Audience Pam Hutton
330 Closing Comments and Wrap Up Rudy Malfabon
A New Tool for Safety
bull We now can see inside the vehicle and on the road to understand what precedes crashes
bull New understanding of crash factors are possible
5272014
Objectives of Implementation
bull Demonstrate use of the SHRP2 Safety Data
bull Increase statesrsquo understanding of its potential
bull Identify countermeasures
bull Reduce crashes
Possible Research Topics
SHRP2 Safety Task Force is particularly interested in proposals addressing one of five research topics
bull Driver Speed How do drivers change or adjust speed as the navigate various roadway and environmental conditions
bull Roadway Features and Driver Performance How do certain roadway features influence driver performance or behavior
bull Preceding Contributory Events What driver characteristics behavior or performance precede crashes and near-crash events
Possible Research Topics
continuedhellip
bull Vulnerable Road Users How do drivers interact with vulnerable road users
bull Intersections How is navigation performance at either rural or urban intersections influenced by roadway elements driver characteristics or driver behaviors
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
A New Tool for Safety
bull We now can see inside the vehicle and on the road to understand what precedes crashes
bull New understanding of crash factors are possible
5272014
Objectives of Implementation
bull Demonstrate use of the SHRP2 Safety Data
bull Increase statesrsquo understanding of its potential
bull Identify countermeasures
bull Reduce crashes
Possible Research Topics
SHRP2 Safety Task Force is particularly interested in proposals addressing one of five research topics
bull Driver Speed How do drivers change or adjust speed as the navigate various roadway and environmental conditions
bull Roadway Features and Driver Performance How do certain roadway features influence driver performance or behavior
bull Preceding Contributory Events What driver characteristics behavior or performance precede crashes and near-crash events
Possible Research Topics
continuedhellip
bull Vulnerable Road Users How do drivers interact with vulnerable road users
bull Intersections How is navigation performance at either rural or urban intersections influenced by roadway elements driver characteristics or driver behaviors
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Objectives of Implementation
bull Demonstrate use of the SHRP2 Safety Data
bull Increase statesrsquo understanding of its potential
bull Identify countermeasures
bull Reduce crashes
Possible Research Topics
SHRP2 Safety Task Force is particularly interested in proposals addressing one of five research topics
bull Driver Speed How do drivers change or adjust speed as the navigate various roadway and environmental conditions
bull Roadway Features and Driver Performance How do certain roadway features influence driver performance or behavior
bull Preceding Contributory Events What driver characteristics behavior or performance precede crashes and near-crash events
Possible Research Topics
continuedhellip
bull Vulnerable Road Users How do drivers interact with vulnerable road users
bull Intersections How is navigation performance at either rural or urban intersections influenced by roadway elements driver characteristics or driver behaviors
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Possible Research Topics
SHRP2 Safety Task Force is particularly interested in proposals addressing one of five research topics
bull Driver Speed How do drivers change or adjust speed as the navigate various roadway and environmental conditions
bull Roadway Features and Driver Performance How do certain roadway features influence driver performance or behavior
bull Preceding Contributory Events What driver characteristics behavior or performance precede crashes and near-crash events
Possible Research Topics
continuedhellip
bull Vulnerable Road Users How do drivers interact with vulnerable road users
bull Intersections How is navigation performance at either rural or urban intersections influenced by roadway elements driver characteristics or driver behaviors
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Possible Research Topics
continuedhellip
bull Vulnerable Road Users How do drivers interact with vulnerable road users
bull Intersections How is navigation performance at either rural or urban intersections influenced by roadway elements driver characteristics or driver behaviors
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Neil Pedersen TRB SHRP 2May 23 2014
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
SHRP 2 Safety Data
bull Description of the NDS and RID databasesndash NDS Naturalistic Driving Study
ndash RID Roadway Information Database
bull Data availability and accessndash Now for use in writing your proposals
ndash January 2015 for use in your analyses
bull Data considerations when preparing proposals
bull Potential research questions
bull Cost of data
bull Where to find more information
9
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Strategic Rationale
bull Driver behavior is keyndash Primary factor in two-thirds of crashesndash Contributing factor in more than 90 of crashesndash Hardest to study the thing we know the least about
bull Opportunity - Naturalistic Driving Study (NDS)ndash Miniaturized sensor technologies and increased computing capacity
