Embed Size (px)
Citation preview
VALIDATION OF THE SSAM TECHNIQUE FOR THE ASSESSMENT 1
OF INTERSECTIONS SAFETY 2
3
4
5
Luís Vasconcelos, MSc. 6
Adjunct Professor, Department of Civil Engineering 7
Polytechnic Institute of Viseu 8
Campus Politécnico de Repeses 9
3504-510 Viseu – Portugal 10
Phone: (+351) 232 480 500, E-mail: [email protected] 11
(corresponding author) 12
13
Luís Neto, MSc. 14
Department of Civil Engineering 15
University of Coimbra 16
Rua Luís Reis Santos - Pólo II 17
3030-788 Coimbra – Portugal 18
Phone: (+351) 239 797 100, E-mail: [email protected] 19
20
Álvaro Maia Seco, PhD. 21
Associated Professor, Department of Civil Engineering 22
University of Coimbra 23
Rua Luís Reis Santos - Pólo II 24
3030-788 Coimbra – Portugal 25
Phone: (+351) 239 797 100, E-mail: [email protected] 26
27
Ana Bastos Silva, PhD. 28
Assistant Professor, Department of Civil Engineering 29
University of Coimbra 30
Rua Luís Reis Santos - Pólo II 31
3030-788 Coimbra – Portugal 32
Phone: (+351) 239 797 100, E-mail: [email protected] 33
34
35
July 2013 36
Submitted for consideration for publication and presentation at the 93rd Annual 37
Meeting of the Transportation Research Board, January 12-16, 2014. 38
39
Resubmitted with reviewers comments addressed: November 10, 2013 40
41
Total number of words: 5580 (text) + 6 (figures/tables) = 7080 words (Max 7500 42
words). 43
1
ABSTRACT 44
45
The Surrogate Safety Analysis Module (SSAM) is a software application that reads 46
trajectory files generated by microscopic simulation programs and calculates surrogate 47
measures of safety. This approach eliminates the subjectivity associated with the 48
conventional conflict analysis technique and allows assessment of the safety of a facility 49
under a controlled environment before the occurrence of accidents. The specific goal of 50
this research is to validate SSAM as a tool for accident prediction at urban intersections. 51
Two methods are used: the first compares the simulated number of conflicts using SSAM 52
and the predicted number of injury accidents using analytic models in three reference 53
intersections layouts (4-leg intersection, 4-leg staggered intersection and single-lane 54
roundabout); the second compares SSAM results with conflicts observed on site in four 55
real intersections, two priority ones and two roundabouts. The results indicate that despite 56
some limitations related to the nature of current traffic microsimulation models, SSAM 57
analysis is a very promising approach to assessing the safety of new facilities or 58
innovative layouts. 59
60
61
62
2
INTRODUCTION 63
In urban areas traffic accidents are usually concentrated at intersections. Traditional 64
approaches to estimating the potential traffic accident risk of intersections, based on 65
historical crash data, include before-and-after analyses and accident prediction models. 66
Both approaches have important limitations related to the complexity of the safety factors 67
and the poor quality of data, thus: (i) compared to other traffic events, accidents are quite 68
exceptional in that they results from a series of unhappy improbable actions and 69
situations; (ii) accidents are rare events, so it is troublesome to base traffic safety analyses 70
at individual sites on accidents alone; (iii) not all accidents are reported and the level of 71
underreporting depends on the accident’s severity and types of road users involved; (iv) 72
information on the circumstances preceding the accident is seldom available (1). 73
Additionally, the fact that accidents must take place before one can determine the risk of 74
locations is, from an ethical point of view, a disadvantage (2). Therefore, specific 75
evaluation techniques are required to take into account changes in traffic regulations or 76
in the geometrical design of infrastructure, in order to estimate the true effects of safety 77
improvements. Moreover, regression models may not be transferable because they 78
implicitly reflect road users' behavior, the vehicle fleet and driving rules which vary from 79
country to country and even from site to site. 80
The traffic conflict technique is an approach that overcomes the lack of good and 81
reliable accident records, relying instead on observations of conflicts. A conflict is 82
defined as a situation in which two or more road users approach each other in time and 83
space such that there is risk of collision if their movements remain unchanged (3). The 84
necessary evasive action is usually braking, but may also be swerving or acceleration, or 85
a combination of them. In other words, conflicts are events that would result in an accident 86
if one of the drivers was not able to make an evasive maneuver in due time, since both 87
vehicles were in course to occupy the same space at the same time. Conflicts are far more 88
frequent than accidents and they are observable in real time at the site, allowing safety 89
assessments without the occurrence of accidents. While interest in the conflict technique 90
has been considerable, its practical use has been limited due to questions of subjectivity 91
in the registration process and the costs of data collection (4). Automatic tracing of 92
trajectories from video recordings is a recent and promising technology that addresses 93
