Sei sulla pagina 1di 15

Accid. Anal. and Prev., Vol. 26, No. 5, pp.

593-607, 1994
Pergamon Copyright 0 1994 Elsevier Science Ltd
Printed in the USA. All rights reserved
0001-4575/94 $6.00 + .00
0001_4575(94)E0002-3

SIMULATION OF TRAFFIC CONFLICTS AT


UNSIGNALIZED INTERSECTIONS WITH TSC-Sim

TAREK SAYED, GERALD BROWN, and FRANCIS NAVIN

University of British Columbia, Vancouver, B.C., Canada V6T 124

(Accepted 27 September 1993)

Abstract-This paper describes a traffic conflicts computer simulation model and graphic display for both
T and 4-leg unsignalized intersections. The goal of the model is to study traffic conflicts as critical-event
traffic situations and the effect of driver and traffic parameters on the occurrence of conflicts. The analysis
extends conventional gap acceptance criteria to describe driver’s behaviour at unsignalized intersections by
combining some aspects of gap acceptance criteria and the effect of several parameters including driver’s
characteristics such as age, sex, and waiting time. The effect of different traffic parameters such as volume
and speed on the number and severity of traffic conflicts is also investigated. The model is unique insofar
as it uses a technique of importance sampling and stores the traffic conflicts that occur during the simulation
for later study. A graphical animation display is used to show how these conflicts occurred and the values
of critical variables at the time. Model results were evaluated against previous work in the literature and
validated by using field observations from four unsignalized intersections. The simulation results correlated
reasonably well with actual conflict observations and should prove useful for assessing safety performance
and feasible solutions for other unsignalized intersections.

Keywords-Traffic conflicts, Simulation models, Unsignalized intersections, Driver behavior

INTRODUCTION mechanism, particularly the crash-avoidance behav-


iour of drivers.
Road traffic analysis is becoming more computation-
The goal of this paper is to outline a traffic
ally intensive in order to better understand the mech-
conflict simulation model for unsignalized intersec-
anism of traffic operations and road safety dynamics.
One emerging area is the study of extreme events, tions and to report on tests undertaken to validate
such as traffic crashes, and the conceptual and math- the model. Computer simulation is employed to
ematical techniques that can be used to study such study the problem because of the uncertainty and
problems (Navin 1991). Traffic crashes at intersec- complex relationships associated with human be-
tions are pervasive road systems failures, but little haviour and the difficulty of controlling different
is presently known about the failure mechanism. aspects of that behaviour and the parameters of
Because of well-recognized measurement and statis- systems operation by the use of conventional mathe-
tical problems with accident records and the site- matical methods. The model is unique insofar as it
specific focus (“black spots”) of many accident uses a technique of importance sampling, that is,
studies, the observation of traffic conflicts as a road- the model saves characteristic information for only
user behavioural factor has been advocated as a those conflicts that are predefined to be “signifi-
procedure to study the vehicle crash failure mecha- cant” events and stores these events as they occur
nism from a somewhat broader perspective than ac- during simulation for later study. A graphical anima-
cident statistics alone (Brown 1991). However, in- tion display is used to show how the conflict oc-
complete conceptualization and the cost of training curred and the value of critical variables at that time.
observers and collecting traffic conflict data have While there is considerable literature on simula-
been factors inhibiting extensive application of the tion models for unsignalized intersections, most of
technique, and thus successful simulation of traffic these models consider intersection capacity and sim-
conflicts appears to have practical benefits for traffic ulate how traffic volume affects level of service and
studies application as well as providing a more com- delay. Only a few authors such as Cooper and Fergu-
plete understanding of the vehicle crash failure son (1976) and McDowell et al. (1983) consider simu-
593
594 T. SAVED et al

