NCHRP Project 3-79a Arterial Performance Measures

Working Paper No. O2-2

Objective 2: Alternative Test Sites for Evaluating Loop Detector Signature Matching for Estimating

Link Travel Time

Prepared for: National Cooperative Highway Research Program

Transportation Research Board National Research Council

Transportation Research Board

NAS-NRC LIMITED USE DOCUMENT

This report is furnished only for review by members of the NCHRP project panel and is

regarded as fully privileged. Dissemination of information included herein must be approved by the NCHRP.

Prepared by:

James V. Krogmeier Darcy M. Bullock, Joseph M. Ernst

Purdue University

June 1, 2009

Objective 2: Open Architecture Multimedia Database for Archiving Signatures, Photos, and Ground Truth Matching

Abstract

Travel time is an important metric for evaluating the performance of transportation segments. Historically link travel times have been estimated using probe vehicles, license plate matching, and analytical models based upon volume, occupancy and speed. More recently, link travel times have been estimated using toll-tag readers, cell phone tracking, and consumer electronic device tracking. Objective 2 of 3-79a is to collect a relatively large library of detector signatures, evaluate alternative detector matching algorithms, and assess the feasibility for using these techniques to estimate link travel times. To achieve this objective, a structured data collection and management plan is essential to ensure the techniques are systematically evaluated. This paper is the second in a series of white papers supporting Objective 2. The first paper in this series described the database design and the collection infrastructure for the database development. This paper will describe three possible sites for a more extensive data collection:

• Signalized intersection in West Lafayette, IN

• Signalized intersection in Noblesville, IN

• Interstate sites on I-70 separated by approximately 1 mile

Each site is described and the advantages and disadvantages are discussed. In conclusion, a recommendation is made to use the West Lafayette site for extensive data collection.

Introduction

To adequately evaluate vehicle re-identification (matching) methods, large data sets are required. The multimedia open architecture database described in the first white paper in this series is designed to store all data necessary for vehicle re-identification research. This database uses a standardized format for storage of different types of data collected from vehicles travelling through existing data collection assets along corridors and highways. There are several characteristics that will be evaluated in order to choose the best data collection site: 1) Microloop and Inductive Loop Stations Available 2) Distance between detectors 3) Video Capture Capabilities 4) Ground Truthing Capabilities

4) Ease of Access/Deployment Many detectors at intersections throughout the United States are either micro-loop or inductive loop detectors. While there are many other types of detectors, few of them produce high resolution numerical data that is desirable for vehicle recognition. It is hypothesized that the micro-loop data will yield signatures showing finer features than will the data from inductive loops. For this reason, it is desirable for the data set to include both types of sensors in order to make a rigorous comparison. Although modest spacing (say 0.5 mile to 1 mile) is desirable for estimating travel times, such spacing is not necessarily required for the evaluation of matching algorithms unless the algorithms use the distance to do some sort of travel time filtering or to evaluate platoons of vehicles instead of individual vehicles. The correlation between two signatures of individual vehicles is not dependent on the distance between the sensors. Video capture capabilities are not required for any of the matching algorithms, but are important for supporting ground truth verification of matches. Without video capture, other quality metrics can be used to evaluate performance, but an actual ground truth cannot be generated. The ease of access, deployment, and acquiring ground truth are important because it will determine the size of a manageable dataset. The next three sections include descriptions of each of the three data collection sites with respect to these characteristics. The final section recommends the West Lafayette site based on a comparison of the three sites.

Site 1: West Lafayette (Northwestern and Stadium intersection)

The Northwestern and Stadium intersection in West Lafayette is shown in Figure 1 with a more detailed view of the Northbound approach in Figure 2. The Northbound approach is the chosen approach for this intersection because it is the only approach with advanced detection. In each of the two Northbound through lanes, data can be collected from

• two stop bar micro-loop detectors (referenced as NA_M1, NA_M2, NB_M1, and

NB_M2 in Figure 2 and Figure 3),

• two advanced mirco-loop detectors(referenced as NA_M5, NA_M7, NB_M5, and

NB_M7 in Figure 2 and Figure 4), and

• two inductive loop detectors(referenced as NA_L6, NA_L8, NB_L6, and NB_L8 in

Figure 2 and Figure 4)