these questions (5; 6) but the problem of predicting the safety benefit of a new geometry 94
or circulation scheme remains. Some functions linking simulated conflicts and crash 95
predictions were obtained (7-9), but the connection between surrogates and crashes is still 96
relatively unknown. 97
Microscopic simulation models are seen as promising tools to evaluate road safety 98
levels of existing and new infrastructures. The core of this new approach is the software 99
developed by the FHWA (Surrogate Safety Assessment Model - SSAM) that automates 100
conflict analysis by processing the vehicle trajectories produced during the simulation 101
(vehicle's position, speed and acceleration profiles). This approach has all the generic 102
advantages of simulation (possibility of assessing the safety of new facilities before the 103
occurrence of accidents, controlled environment, etc.) but also has some limitations: 104
common microscopic simulation models are developed for traffic-flow analyses and lack 105
3
some operational sub-models that are essential for safety analyses. Some authors have 106
proposed specific procedures to calibrate simulation models for safety assessment (10; 107
11) but this is still an ongoing research field. 108
Practical experience with SSAM is limited and the results are not consensual. 109
Gettman et al. (6) found a significant correlation between simulated conflicts and crashes 110
reported at 83 four-leg signalized intersections. Kim and Sul (12) tested the effect of 111
changing the speed limit of an arterial road in Sungnam, Korea and concluded that 112
VISSIM microsimulation represents speeds and flows well but is insufficiently detailed 113
in terms of safety analysis. Dijkstra et al. (13) modeled a 300 km2 road network in 114
PARAMICS and concluded that there was a significant relationship between the observed 115
crashes and the simulated conflicts. Huang et al. (14) compared observed and simulated 116
conflicts at ten signalized intersections in the Nanjing area in China. Their results show a 117
reasonable goodness-of-fit between simulated and recorded rear-end conflicts but they 118
also found that the simulated conflicts are not good indicators for traffic conflicts 119
generated by unexpected driving maneuvers, such as illegal lane-changes. 120
The specific goal of this research is to validate SSAM as a tool for accident 121
prediction in urban intersections. Two methods are used. The first compares the simulated 122
number of conflicts using SSAM and the predicted number of accidents using analytic 123
models in three fictitious intersections – one 4-leg priority intersection, one staggered 4-124
leg intersection, and one single lane roundabout – with simple geometry and idealized 125
traffic demand distributions. The second method compares the observed conflicts in four 126
selected intersections and the simulated conflicts using SSAM, emulating (as much as 127
possible) the real conditions observed at the different locations. 128
THE SSAM APPROACH 129
SSAM operates by processing data describing the trajectories of vehicles driving 130
through a traffic facility and identifying conflicts. For each vehicle-to-vehicle interaction 131
SSAM calculates surrogate measures of safety and determines whether or not that 132
interaction satisfies the criteria to be deemed a conflict. In the present analysis the 133
research team used time to collision (TTC) as a threshold to establish whether a given 134
vehicle interaction is a conflict and the relative speed (DeltaS) as a proxy for the accident 135
severity. Their definition is (6): 136
• TTC is the minimum time to collision value observed during the interaction of 137
two vehicles on collision course. If at any time step the TTC drops below a given 138
threshold (1.5 s in this work, as suggested for urban areas (15)), the interaction is tagged 139
as a conflict; 140
• DeltaS is the difference in vehicle speeds as observed at the instant of the 141
minimum TTC. More precisely, this value is mathematically defined as the norm of the 142
velocity vector of each vehicle, thus accounting for the differences in the vehicles’ 143
absolute speeds and headings. For example, if the speeds of two vehicles at the instant of 144
minimum TTC were 50 km/h and 30 km/h, then DeltaS would be 20 km/h if they were 145
following the same direction (rear-end accident), 80 km/h if they were driving in opposite 146
4
directions and 58.3 km/h if there were in following perpendicular directions (law of 147
cosines). 148
Aimsun (16) was the simulator used in this work. With the exception of the 149
parameters that control the drivers’ desired speed on links, most parameters kept their 150
default values and a single vehicle type (car) was considered. To increase the accuracy of 151
the surrogate measures (and specifically, for each vehicle conflict, the value of the 152
minimum TTC and the vehicles’ speeds at that instant) the simulation step was set to 0.25 153
seconds and drivers’ reaction time was set to 0.75 seconds. Further details about 154
alternative surrogate measures can be found elsewhere (1; 17). 155
CONCEPTUAL VALIDATION 156
Evaluation framework 157
The SSAM would ideally be validated against accident records. However, as 158
accidents are very rare, it would be very difficult to obtain enough data from a single site 159