lating traffic conflicts at unsignalized intersections. two vehicles to collide if they continue at their pres-
Models that deal with traffic conflicts at intersections ent speed and on the same path.” The value of TTC
tend to oversimplify the driver’s behaviour; they is infinite if the vehicles are not on a collision course;
adopt fixed gap-acceptance criteria to describe how but if the vehicles are on collision course, the value
drivers make decisions and they neglect to take into of TTC is finite and decreases with time. The mini-
account important aspects such as the effect of the mum TTC reached as the vehicles approach on the
stopped delay on the driver’s behaviour. Conse- collision course is taken as the critical conflict sever-
quently, the simulation model reported on here is ity. Hayward suggested a minimum TTC value of
multidimensional, incorporating some aspects of gap one second as a threshold value. Since Hayward, a
acceptance criteria as well as the effect of several number of field and experimental studies using time-
other parameters that are, a priori, considered useful based measures indicate a desirable driver safety
to describe that behaviour. “space” of 1.5 seconds and a minimum space of
1.0 seconds (van der Horst 1990). A closed-course
experiment conducted by van der Horst and Brown
TRAFFIC CONFLICTS SIMULATION
(1989) to confirm these values resulted in critical
The traffic conflicts concept was first proposed TTC measures of t(,) = 1.6 seconds and t(,inJ = 1.1
by Perkins and Harris (1967). As an alternative to seconds. From these results, it was decided to use
accident data, which in many cases are scarce, un- values of t(,) = 1.5 seconds for the critical event in
available, or unsatisfactory, their objective was to the simulation process.
define traffic events or incidences that occur fre-
quently, can be clearly observed, and are related
THE SIMULATION MODEL
to accidents. A traffic conflict was defined as any
potential accident situation leading to the occur- Because there are few applications of traffic
rence of evasive actions such as braking or swerving. conflict studies, there are few data available to de-
This simple definition has since been refined to incor- velop mathematical safety models; and the lack of
porate categories of vehicle manoeuvres and mea- conceptual precision of existing models attests to
sures of time and space proximity between vehicles the complexity and stochastic nature of the traffic
at the moment of conflict. An internationally ac- process. Simulating traffic conflicts assumes that the
cepted definition of a traffic conflict is given by incidence and severity of traffic conflicts have sto-
Amundson and Hyden (1977), “A conflict is an ob- chastic event characteristics similar to other critical
servable situation in which two or more road users events of the traffic process such as vehicle arrivals
approach each other in space and time to such an and accidents. The simulation is meant to emulate
extent that there is a risk of collision if their move- this traffic process at unsignalized intersections as
ments remain unchanged.” An expanded concept follows: (i) vehicles with a random set of characteris-
can be found in Brown (1994). tics are generated on the approaches; (ii) on arriving
For simulation, a traffic conflict needs to be at the intersection from a subject approach, an ac-
described and scaled as an unequivocal, observable ceptance criterion is determined based on a priori,
measure of the systematic variability of risk as per- realistic assumptions; (iii) a “consistent” behaviour
ceived by the driver in a given traffic situation. A model assumes a minimum gap that is acceptable to
variety of observation methods have been developed each driver at all times with variation across drivers
to scale the severity of risk as measured by traffic based on the type of traffic control, approach speed,
conflicts. Conventionally, these methods are driver age and sex, and stopped delay. A traffic
thought of as subjective methods and objective conflict is recorded by the simulation when a driver
methods. In the former, one can find terms such as accepts a gap, that, given the closing speed of the
euasive uction or s~~~~~ ~e~uui~ur as part of the conflicting vehicles, puts him/her at risk of collision
definition. These methods include considerable with the other vehicle.
judgement by the observer and are criticised by re-
searchers such as Hauer (1978) and Allen, Shin, and Structure, ussumptions, and input parameters
Cooper (1977) because the grading of the severity The model is built using a microcomputer ver-
of the evasive action can vary greatly from one ob- sion of the discrete event simulation language, Gen-
server to another, particularly if the observers are eral Purpose Simulation System, or GPSSlH (Wol-
not well trained. Objective methods incude a verine Software Corporation, Annandale, VA,
cardinal or ordinal time-proximity dimension in the USA), a specialized language. The basic elements
severity scale. Hayward (1972) used the time-to- of GPSSlH models are blocks and transactions in
collision (TTC) measure defined as “the time for which transactions are dynamic entities (such as
Simufation of traffic conflicts 595