can observe real-world drivingndash Method proven with 100-car study at Virginia Techndash Crash pre-crash near-crash and ldquonormalrdquo driving data
bull SHRP 2 research focused on collecting databull These data are expected to be used for the next 20-30 years
to better understand driver behavior and interactions between drivers their vehicles and the roadways
10
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Naturalistic Driving Studies
bull Instrument volunteer driversrsquo vehicles and collect data continuously during their normal driving
bull Whyndash What do drivers really do Speeding tailgating cell phone alcoholhellipndash What were they doing just before they crashed
bull Usual crash studies can only guessbull We can see fraction of second by second what happenedbull How did they avoid a crash
ndash How do the roadway vehicle and environment impact driving
bull SHRP2 Naturalistic Driving Studyndash 40 times larger than 100 car studyndash National in scopendash All ages both genders
11
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
SHRP 2 Naturalistic Driving Study amp Roadway Information Databases
NDSData
RID(GIS)
Data from 3147 volunteer drivers and their vehiclesin six sitesPassenger cars vans
SUVs pickups
New data collected 12500 centerline milesconsistent across six sites
Acquired data (DOTs others) bull 200000 centerline milesbull Roadway weather traffic
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
NDS Study Design
bull Largest naturalistic driving study ever undertakenndash 3147 drivers all agegender groupsndash 3958 data years 5 M trip files 497 M vehicle milesndash 3 years of data collection
bull Most participants 1 to 2 yearsndash Vehicle types All light vehicles
bull Passenger Carsbull Minivansbull SUVsbull Pickup Trucks
ndash Six data collection sitesbull Mix of urban and rural
13
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
NDS Data Overview
bull Driver demographics assessmentsbull Vehicle descriptorsTRIP DATA
bull Multiple Videosbull Machine Vision
ndash Eyes Forward Monitorndash Lane Tracker
bull Accelerometer Data (3 axis) bull Rate Sensors (3 axis)bull GPS
ndash Latitude Longitude Elevation Time Velocity
bull Forward Radarndash X and Y positionsndash X and Y Velocities
bull Cell Phone Recordsndash Beginning and end of calls on
major carriers
bull Passive Alcohol Sensorbull Illuminance sensorbull Infrared illuminationbull Incident push button
ndash Audio (only on incident push button)
bull Turn signalsbull Vehicle network data
ndash Acceleratorndash Brake pedal activationndash ABSndash Gear positionndash Steering wheel anglendash Speedndash Hornndash Seat Belt Informationndash Airbag deploymentndash Many more variableshellip
14
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
NDS Example Video Data(not an actual participant)
15
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
NDS Data Overview
bull Size the file is hugendash 2 petabytes = 2 million 1 GB flash drives (12 PB video 08 PB sensor)ndash ldquoGive me the whole raw data filerdquo isnrsquot possible or sensible
bull Complexity different data types ndash Categorical data constant over a trip driver age vehicle typendash Sampled data collected at original resolution (once a trip up to 640 Hz
during a crash) speed acceleration GPS position radar vehicle network information
ndash Video data from 4 cameras must be codedbull Automated reduction lane trackerbull Manual reduction all other items for specific analyses
bull Privacy considerations personally-identifying data (PII)ndash Face video and other personal information access only with Institutional
Review Board (IRB) approval for qualified researchers in a secure location16
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
NDS Tools for Data Users
bull Trip summary filesbull Crash near-crash and baseline event and epoch
filesbull InSight websitebull Linking NDS and RID data
17
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Trip Summary File
bull Trip summary file - categorical data on each trip ndash Think of a spreadsheet with 1 row per trip 5M rowsndash Identify trips of interest also can be analyzed directly
bull Variablesndash Driver data ndash demographics driver assessmentsndash Vehicle data ndash descriptive ndash Roadway data ndash roadway type (incl ruralurban) posted speed limit
number of intersections data source (mobile van state inventory) ndash Trip data ndash duration speed accelerations headway etcndash Variables that change during a trip in bins counts or maxmin
bull speed bins 0-10 mph 10-20 mph etc ndash time or of trip in eachbull number of accelerations higher than threshold value
18
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Crash Near-crash Baseline Files
bull Crashes expect 700 varying severitybull Near-crashes ldquoalmostrdquo crash but for hellip expect 7000
ndash Crash surrogates how did driver avoid a crash
bull Baseline randomly selected across all vehicles 30000ndash Denominator for risk calculations measure overall prevalence
bull Event files for each ndash categorical datandash Coded from last 6 seconds of data before precipitating eventndash Manual video reduction (eg driver distraction)
bull Epoch files for each ndash sensor and video datandash 30-second data segments (20 before precipitating event 10 after only
20 for baseline)ndash Manual eye-glance coding
19
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Website