under controlled external variables such as weather or traffic conditions to define a 160
reference scenario. As an alternative, conventional accident prediction models (APMs) 161
were taken as reference. These models are usually based on a large number of accident 162
records and rely on advanced statistical techniques to identify the significant variables 163
and calibrate the respective coefficients. A simple evaluation framework was designed to 164
assess the performance of SSAM: it consists of comparing the SSAM outputs against 165
accident estimates from the APMs for a set of layouts under the same traffic demand (see 166
Figure 1). It starts with a 4-leg yield-controlled intersection with a typical traffic demand 167
(70% traffic in the major E-W direction, 30% in the minor (N-S); at each entry, 30% turn 168
right, 50% move through and 20% turn left, Figure 1-a). We were interested in seeing if 169
the safety models can replicate the safety improvements arising from a left-right 170
staggering (reconfiguration of crossroads as two T-junctions, Figure 1-b) and with the 171
conversion to a single-lane roundabout (Figure 1-c). 172
The three layouts have standard geometric characteristics for urban conditions. The 173
4-leg intersection has left-turn lanes in the major direction and short right-turn lanes from 174
the minor direction. A conventional single-lane roundabout with four legs was also 175
considered, designed according to the Portuguese Roundabout Design Guidelines (18). It 176
has the following main geometric characteristics: inscribed circle diameter (ICD) – 40 m, 177
carriageway width (between lane markings) – 6.5 m, approach lane width – 3.5 m, entry 178
width – 4.5 m, entry radius – 30 m. The roundabout is symmetrical in both the N-S and 179
E-W directions. A 50km/h speed limit was assumed for all layouts. 180
These layouts were chosen because, first, there are several directly applicable 181
APMs (presented in the next section) and, second, their effects on safety are well 182
documented trough a large number of studies. With respect to the staggering, according 183
to Elvik et al. (19), junctions with four approaches (32 conflict points) make greater 184
demands on road user alertness and behavior than junctions with three approaches (9 185
conflict points). The effects of staggered junctions depend on the proportion of minor 186
road traffic at the crossroads before staggering and on the staggering side. For the 187
5
conditions presented on Figure 1-b (left-right staggering, heavy minor road traffic) one 188
can expect a 33% reduction in injury accidents and 10% reduction in property damage 189
accidents (19). Converting a crossroads to a single-lane roundabout is also a well-190
recognized way to improve safety. Several factors contribute to this: i) the number of 191
conflict points is reduced from 32 to 20; ii) all drivers must give way to circulating traffic 192
and are thus forced to watch traffic at the roundabout more carefully; iii) all traffic comes 193
from one direction, which facilitates finding a suitable gap to enter the roundabout; iv) 194
roundabouts make users drive around a circular traffic island located in the middle of the 195
intersection. This reduces the entry and circulating speeds, and the potential impact angle. 196
A meta-analysis of several studies indicates that the total number of severe accidents is 197
significantly lower on roundabouts (46% reduction for injury accidents) but property 198
damage accidents increase (+10%) (19). A recent before-and-after study in Denmark 199
reports decreases of 47% for injury and 16% and property-damage-only accidents (20). 200
For simulation purposes the daily traffic demand was split into 24 O/D matrices 201
based on a typical distribution throughout one day. Traffic demand assumed only cars. 202
203
4.5%
7.5%
3%
15%
28%
7%
35%
10.5%
31% 20.5%
4.5%
7.5%
3%
15%
28%
7%
35%
10.5%
31%20.5%
17.5%
7%
35%
10.5%
4.5%
7.5%
3%
15%
4.5%
7.5%
3%
15%
17.5%
7%
35%
10.5%
17.5%35%
10.5%
4.5%
7.5%
3%
15%
4.5%
7.5%
3%
15%
17.5%
7%
35%
10.5%
7%W
S
E
N
W
E
N
S
W
N
S
E
b)
a) c)
204 205
Figure 1 - Conceptual validation - geometric layouts: a) 4-leg priority intersection, b) staggered 4-206 leg-intersection (modeled as two 3-leg intersections), c) single-lane roundabout 207
6
Accident Prediction Models 208
At least two different models were used for each layout analyzed. The accidents in 209
the priority intersections (3-leg and 4-leg, see Table 1) were estimated using models 210
developed in the United States (21), United Kingdom (22) and Portugal (23). Each of 211
these models gives the predicted number of total intersection-related injury accidents per 212
year, excluding pedestrians. The data samples vary considerably between studies: the US 213
models are based on a sample of 324 four-leg STOP-controlled intersections and 382 214
three-leg STOP-controlled intersections in Minnesota and they include 5 years of accident 215
data (1985-1989) for each intersection. The UK model was based on data for 3800 km of 216
highway in the UK including more than 5000 minor junctions. The Portuguese dataset 217
included 44 locations for the 3-legged intersections and 50 locations for 4-legged 218
intersections, located in Lisbon. 219