for a total of 8 micro-loops and 4 inductive loops. Data can be collected from all 12 sensors concurrently. The stop bar inductive loops are not suitable for this data set because they are connected in series to one detector channel instead of each being connected to an individual detector channel. The video available for the Northbound lanes is shown in Figure 3 for the stop bar detectors and Figure 4 for the advanced detectors. These two cameras provide sufficient coverage for ground truth of all 12 sensors. The distance between the advanced detectors and the front loops is 125 ft. This is sufficient distance for velocity change between sensors and some lane change. It is not a large enough distance for significant re-ordering of vehicles. This distance should be sufficient for evaluation of algorithms that attempt to pair individual vehicles. If an algorithm incorporates matching platoons of vehicles, it would likely perform better in this test site than at a site with more separation. This site is ideal for ground truthing data. This is one aspect that makes the 125 ft distance between sensors desirable. The other reason is that this site can be easily ground truthed with existing camera infrastructure. This site lends itself to easy deployment because of its close proximity to the Purdue campus and also because it is the site where the prototype has been developed. Because of the ease of deployment and ground truthing, the data set can be expected to have many more records than the other two locations.

Site 2: Noblesville (SR 37 & SR 32)

The Noblesville site is shown in Figure 5. Because the Northbound site has advanced detection, it would be selected for collecting data. A more detailed diagram of the Northbound approach is shown in Figure 6. The sensors at this intersection also include

• two stop bar micro-loops (referenced as NA_M1, NA_M2, NB_M1, and NB_M2 in

Figure 6 and Figure 7 ),

• two advanced micro-loops(referenced as NA_M5, NA_M7, NB_M5, NB_M7 in Figure 6

), and

• two advanced inductive loops in each lane (referenced as NA_L6, NA_L8, NB_L6,

NB_L8 in Figure 6 ).

Similar to the West Lafayette site, the stop bar inductive loops are not suitable for this data set because they are connected in series to one detector channel instead of each being connected to an individual detector channel. The advantage of the Noblesville site as compared with the West Lafayette site is that the distance between the stop-bar and advanced detector is about 400 ft instead of about 125 ft at the West Lafayette site. However, at Noblesville, there is only one camera for the Northbound approach. A screenshot from this camera is shown in Figure 7. When the camera is zoomed out sufficiently for capturing the stop-bar detectors, the resolution of the vehicles travelling over the advanced loops makes accurate ground truthing less than ideal. If this site is selected, the project team would devote more effort to the data collection and less to the algorithm development.

Site 3: I-70 (MM 66.6 and MM 67.3)

The I-70 site is shown in Figure 8 with a diagram of the sensors in Figure 9. Sensors will be used at both MM 66.6 and MM 67.3. Each of the two sites has two micro-loops (lead and lag) in each of the two lanes for a total of eight micro-loop detectors. There are no inductive loop detectors at this site. Figure 10 shows screen shots from two cameras that can be used to ground truth the data collected at these sites and Figure 11 shows the proposed temporary data collection infrastructure. The main advantage of this site is that the detectors are about one mile apart. This is about 10 times the distance as the Noblesville site. Video can be captured from both sites as shown in Figure 10, but it requires coordination on a day by day basis with INDOT to move cameras for a limited time. This means that the data set from this site would be limited to several hours during one collection day. Since the travel times will have a much higher range, ground truth will be much more time consuming to produce. Also, this segment of freeway experiences limited congestion so there would be little variation in speed that is typically encountered on a signalized arterial.

Recommendation

A summary of the advantageous and disadvantages of each site is shown in Table 1. Although it would be desirable to have longer detector spacing, the project team recommends the West Lafayette site for the data collection effort. The vehicle matching algorithms are hypothesized to perform comparably at 125 ft and 500 ft. Therefore, the Noblesville site is not likely to have enough distance to yield significantly different results than the West Lafayette site. The Noblesville site will have limited ground truth because the available video streams do not cover the advanced detectors. If a larger distance is required, the research team would need to resort to setting up temporary infrastructure for a limited collection at the I-70 site.

Figure 1: West Lafayette Intersection

Figure 2: West Lafayette Proposed Sensors for Data Collection

Figure 3: Camera NBC4 for West Lafayette Intersection

Figure 4: PTZ Camera at West Lafayette Intersection

Figure 5: Noblesville Intersection (SR 37 & SR 32)

Figure 6: Noblesville Proposed Sensors for Data Collection

Figure 7: Camera NBC1 for Noblesville Intersection

Figure 8: I-70 Highway Site

Figure 9: I-70 Proposed Sensors For Data Collection

Figure 10: Microloop Locations for I-70 data collection site

Figure 11: I-70 Data Collection Infrastructure

Table 1: Comparison of Sites