bull Study description and overview videobull Data access guidebull Data dictionariesbull Sample data bull Data
ndash Descriptive data for whole data filendash Categorical data on trips from trip summaries query capabilityndash Event data from crashes near-crashes baselinendash Viewer - forward video and time series data for crashes near-crashes
bull Data de-identified no PII easy IRB certification on-line access
bull httpsinsightshrp2ndsus20
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Website Homepage
21
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Website Data Categories
22
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
RID Data Overview
bull Size the file is manageablendash 50-60 GB without video 6-8 TB with video and aerial images
bull Complexity 4 different data sourcesndash Mobile van data very detailed about 12500 centerline miles 43195
intersections 518570 MUTCD signs includes forward videondash State roadway inventory data from 6 study States HPMS datandash National base map database for entire countryndash Supplemental data from 6 study States data vary by State
bull Privacy considerations should be nonendash In most cases RID data should not require IRB approval Video data may
require IRB to determine exemption if used in conjunction with NDS data
bull When the NDS and RID are linked match trips with road segments match road segments with trips
23
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
North Carolina Site (RaleighDurham)New roadway data4558 miles = 2279 centerline miles
24
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Acquired data from NCDOT coverage(extends beyond lsquoboxrsquo)
25
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Mobile Van Data
bull New data SHRP 2 collected consistently across 6 statesbull Quality assured to meet project specsbull 25000 driven12500 centerline miles across the six NDS
sitesndash FL 4366 miles 45 rural55 urbanndash IN 4635 miles 64 rural36 urbanndash NC 4558 miles 59 rural41 ruralndash NY 3570 miles 68 rural32 urbanndash PA 3670 miles 83 rural17 urbanndash WA 4277 miles 31 rural69 urban
bull Total 25076 directional miles
26
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Types of Mobile Van Data
bull Horizontal Curvature Radius Length PC PT Direction bull Gradebull Cross Slopebull Lane in terms of the number width and type ( turn passing acceleration
car pool etchellip) bull Shoulder typecurb paved width if existsbull Intersection location number of approaches and control (uncontrolled all-
way stop two-way stop yield signalized roundabout) Ramp termini are considered intersections
bull All MUTCD signsbull Barriersbull Median presence (YN) type (depressed raised flush barrier) bull Rumble Strip presence (YN) location (centerline edge line shoulder) bull Lighting presence(YN)
27
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Acquired Roadway Data
bull Existing roadway inventory data acquired from agencies such as the six State DOTs (data items not consistent) ~200000 centerline miles
bull Includes HPMS files for the six states plus bull Functional Classificationbull Signalsbull Intersectionsbull Access Controlbull Pavement Conditionbull Bridge Locationbull Vertical Alignmentbull Interchangesbull Rest Areasbull Terrainbull Tunnelsbull FRA grade crossings
28
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Acquired Supplemental Data
bull Existing data and information from State DOTs Public Agencies and Private Sources
bull ~ 200 000 centerline miles bull Crash history databull Traffic information ndash AADTbull Traffic Data - continuous counts
(ATR)bull Traffic Data -short duration countsbull Aerial imagery bull Speed limit databull Speed limit lawsbull Cell phone and text messaging
lawsbull Automated enforcement laws
bull Alcohol-impaired and drugged drivers laws
bull Graduated driver licensing (GDL) laws
bull State motor cycle helmet use lawsbull Seat belt use lawsbull Local climatological data (LCD)
NOAAbull Cooperative weather
observerother sourcesbull Winter road conditions (DOT)bull Work zonebull 511 informationbull Changes to existing infrastructure
conditionbull Roadway capacity improvements
29
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Linking NDS and RID
bull Identify all trips passing over a given roadway segmentbull Identify all roadway segments over which a given trip travelsbull Link file matches trip IDs and roadway segment IDsbull To be complete by Dec 31 2014
30
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Data Availability
bull NDS data collection completedndash April 2014 47 million trips through quality controlndash Quality control rest of trips add radar cell phone data by December 31 2014
bull RID data collection completedndash Complete quality control add supplementary data by October 31 2014
bull Trip summary filesndash April 2014 1143033 trips not all variablesndash December 31 2014 complete
bull Crash near-crash and baseline event and epoch filesndash April 2014 113 crashes and near-crashesndash December 31 2014 all crashes near-crashes baseline
expect about 700 crashes 7000 near-crashes 30000 baseline
bull Linking NDS and RID data complete December 31 2014
31
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Data Access Three Levels
bull Websitendash Previously discussed