220 Table 1 – Accident prediction models for 3-leg and 4-leg priority intersections 221
Model type Origin Expression
3-leg
priority
intersection
USA:
Harwood and Council
(21)
1 210,9 0,79ln 0,49ln
1 2 3 4 ADT ADT
A AMF AMF AMF AMF e
UK: Maher and
Summersgill (22)
0.80 0.36
1 2
1000 10 0.0
049
0
IF IFA
PORTUGAL:
Vieira Gomes (23) 6 1.18844.7078 10A IF
4-leg
priority
intersection
USA:
Harwood and Council
(21)
1 29,34 0,6ln 0,61ln
1 2 3 4 ADT ADT
A AMF AMF AMF AMF e
PORTUGAL:
Vieira Gomes (23) 5 1.1673.8765 10A IF
Where: A is the predicted number of total intersection-related accidents per year. USA - AMF1 is the 222 accident modification factor for there being a left-turn lane on major road (0.78 for one major-road 223 approach, 0.58 for both major-road approaches), AMF2 is the accident modification factor for there being a 224 right-turn lane (0.95 for a right-turn lane on one major-road approach, 0.90 for right-turn lanes on both 225 major-road approaches), AMF3 is the accident modification factor for the sight restrictions, AMF4 the 226 accident modification factor for the conversion from minor road to all-way stop-control, ADT1 is the annual 227 average daily traffic on the major road, ADT2 is the annual average daily traffic on the minor road. Portugal 228 and UK: IF, IF1 and IF2 – total inflow in the intersection, major direction and minor direction, respectively 229 (veh/day, annual average) 230
231
The accidents in the roundabout were estimated using models developed in the 232
United Kingdom (24), Australia (25) and Portugal (23) (see Table 2). The British model 233
is based on accidents that occurred over a six-year period on 84 roundabouts in the UK. 234
Some of the sub-models obtained were excluded because they refer to causes that are 235
beyond the scope of the SSAM methodology (namely the sub-model A3 - single-vehicle 236
accidents and sub-model A5 – accidents involving a pedestrian casualty). Arndt (26) 237
analyzed one hundred Queensland (Australia) roundabouts, where 492 major accidents 238
were recorded, generally over a five-year analysis period. The main objective of that study 239
was to determine the effect of roundabout geometry on accident rates. Models were 240
developed for single vehicle accidents, approaching rear-end accidents and 241
7
entering/circulating vehicle accidents. For the reasons stated in the previous point, the 242
first type was excluded from the current analysis. Vieira Gomes (23), using data from 15 243
roundabouts in Lisbon, Portugal, developed models to estimate the frequency of accidents 244
with injuries on urban road networks. 245
246
Table 2 – Accident prediction models for roundabouts 247
Origin Conflict
type Expression
UK:
Maycock
and Hall
(24)
Entry-circ.
0.7 0.4
1
10.052 exp 40 0.14 0.007 0.2 0.01
1 exp 4 7e c e mA Q Q C e ev P
R
Approach 1.7
2 0.0057 exp 20 0.1e eA Q C e
Other 0.8 0.8
4 0.0026 exp 0.2e c mA Q Q P
Australia:
Arndt
(25)
Rear-end 11 0.5 29.62 10r a cA Q Q S
Entry-circ. 12 0.5 23.45 10c a ci iA Q Q S
Portugal:
Vieira
Silva (23)
All 8 1.5084 0.52482.3845 10 e LEGA FT
Where : UK - Ai are personal injury accidents (including fatalities) per year per roundabout approach (A1 248 entering-circulating accidents, A2 approaching accidents (mostly rear-ends, but also changing lane 249 accidents, A4 other accidents (variety of non-pedestrian accidents)), Qe and Qc are the entering and 250 circulating flow, respectively (1000s of veh/day), Ce is the is entry curvature (Ce = 1/Re and Re is the 251 entry path radius for the shortest vehicle path (m)), e is the entry width (m), v is the approach width, R is 252 the radius of the inscribed circle diameter, Pm is the proportion of motorcycles (%) and θ is the angle to 253 next leg measured centerline to centerline (degrees, °). Australia - Ar / Ac are, respectively, rear-end and 254 entering/circulating accidents per year (over $1000 property damage and/or personal injury), Qa is the 255 average annual daily traffic on the approach, i.e., one-way traffic only (veh/d), Qci are the various average 256 annual daily traffic flows on the circulating carriageway adjacent to the approach, i.e., one-way traffic only 257 (veh/d), Si are the various relative 85th percentile speed between vehicles on the approach curve and 258 vehicles on the circulating carriageway from each direction (km/h). Portugal - A is the estimated number 259 of accidents with injuries per year in the roundabout, FT is total entering traffic flow in vehicles per day 260 (AADT), LEG is the number of legs of the roundabout 261
Results 262
The purpose of this first test was to compare the SSAM and APM results for 263
uniform traffic growth, when the total AADT (entering traffic) varies from 0 to 20000 264
veh./day in the three layouts. 265
Figure 2 shows the results for both the simulated conflicts and predicted accidents 266
in all three types of intersections. We can infer that although the regression models relate 267
to different countries and are based on different assumptions, they still provide consistent 268
estimates. The SSAM conflict-flow points are almost perfectly fitted by a convex curve, 269
similar to the accident-flow curves from the regression models, although with a more 270
pronounced curvature, which suggests a good correlation between simulation conflicts 271
and actual accidents. 272
All APMs predict a reduction in the number of injury accidents with the staggering 273
of the 4-leg intersection and a further reduction following the conversion to a roundabout, 274
which agrees with expectations and with international experience. The SSAM method 275
8
predicts a reduction in the number of conflicts in the staggered intersection (sum of the 276
two T-intersections), as expected, but it also predicts an increase on the roundabout. To 277
assess the severity level of the simulated conflicts, the average speed differential (DeltaS) 278
was calculated for each intersection and the following values were obtained: 4-leg 279
intersection - 6.46 m/s, staggered junctions - 6.19 m/s, roundabout – 4.40 m/s. SSAM 280
predicts a large number of low severity conflicts at roundabouts. This agrees with the 281
above mentioned typical significant reduction of injury accidents at roundabouts, 282
associated with an increase of minor accidents. 283
284
285
286
287
Figure 2 – Accident – Flow relationships from analytic regression models and Conflict – Flow 288 relationship from the SSAM technique 289
0
500
1000
1500
2000
0
1
2
3
4
5
0 5000 10000 15000 20000
Co
nfl
icts
per
da
y
(SS
AM
)
Acc
iden
ts p
er y
ear
(AP
M)
Total demand flow (veh/day)
4-Leg intersection
Harwood & Council
Vieira Gomes
SSAM
0
500
1000
1500
2000
0
1
2
3
4
5
0 5000 10000 15000 20000
Co
nfl
icts
p
er d
ay
(S
SA
M)
Acc
iden
ts p
er y
ear
(AP
M)
Total demand flow (veh/day)
Staggered junctions
Harwood & Council
Maher and Summersgill
Vieira Gomes
SSAM
0
500
1000
1500
2000
0
1
2
3
4
5
0 5000 10000 15000 20000
Co
nfl
icts
per
da
y
(SS
AM
)
Acc
iden
ts p
er y
ear
(AP
M)
Total demand flow (veh/day)
Roundabout
Maycock and Hall
Arndt
Vieira Gomes
SSAM
9
The graphs show that the relation between conflicts and accidents is not linear, as 290
it varies with the entry flow, the intersection type and with the APM in question. 291
Nevertheless, it is possible to obtain some approximate figures that may be useful to 292
narrow down the uncertainty in this research area. For example, for a 12000 vehicles/day 293
entry flow, depending on the APM we can expect to find a conflict/accident ratio of 25000 294
– 30000 in a 4-leg intersection, 15000 – 60000 in the staggered 4-leg intersection and 295
180000 – 500000 on the roundabout. These ratios increase with the demand flow, as 296
expected, due to the increasing difficulty of maintaining high speeds, which leads to lower 297
probability of a conflict resulting in an accident with injuries. 298
FIELD VALIDATION 299
The objective of this section was to understand how conflicts predicted by the 300
Aimsun-SSAM approach correlate with real conflicts observed in urban intersections. 301
Four intersections were selected within the city of Coimbra, Portugal: one three-leg stop 302
controlled intersection, one four-leg stop controlled intersection, one four-leg single-lane 303
roundabout, and one five-leg two-lane roundabout. The main geometric and traffic 304
characteristics of these intersections are presented in the next section. 305
Observation of conflicts 306
The conflict observation followed the guidelines presented in the FHWA Traffic 307
Conflict Observers Manual (27), with minor adjustments aiming at comparability with 308
the simulation outputs. Therefore, only vehicle-vehicle interactions were considered. 309
Pedestrians were not regarded as road users but could be a cause of a conflict, particularly 310
rear-end conflicts between two vehicles forced to stop for a pedestrian in a crosswalk. 311
The severity of conflicts was not assessed, meaning that there was no limit on time-to-312
collision or post-encroachment time for a conflict to be registered. As such, observers 313
were expected to register a conflict every time a road user with the right of way was forced 314
to modify their behavior, either by braking or swerving. 315
Each observer was responsible for a certain direction of vehicles with the right of 316
way (e.g.: one observer would only register conflicts involving northbound vehicles in 317
the major road while the other was responsible for the southbound vehicles). This meant 318
that conflicts in minor roads and after the intersection (from each observer’s point of 319
view) were ignored. In roundabouts, each observer was responsible for a quarter of the 320
circulatory carriageway. Rear-end conflicts in the approaches were disregarded. Because 321
of the methodology used for conflict observation on site and in order to properly compare 322
the number of observed and simulated conflicts, only the conflicts that took place in the 323
intersection itself or in the major road immediately before the intersection were 324
considered. In the case of roundabouts, only the conflicts taking place in the circulatory 325
carriageway were counted. 326
There was a trial period during which we tested this method and made sure that 327
every observer was on the same page, so that the results were as similar as possible. At 328
least one of the authors was permanently on the field accompanying the sessions and 329
helping observers to decide on the less clear cases. 330
10
The observations were taken for nine hours per location, each on a different day, 331
divided into three periods: the morning period from 7:30 am to 10:30 am, the noon period 332