bull Full data without PIIndash All data not requiring driver face video or trip start or end datandash IRB review required but should be straightforward may be exemptndash Data can be exported to userrsquos institutionndash Data Sharing Agreement requiredndash Analyses not requiring PIIndash No driver eye glances facial expressions
bull Full datandash All datandash IRB review and Data Sharing Agreement requiredndash PII data must be viewed in secure enclavendash All analyses
32
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Data Access May-June 2014
bull NDS ndash through InSight websitendash Learn about the data study documentation data dictionaries ndash Work with sample datandash Use Queries to estimate sample sizes for your research questions
bull NDS ndash through data sharing agreement with VTTIndash Will require IRB approval or exemptionndash Miguel Perez to provide more details
bull RID data ndash read-only copy of filendash North Carolina data currently availablendash All 6 states available July 1ndash All supplemental data available Oct 31ndash Available from Omar Smadi at CTRE and Charles Fay at TRB
33
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Data Access in 2015
bull NDS datandash From InSight website as described previously
ndash From VTTI through a data sharing agreement to be described by Miguel Perez
bull PII data must be viewed in secure enclave
bull RID datandash Available through InSight website
bull NDS-RID Linking Filendash Available through InSight website
34
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Three Pilot Studies (S08)
bull Run-off-road crashes on rural 2-lane curvesndash Effects of signs markers chevronsndash CTRE Iowa State University
bull Offset left-turn lanesndash Effect on safety design guidancendash MRIGlobal
bull Driver glance behaviorndash What glance patterns are saferndash SAFER
bull Statusndash Phase 1 proof of concept completedndash Phase 2 contract awards spring 2013ndash Final reports July 2014
35
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Data Considerations When Preparing Proposals1 Understand the data ndash learn as much as you can before you start This
includes missing data representativeness biases link between NDS and RID and the like
2 Think through your analysis plan carefully ndash research questions data needed sample sizes how to identify the data etc
3 Expect to talk to VTTI and CTRE early and often after research has begun
4 IRB approvals and data sharing agreements probably take longer than you expect (sometimes much longer) ndash work this into your schedule
5 In Phase I you will get a small pilot data set and run a proof-of-concept trial This helps immensely in understanding the data
6 You may need to get a second pilot data set after yoursquove worked with the first and learned what you should have asked for
7 If you plan to use information from the video data think through carefully how the videos can best be accessed and coded
36
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Categories of Research Questions (S02)
bull How do driver characteristics (eg demographic differences) influence crash likelihood
bull What impacts do roadway countermeasures have on lane-keepingbull How does driver distraction influence crash likelihoodbull How do aggressive driving behaviors influence crash likelihoodbull How does driver fatigue influence the likelihood amp types of crashesbull What is the effect of driver impairment on crashes and driver errorbull How does turn lane configuration influence behavior and crash riskbull How do roadway features influence crash likelihoodbull How do signage lighting conditions and other traffic control-related
countermeasures influence crash likelihood amp driver performancebull How do traffic and traffic volume influence intersection negotiation lane-
keeping performance and crashes
37
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Cost of Data
bull For Phase 1ndash If all data is available on the website there will be no costndash If a data sharing agreement with VTTI is required the cost will depend on the level of
effort to extract datandash Based on the S08 pilot project experience costs for data projects requiring complex
data extraction can be expected to be about a third of the total Phase 1 costndash Projects that have less complex data requirements could be expected to have a lower
percent budget for data extractionndash Detailed cost estimates require multiple iterations of refining the data request with VTTI
before a final cost estimate can be preparedndash There will not be time to get a detailed cost estimate from VTTI before June 27 only
whether a data request can be expected to be complex or less complexndash Phase 1 funding should be sufficient to enable extraction of a small sample data set for
testing the proof of concept as part of the Phase 1 budget
bull For Phase 2ndash A detailed data cost estimate for Phase 2 will be required as part of the Phase 1 report
38
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Where to Find More Information
bull About the NDS bull at the InSight website httpsinsightshrp2ndsusbull at the recorded NDS webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About the RID bull At the recorded RID webinar
httpwwwtrborgStrategicHighwayResearchProgram2SHRP2SafetyWebinarsaspx
bull About potential research topics and example work plansbull In the S02 report Integration of Analysis Methods and Development of
Analysis Plan httpwwwtrborgPublicationsBlurbs166051aspx
bull About the three current analysis projectsbull In the S08 summary report Initial Analyses from the SHRP 2