from 12:00 pm to 15:00 pm, and afternoon period from 16:00 pm to 19:00 pm. These 333
periods cover different traffic demand levels and directional splits. The total workload 334
was 108 person×hours. 335
336
Development of the microscopic models 337
Other observers were responsible for collecting the data required for the 338
microsimulation modeling: traffic volumes per movement, pedestrian movements in 339
crosswalks, bus stop frequency (if needed), and vehicle approach speeds using a laser gun 340
(LIDAR). The parameters that control the speed of vehicles under free-flow conditions 341
(desired speeds and levels of compliance with speed limits) were adjusted in the model 342
to correspond to field measurements. 343
Pedestrians are traditionally simulated in Aimsun by using an embedded pedestrian 344
simulator engine (a plugin provided by Legion, a company specializing in the field of 345
pedestrian simulation). Unfortunately, SSAM is unable to assess pedestrian-pedestrian 346
and vehicle-pedestrian interactions, so we decided to emulate pedestrian crossing events 347
using Aimsun’s “periodic section incidents” feature. This feature allows a specified 348
segment of a road (the crosswalk) to be blocked to vehicles during short time intervals, 349
with a duration and frequency typical of the observed crossings. 350
The default values of the remaining parameters that control the car-following and 351
gap-acceptance behavior were not changed. A comprehensive calibration was not deemed 352
necessary as the main objective of this analysis was to compare alternative layouts, so the 353
absolute values are not of significant importance. Furthermore, keeping default values 354
facilitates future comparisons with similar studies. 355
356
Locations 357
Three-leg intersection 358
The chosen intersection is located on an arterial road within an urban environment. 359
The main street is north-south oriented and has an estimated total entering AADT of 360
14483 vehicles per day, with evenly balanced flows for most of the day, but mainly 361
northbound in the morning peak hour. The street is approximately 7 meters wide and has 362
one lane each way. No left-turn lane is provided on the major road. The minor street 363
approaches the main one from west and has a considerable entering AADT of 8245 364
vehicles per day, with most of the movements being northbound. The street is 365
approximately 12 meters wide and has one lane for the westbound traffic and two lanes 366
for the eastbound traffic, thus allowing segregated right and left turns. There is also a one 367
way street approaching the main road from the east, just a few meters to the south of the 368
intersection, which was disregarded due to its low traffic volume. Additionally, there is 369
one bus stop on each side of the major street, located approximately 60 meters to the north 370
11
of the intersection. This is a very significant factor for the traffic flow, both southbound 371
and northbound, since there is no bay and the buses have to stop in the main road. Other 372
constraints are a couple of crosswalks immediately to the west and south of the 373
intersection. Both have residual pedestrian traffic, crossing every other 10 or 15 minutes. 374
The average free flow speed while approaching the intersection was 46.0 km/h with a 375
standard deviation of 5.4 km/h. 376
Four-leg intersection 377
The chosen intersection is located in the heart of a residential district, where most 378
of the neighborhood’s commerce can be found. The main street is north-south oriented 379
and has an estimated entering AADT of 5135 vehicles per day, mainly northbound, 380
particularly in the morning peak hour. The street is approximately 9 meters wide, with 381
one lane each way, and parallel parking is allowed in both ways. The minor street is east-382
west oriented and has an estimated entering AADT of 1924 vehicles per day. The street 383
is approximately 8 meters wide, with one lane each way, allowing parallel parking for the 384
eastbound vehicles on the eastern approach and diagonal parking for both ways on the 385
western approach. Perpendicular parking is allowed within the intersection in the 386
northeast and southeast corners. This fact was ignored during the traffic conflict 387
observations since it was very difficult to emulate in the microsimulation software. There 388
are crosswalks in all of the four approaches. Since it is a residential area pedestrian traffic 389
is a significant factor. As such, we have registered the pedestrian crossing frequency for 390
each crosswalk. The average free flow speed while approaching the intersection was 39.6 391
km/h with a standard deviation of 4.9 km/h. 392
One-lane roundabout 393
The chosen roundabout is located in one of the most densely populated areas of the 394
city. This node connects three arterial streets and one local street. The roundabout has a 395
diameter of 36 meters and a 6 meters wide circulatory lane. The total estimated AADT is 396
22938 vehicles per day. There are crosswalks in all of the four approaches. Nevertheless, 397
even though it is located in a central area, there is not much pedestrian movement in this 398
intersection. Given that, crosswalks were not considered in the model. The average free 399
flow speed while approaching the roundabout was 46.4 km/h with a standard deviation 400