Naturalistic Driving Study httpwwwtrborgPublicationsPubsSHRP2ResearchReportsSafetyaspx
39
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Accessing the SHRP 2 Safety Data
Miguel PerezSuzie Lee
SHRP2 Safety Data IAP WorkshopMay 23 2014
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Outline
bull Data access life-cyclendash InSight access (website)ndash InDepth access (researcher-specific data needs)
bull Data access ndash IRB and DSAndash What and Why of IRBsndash Practical Implications
bull PIIbull What should you expect to receivebull QampA
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
InSight Data Access Life-Cycle
bull Access website (httpinsightshrp2ndsus)bull Register
ndash Provide some basic information about yourselfndash Verify that your e-mail address is validndash Agree to Terms of Service = data sharing agreement
bull Gain ldquoQualified Researcher Statusrdquondash Obtain IRB training and submit certificate for verificationndash Free of charge at NIH or VT (and possibly at your institution)
(httpsphrpnihtrainingcomusersloginphp)(httpwwwirbvtedupagestutorial_introhtm)
bull Explore testbull Before engaging in research activities (exploring is OK) confirm
exempt status with your institutionrsquos IRBbull Submit requests for downloads to VTTI ndash proof of IRB
exemptionapproval and data sharing agreement will be required at this point
ndash Should be a very quick process
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
InDepth Data Access Life-Cycle
bull Initial contactbull Data specificationbull Statement of workbull Contracting bull IRB approval and data sharing agreementbull Generation of pilot datasetbull Data specification refinementbull Generation of ldquofinalrdquo datasetbull Supportbull Project close-out
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
DATA ACCESS ndash IRB AND DSA
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
US Institutional Review Board (IRB) History and Perspective
bull Similar to Research Ethics Boards or Ethics Review Panels in other countries
bull Modern era began with Belmont Report (httpwwwhhsgovohrphumansubjectsguidancebelmonthtml )
bull Carries the force of law in the USndash 45 CFR 46ndash Common Rule (ie other federal agencies and
departments abide by 45 CFR 46)
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
A Word from our Sponsor(s)
bull Not AASHTOStatesFHWASHRP 2 ndash but the participantsbull According to the Belmont Report there are three principles
by which human subjects research should be conductedndash Respect for persons autonomyndash Beneficencendash Justice
bull These principles are generally accepted and followed internationally as well
bull Participants are what makes the SHRP 2 NDS possible and we must continue to respect the promises we made them (and the data they provided to us) into the future
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
What IRB Isnrsquot ndash and What it Is
bull IRB isnrsquotndash Just more paperworkndash An impediment to your research and careerndash A hurdle to jump over
bull IRB isndash Following the three principles ndash autonomy beneficence
justicendash Honoring promises to participantsndash ldquoNamely an ethos that should permeate the way those
engaged in human research approach all that they do in their researchrdquo (Australian National Statement on Ethical Conduct in Human Research 2007)
bull Anyone who uses the SHRP 2 NDS data is engaged in human subjects research
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
IRB - Practical Implications
bull Determine if your institution has an IRBndash If it doesnrsquot - ldquofor hirerdquo IRBs are availablendash Virginia Techrsquos IRB cannot be used unless VTTIrsquos role includes
data analysisbull Familiarize yourself with your IRBrsquos policies and procedures
ndash Most IRBs offer a simplified application for this type of data mining project
bull Complete application and wait for IRB decisionndash Most likely InSight data use will be deemed ldquoexemptrdquo by the
IRB (meaning that no further approvaloversight is required)ndash InDepth data use will likely either be ldquoexemptrdquo (if no PII is used)
or go through ldquoexpedited reviewrdquo (if some PII use is required implies that the approval decision is within the purview of the IRB chairperson)
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Data Sharing Agreements (DSAs) ndash Practical Implications
bull Agreement between the data steward and the research teambull Formalizes that data will be shared and used in accordance with the
promises made to the participants in the informed consentbull Brief (~2-3 page) document that requires
ndash Scope of analysis (1 paragraph) ndash Specific dataset requested (1 paragraph) ndash Researcher qualifications (1 paragraph per researcher)ndash Data retention plan ndash Proof of human subjects training for the researchers ndash IRB approval (or proof of exemption) andndash Agreements to not attempt to learn identity of participants and to
constrain data use to the activities outlined in the DSAbull Template is available in the data access guide (downloadable from
the InSight website)ndash Use the template
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
PII
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Why the Focus
bull The IRB system was put in place to protect humans participating in research activities