of 6.7 km/h. 401
Two-lane roundabout 402
The chosen roundabout is located in one of the most densely populated areas of the 403
city. This node connects three very important arterial streets and two local streets, one of 404
which is a one way street towards the intersection. This intersection is within Coimbra’s 405
ring road and there is a considerable amount of heavy vehicle traffic. Based on a partial 406
observation, it was assumed that heavy vehicles represented 5% of the total traffic. The 407
total estimated AADT is 33625 vehicles per day. There are crosswalks in all of the four 408
approaches, however, for the abovementioned reasons, crosswalks were not considered 409
12
in the model. The average free flow speed for cars while approaching the roundabout was 410
60.0 km/h with a standard deviation of 7.8 km/h. 411
Results 412
The SSAM analysis produced the results illustrated in Figure 3 for the noon period. 413
The graphical display of conflicts agrees reasonably with observations: i) most approach 414
legs have low severity rear-end conflicts. These occur at peak periods when, for several 415
reasons, priority drivers slow down and delay upstream vehicles; ii) at the 4-leg 416
intersection the model predicts a high-percentage of severe conflicts, which actually occur 417
and are related to perpendicular trajectories and to sudden stops in response to pedestrians 418
at crosswalks; iii) at the roundabouts, the model predicts entering-circulating and weaving 419
conflicts where they actually occur. 420
421
422
Figure 3 - SSAM results (noon period): graphical display of conflicts, colored by relative speed 423 (proxy for the accident severity) 424
DeltaS (Relative Speed) < 5 m/s 5 - 10 m/s > 10 m/s
a)
b)
c)
d)
13
The comparison between observed and simulated conflicts is presented in Figure 4. 425
As noted above, only the conflicts that take place in the priority links are registered (major 426
direction links in the priority intersections, circulatory lanes on the roundabouts). SSAM 427
presents a systematic underestimation trend, probably related to the TTC threshold 428
adopted, but in global terms it is able to replicate the pattern of real conflicts in the four 429
intersection types and for the different demand periods. The least satisfactory results were 430
obtained at the single-lane roundabout and are related to a pedestrian crossing in the south 431
leg, not included in the model, which was sometimes responsible for upstream queues 432
that reached the circulatory lane. 433
434
Figure 4 - Comparison of observed and simulated total conflicts at the four sites 435
436
SUMMARY AND CONCLUSIONS 437
Safety assessment based on conventional accident prediction models raises 438
questions about the availability and quality of crash data to serve as reference for model 439
development, and it is not feasible for studying new layouts or facilities operating outside 440
the models’ calibration domain. Recently, efforts have been made to expand the use of 441
microscopic simulation models to safety assessment problems. Our study evaluated the 442
potential of the SSAM approach to assess urban intersections’ safety levels. A conceptual 443
validation, based on conventional accident prediction models (APMs), showed a strong 444
0
40
80
120
160
200
Morning Noon Afternoon
3-Leg Intersection
Simulated conflicts Observed Conflicts
0
40
80
120
160
200
Morning Noon Afternoon
Single-lane Roundabout
Simulated conflicts Observed Conflicts
0
40
80
120
160
200
Morning Noon Afternoon
4-Leg Intersection
Simulated conflicts Observed Conflicts
0
40
80
120
160
200
Morning Noon Afternoon
Two-lane Roundabout
Simulated conflicts Observed Conflicts
14
relation between accidents predicted by the regression models and conflicts predicted by 445
simulation models. Conflict/accident ratios were found to vary according to the 446
intersection type, the entry flow and the APM in question. For a moderate traffic flow 447
(entering AADT = 12000 vehicles/day) that ratio is about 25000-30000:1 in a 4-leg 448
intersection and 180000-500000:1 on a single-lane roundabout. 449
A field validation compared simulated and observed conflicts in four types of 450
intersection. Despite a systematic underestimation trend, SSAM was able to replicate the 451
hourly evolution of conflicts and to identify the hazardous areas of each intersection. The 452
sub-optimal results at the single-lane roundabout indicate that this safety approach is quite 453
sensitive to inaccurate modeling. Nevertheless, it can be concluded that SSAM analysis 454
is a very promising approach to assessing the safety of new facilities, of innovative 455
layouts (e.g. turbo-roundabouts (28)) or of traffic regulation schemes (e.g. limiting the 456
driving degree of freedom on freeways(8)) . 457
ACKNOWLEDGMENTS 458
This research is partially funded by FEDER (Programa Operacional Factores de 459
Competitividade – COMPETE) and FCT (Fundação para a Ciência e a Tecnologia) in the 460
scope of projects PTDC/SEN-TRA/122114/2010 (AROUND – Improving Capacity and 461
Emission Models of Roundabouts) and EMSURE - Energy and Mobility for Sustainable 462
Regions (CENTRO-07-0224-FEDER-002004). 463
464
465
REFERENCES 466