bull In data mining efforts where the data have been collected the main risk to the original participants is the release of information that could be used to link them to their data
bull For example the SHRP 2 NDS contains information aboutndash Drug usendash Medical conditionsndash Home and office locationsndash Places that participants frequent (eg church family amp
friends households commercial establishments)bull Participants would typically prefer to keep at least some of this
information private but they entrusted us with it to help answer research questions
bull Note Many research questions do not require PII
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Personally Identifying Information(and workarounds)
bull Dates and times ndash Transform (eg March 15 2010 0645am to March Monday 6am-12pm)
bull Voiceprintsndash Pitch and modulation adjustments
bull Full face photos videos amp comparable imagesndash Blur
bull GPS coordinatesndash Cut off trip start and end locations (and surrounding areas)
bull Full trip files with starting and ending locations shown via forward videondash Cut off trip start and end locations (and surrounding areas)
bull Files with identifiable highway signs and footage of a high-profile incidentndash Remove or blur
bull Any other types of data that could be used to identify a research participant
bull The main mechanisms for protection are the use of the secure data enclave andor data processing occurring within the data warehouse
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
WHAT SHOULD YOU EXPECT TO RECEIVE
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Data Formats
bull The answer to ldquowhat will I receiverdquo is that ldquoit dependsrdquobull In most cases a text file (eg csv) will contain all of the
information you need (eg a website query export)bull In some cases an export will consist of many text files
ndash These will contain the time series data for the variables that you requirehellipusually one text file per trip or trip segment of interest
bull When video is required (and non-identifying) the different views will be provided in separate files that will share a name with the file containing the corresponding time series data
bull See the InSight website for sample files(httpsinsightshrp2ndsusprojectBackgroundindex scroll down to ldquoExample
Drivesrdquo)
bull Other access mechanisms to be developed beyond 2015
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
QUESTIONS AND DISCUSSION
Miguel A PerezDirector Center for Data Reduction and Analysis Support(540) 231-1537miperezvttivtedu
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Shauna Hallmark CTREMay 23 2014
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
SHRP2 S08d Evaluation of Lane Departure Risk on Rural 2-lane Curves
bull Define relationship between driver distraction other driver roadway and environmental characteristics and risk of lane departure
‒ Impact of countermeasures ‒ Impact of specific roadway features (ie radius)‒ Impact of distractiondriver characteristics
57
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
SHRP2 S08d ndash Rationale for Selection of TopicMethodologies
bull SHRP 2 S02 identified rationale for selection of research questions using SHRP2 NDS data
bull Does it matter to stakeholders‒ State DOTs‒ Lane departure crashes ~ 50 fatal crashes‒ Curves have 3 x crash rate of significant interest to
Statelocal agencies
58
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Are NDS data the right data (ie would simulation crash data etc be better)
Image source SHRP 2 S02
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
Source ESRI Florida DOT
bull Can the data support the research question (evolving as we learn more about the dataRequested small dataset firstReviewed data dictionaries and available documentation
o eg was lane tracker accurate enough to identify lane departure
Reviewed how variables were collected
Adjusted expectations as new information became availableo Steering wheel position not
as available as expected
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
bull Is research question structured so results help stakeholders‒ Tried to structure so results reflected what DOTslocal agencies
expect (odds ratio rather than complicated statistical equations - as appropriate)
bull Consider time and resources needed to address fully address research question
‒ Identified sample size necessary for research question‒ ie calculated how long to manually reduce driver face video and
process datasets‒ Weighed sample size against practicality
of reducing data‒ Selected team with necessary skillset
‒ GIS‒ Roadway engineering‒ Human factors
SHRP 2 S08d ndash Rationale for Selection of TopicMethodologies
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
bull Assess the relationship between driver behavior and characteristics roadway factors environmental factors and likelihood of lane departures using NDS and roadway data
‒ Develop models to quantify the relationship between driver behavior and the roadway environment
‒ Focus on curves on rural 2-lane paved roadwaysbull 2-lane roadways are particularly problematicbull Lane tracker does not function
on unpaved roadways
Objective
62
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Identification of Rural Curves
bull NDS and RID not yet mergedbull Queried RID for rural 2-lane curves
of interestbull radius presence of RS etc
bull Reviewed other data (ie Google Earth)
bull Identified potential curves developed buffer in GIS
bull Provided curve buffers to VTTIbull VTTI provided data for trips through
identified curves
image and data source Google ESRI VTTI
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
bull driver facendash glance locationndash distractionndash state (ie drowsy)
bull forward videondash on-coming vehiclesndash vehicle-following ndash confirmation of
roadway featuresndash environmental (ie
rain dark)
bull rear video
bull over-shoulderndash distraction
Image sources ESRI VTTI does not show unconsented driver
Reduced in-house
bull RIDndash roadway
characteristicsbull times series from
NDSndash GPSndash vehicle kinematics
bull cabinndash of passengers
Reduced at VTTI secure data enclave
SHRP 2 S08d ndash Data Requested
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
bull Represented vehicle position at 01 sec increments (time series data)
bull 1 event = 1 driver trip through 1 curvebull Spatially correlated vehicle trace to curve
‒ Identified PCPT and then location within tangent or curve (ie 50 m upstream)
Raw Vehicle Kinematic Data
vttitimestamp vttiaccel_x vttiaccel_y vttiaccel_z LocationDistance (meters) vttigyro_y vttigyro_x vttigyro_z speed (ms)
256299 -00029 -00029 -10063 upstream -24 0 0 0 3056256400 -00029 -00029 -10034 upstream -21 0325195 0 0 3056256499 -0029 -00174 -09744 upstream -18 0325195 -0325195 -0325195 3056256600 -00029 -00174 -10063 upstream -15 0975586 -0650391 -0650391 3056256699 -00058 -00029 -09802 upstream -12 0325195 0 -0650391 3056256800 -00116 -00145 -09628 upstream -9 0 0 -0325195 3056256900 -00261 -00174 -09976 upstream -6 0 0 -0650391 3056257000 -00319 -0029 -09918 upstream -3 3056257099 0 -00116 -09947 PC A 0 0650391 -0975586 -0650391 3049257200 -00087 -00377 -09947 in curve 3 0 -1625977 -0325195 3042257299 -00116 -00319 -09541 in curve 6 -0975586 -1625977 -0975586 3035257400 -0029 -00058 -09744 in curve 9 -1300781 -0650391 -0975586 3028257499 -00029 -00232 -09947 in curve 12 0 0 -1300781 3021257600 00058 -00464 -09483 in curve 15 0 0325195 -0975586 3014257699 -0029 -00696 -09512 in curve 18 0 0 -1300781 3007
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Reduction of Kinematic Driver Factors
bull Reduced driver distractionglance behavior at VTTI secure data enclave
bull Linked to times series file via time stampbull Where were drivers
lookingbull When drivers were
looking away from forward roadway what distractions were present
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Crash Surrogates
bull Were not able to identify crashes first (would have been ideal)ndash Sufficient crashes may not be available in full dataset
bull Reviewed literature and available data (type and quality) to identify best lane departure crash surrogates
ndash Surrogates such as lane position could not be supported with available data
bull Identified events of interest (eg hard side accel speeding braking) and normal driving events for selected curves
bull Selected 2 surrogatesndash Lane crossing (left or right)ndash Curve entry ge 5 mph and ge 10 mph over advisory speed if
present or tangent speed if no advisory speed
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Analyses (results are preliminary)
upstream curve
bull Research Question 1 What defines the curve area of influence (eg where do drivers begin reacting to the curve)
ndash Used times series data for 43 curves (IN NY NC)ndash Used 400 meters upstream of the curve ndash Identified point at which speed and pedal position changed
bull average reaction point was 164 meters upstream of PCndash Used Bayesian Hierarchical model
bull reaction point changes with curve radius and direction (inside versus outside of curve)
ndash Reaction point is ~ 009 m greater for each +100 m in radius
ndash Drivers react ~ 2 m sooner when traveling on outside of curve
bull Implications for curve signing
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Analyses (results are preliminary)
Image source Charlton (2007)
bull Research Question 2 How do drivers normally negotiate curves
ndash Comparison of lane offset and speed at six key points along the curve (PC PT and 4 equally spaced points)
ndash Used panel data model with random effects and ordinary least squares
bull Lateral Position ndash offset in upstream and insideoutside of curve appear to be most significant
bull Speed ndash speed in upstream curve radius and whether they were following another vehicle appear to be most significant factors
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Analyses (results are preliminary)
bull Research Question 3 What is the relationship between driver amp roadway characteristics and lane departure risk
ndash Response variables probability of left or right side lane departure or probability of entering curve 5 or 10 mph over advisory or tangent speed
ndash Event based analysis (one driver trip through curve or tangent section)
bull Reduced data 200 m upstream and through curvendash Logistic analysis (in progress)
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Analyses (results are preliminary)
bull Research Question 4 How does different factors (driver distractiononcoming vehiclecurve geometry) influence lane keeping behavior
ndash Times series datandash Dynamic linear model (DLM)ndash Identifies driver response to roadways and traffic cuesPreliminary Findingsndash Short period of distraction does not influence lane deviationndash oncoming vehicle and curve geometry had significant impact on lane