[1] Laureshyn, A., Å. Svensson, and C. Hydén. Evaluation of traffic safety, based on micro-level 467 behavioural data: Theoretical framework and first implementation. Accident Analysis & Prevention, Vol. 468 42, No. 6, 2010, pp. 1637-1646. 469 [2] Güttinger, V. A. Conflict Observation in Theory and in Practice.In International Calibration Study of 470 Traffic Conflict Techniques, No. 5, Springer Berlin Heidelberg, 1984. pp. 17-24. 471 [3] Amundsen, F., and C. Hyden. Proceedings of first workshop on traffic conflicts, Oslo: Institute of 472 Transport Economics, 1977. 473 [4] Asmussen, E. Joint International Study for the Calibration of Traffic Conflicts Techniques.In 474 International Calibration Study of Traffic Conflict Techniques, No. 5, Springer Berlin Heidelberg, 1984. 475 pp. 1-6. 476 [5] Ardö, H., M. Nilsson, A. Laureshyn, and A. Persson. Enhancements of traffic micro simulation 477 models using video analysis. Presented at The 17th European Conference on Mathematics for Industry, 478 Lund, Sweden, 2012. 479 [6] Gettman, D., L. Pu, T. Sayed, and S. G. Shelby. Surrogate safety assessment model and validation: 480 final report.In, FHWA, 2008. 481 [7] Tarko, A. P. Use of crash surrogates and exceedance statistics to estimate road safety. Accident 482 Analysis & Prevention, Vol. 45, No. 0, 2012, pp. 230-240. 483 [8] Habtemichael, F. G., and L. de Picado-Santos. The impact of high-risk drivers and benefits of limiting 484 their driving degree of freedom. Accident Analysis & Prevention, 485 http://dx.doi.org/10.1016/j.aap.2013.05.013. 486 [9] Al-Ghandour, M., B. Schroeder, B. Williams, and W. Rasdorf. Conflict Models for Single-Lane 487 Roundabout Slip Lanes from Microsimulation. Transportation Research Record: Journal of the 488 Transportation Research Board, Vol. 2236, No. -1, 2011, pp. 92-101. 489
15
[10] Cunto, F., and F. F. Saccomanno. Calibration and validation of simulated vehicle safety performance 490 at signalized intersections. Accident Analysis & Prevention, Vol. 40, No. 3, 2008, pp. 1171-1179. 491 [11] Duong, D., F. F. Saccomanno, and B. R. Hellinga. Calibration of microscopic traffic model for 492 simulating safety performance. Presented at 89th Annual TRB Conference, Washington, DC, 2010. 493 [12] Kim, K.-J., and J. Sul. Development of Intersection Traffic Accident Risk Assessment Model. 494 Presented at 4th IRTAD Conference, Seoul, Korea, 2009. 495 [13] Dijkstra, A., P. Marchesini, F. Bijleveld, V. Kars, H. Drolenga, and M. van Maarseveen. Do 496 Calculated Conflicts in Microsimulation Model Predict Number of Crashes? Transportation Research 497 Record: Journal of the Transportation Research Board, Vol. 2147, No. -1, 2010, pp. 105-112. 498 [14] Huang, F., P. Liu, H. Yu, and W. Wang. Identifying if VISSIM simulation model and SSAM provide 499 reasonable estimates for field measured traffic conflicts at signalized intersections. Accident Analysis & 500 Prevention, Vol. 50, No. 0, 2013, pp. 1014-1024. 501 [15] Svensson, A. A method for analysing the traffic process in a safety perspective.In, No. Doctoral 502 Dissertation, Lund Institute of Technology, 1998. 503 [16] Aimsun. Aimsun Version 7 User's manual, TSS - transport simulation systems.In, Barcelona, Spain, 504 2012. 505 [17] Gettman, D., and L. Head. Surrogate Safety Measures from Traffic Simulation Models. 506 Transportation Research Record: Journal of the Transportation Research Board, Vol. 1840, No. -1, 507 2003, pp. 104-115. 508 [18] Bastos Silva, A., and Á. J. M. Seco. Dimensionamento de Rotundas - Documento Síntese.In 509 Disposições Normativas, INIR - Instituto de Infraestruturas Rodoviárias, 2012. 510 [19] Elvik, R., A. Høye, T. Vaa, and M. Sørenson. Handbook of Road Safety Measures (2nd Edition).In, 511 Emerald, Inc., 2009. 512 [20] Jensen, S. U. Safety effects of converting intersections to roundabouts. Presented at TRB 2013 513 Annual Meeting, Washington DC, 2013. 514 [21] Harwood, D. W., F. Council, E. Hauer, W. Hughes, and A. Vogt. Prediction of the expected safety 515 performance of rural two-lane highways.In, No. FHWA-RD-99-207, Federal Highway Administration 516 (FHWA), 2000. 517 [22] Maher, M. J., and I. Summersgill. A comprehensive methodology for the fitting of predictive 518 accident models. Accident Analysis & Prevention, Vol. 28, No. 3, 1996, pp. 281-296. 519 [23] Vieira Gomes, S. The influence of the infrastructure characteristics in urban road accidents 520 occurrence. Accident Analysis & Prevention, http://dx.doi.org/10.1016/j.aap.2013.02.042, No. 0, 2013. 521 [24] Maycock, G., and R. Hall. Accidents at 4-arm roundabouts.In Highway appraisal and design - PTRC 522 Summer Annual Meeting, Sussex, England, 1984. 523 [25] Arndt, O., and R. Troutbeck. Relationship between roundabout geometry and accident rates. 524 Transportation Research Circular, No. E-C003, 1998, pp. 28:21-16. 525 [26] Arndt, O. Relationship Between Unsignalised Intersection Geometry and Accident Rates. 526 Queensland University of Technology, Brisbane, 2004. 527 [27] Parker Jr, M., and C. V. Zegeer. Traffic Conflict Techniques for Safety and Operations. Observers 528 Manual.In, FHWA, 1989. 529 [28] Vasconcelos, A. L. P., A. Bastos Silva, Á. J. M. Seco, P. J. Fernandes, and M. C. Coelho. Turbo 530 roundabouts: A multi-criteria assessment on intersection capacity, safety and emissions. Presented at TRB 531 93rd Annual Meeting, Washington, D.C., 2014. 532 533