deviationndash Model prediction could be
potentially used for lane departure warning system
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Lessons Learned
bull Research questions should bendash suited to NDS datandash suitable for time frame and resources
bull ldquoEvaluate factors leading to lane departures on rural 2-lane roadwaysrdquo vs ldquoEvaluate factors leading to lane departuresrdquo
bull Understand IRB processbull Review available material about data first
ndash how were data collected linked etcndash What is the accuracy resolution and availability of needed sensors
(ie steering was only available in fraction of vehicle)bull Formulate specific data requests (work with VTTI amp CTRE to fine tune)bull Work with small dataset firstbull Be patient Start early Have back-up plan
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Concept to Countermeasure ndash Research to Deployment Using the SHRP2 Safety DatabasesNDS and RID Data Presentation Pam Hutton AASHTOMay 23 2014
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Implementation Assistance
bull FHWA AASHTO are pursuing a phased approach to reduce risk and cost to states
bull However states accept responsibility to manage research implement findings
bull States encouraged to partner with researchers particularly ones with IRB process
bull States must deliver the authorized research
bull If the concept phase is successful in-depth research phase with a large data set may be authorized
74
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Safety Implementation Process
Phase I - Proof of Concept with a sample reduced data set
Phase II full data set and in-depth analysis amp CM identification
Possible Phase III to adopt or implement countermeasure nationally
Decision
Decision
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Phase I Budget Issues
bull Funding up to $100000 for Phase I effortsbull Applicants can augment the budget with matching funds if
they wishbull Applicants must acquire their own datasetsbull Applicants must account in their budget and their
schedule for the time and cost of acquiring databull Applicants must include in their budget the cost of travel
to panel meeting at end of Phase Ibull Potential Phase II budgets and schedules will be
negotiated if Phase II is approved
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Application Selection Process
bull The online application must include a concise research approach addressing high-priority issues with practical findings
bull Partnership with strong researchers recommended
bull A commitment to pilot or champion the findings
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
IRB Issues
bull We emphasize the importance of compliance with data privacy
bull Applicants must have a certified Institutional Review Board Process
bull Teaming with a researcher with an IRB process could address this issue
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Phase II Actionable Research
bull Only occurs with successful completion of Phase I and authorization by FHWA in conjunction with AASHTO
bull Task Force must provide agreement that continuing research will lead to actionable findings
bull Efforts will consist of in-depth and detailed analysis of the proposed research question using SHRP2 Safety data
bull Results should be findings and recommendations to potential new insights andor countermeasures
bull Deliverables should includendash Detailed plan for Phase IIIndash Cost estimate for Phase III
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Phase III Deployment
bull May be authorized by FHWA in conjunction with AASHTO
bull Will be subject to findings in Phase II researchbull Will NOT include additional researchbull Activities may include
ndash Integration of findings into manuals guidelines policies
ndash Countermeasure developmentndash Pilot testing
bull Countermeasures may be included in future rounds of the Implementation Assistance Program
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Other Useful Information
bull Cost sharing will be considered this may include eligible Federal-aid funds pooled funds state resources in-kind contributions or other sources
bull Budgets qualifications resumes and past experience do not count toward the 5000 word application limit
bull State transportation agencies may submit multiple applications Researchers may be on multiple teams
bull Proposals should include costs to travel to meet with the task force to discuss Phase I findings and proposals for Phase II research
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Additional Resources
bull Substantial information exists for preparing an application including The SHRP2 Solutions page at httpwwwfhwadotgovgoshrp2SolutionsSafetyList
bull FAQs are available at httpshrp2transportationorgDocumentssafetySHRP2_Safety_IAP_FAQs_FINALpdf
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Application Schedule
bull Applications May 30 to June 30
bull Selections Late Summer 2014
bull Expected Phase I start date January 2015
bull Phase I complete September 2015
bull Phase II andor III dependent on results of Phase I
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84
Questions
Implementation Assistance wwwfhwadotgovgoSHRP2
For more information Aladdin BarkawiFHWA Safety Implementation LeadAladdinBarkawidotgov
Kelly HardyAASHTO Safety Implementation LeadKHardyaashtoorg
84