road vehicles) that move from one block to another. of c = 1.5 seconds is used as a minimum allowable
Blocks are actions (e.g. GENERATE, ADVANCE, headway between vehicles in the same lane. If the
QUEUE, LEAVE) that affect the transactions and generated headway is less than the minimum allow-
other system entities, such as statistical, computa- able headway, it is set to the minimum. In this case,
tional, and resource entities. A full description is the vehicle is considered a member of a platoon.
found in Schriber (1974). The model also has some The model uses a random number generator and a
features not normally part of GPSS/H and therefore shifted exponential function to produce the required
has been labelled “TSC-Sim,” for Traffic Systems headway. Va~ability in the random number genera-
Conflict Simulation. tor may generate a vehicle flow rate different from
The model is microscopic since it deals with that required. This problem is solved by calculating
individual vehicles as they approach, go through, the llow rate generated by the mode1 every four
and depart the intersection. Actions for vehicles in minutes. The difference between this flow rate and
the model include vehicle generation, approach to the required one is calculated and then used to adjust
the intersection, choosing a gap (lag), and proceed- the average headway used in the GENERATE block
ing to depart. The assumptions of the model TSC- of the model.
Sim are as follows: (i) there is no overtaking or lane After vehicles are generated, a lane-selection
changing allowed at the intersection, (ii) the subject process is used to ensure that vehicles will reach
intersection is isolated, (iii) all drivers have an unob- their desired destination. When a lane is selected,
structed view of the intersection, (iv) there is no vehicles proceed and try to enter the intersection.
pedestrian inte~erence, (v) all drivers must maintain The minor road consists of two sections: the ap-
at least a minimum headway between their vehicle proach section and the decelerating and/or queuing
and the vehicle in front, (vi) all drivers looking for section. At the approach section vehicles have lower
an acceptable gap have perfect knowledge about the speeds than the desired speed, as they are preparing
movement of vehicles having higher priority and to negotiate the intersection. The queuing section is
fully understand the rules of the road. A gap is de- the location in which vehicles decelerate and in some
cases have to stop according to the number of vehi-
fined at the time headway between two successive
cles in the queue and the type of intersection control.
vehicles in the major road traffic stream. A lag is
Vehicles on the major road are either free movers
defined as the time remaining between an ap-
or a platoon member. Free-moving vehicles can,
proaching vehicle on the major road and a vehicle
under certain conditions, achieve their desired
entering the major road from a minor road.
speed, which is controlled by the speed limit. The
The important input parameters to the model
speed of all platoon members is set by the platoon
include (i) traffic volumes of all traffic streams, (ii>
leader.
percentage of heavy vehicle traffic to the total traffic
Major road vehicles are assumed to be unaf-
volume, (iii) type of intersection control (yield or
fected by minor road vehicles, so they can proceed
stop), (iv) speed limit on the major road, (v) percent-
and enter the intersection directly without any speed
age of each driver type in the driver population, (vi)
change. In some situations vehicles on the major
number of lanes for both major and minor roads,
road have to decelerate or stop as other major road
and (vii) total default simulation time. vehicles in the same lane attempt to turn into the
Several other input parameters such as move-
minor road.
up time, minimum allowable headway, turning speed
Minor road vehicles have to find a suitable gap
of vehicles, and maximum queue lengths are given
(lag) to join or cross the major road. If the intersec-
as constants to the model. However, it is possible
tion is Stop-controiIed, vehicles have to come to a
to change the values of these parameters between
complete stop before looking for a suitable gap (lag).
simulation runs.
In the case of a Yield-controlled intersection, minor
road vehicles may proceed and enter the intersection
Vehicle movemenf si~~lat~~~
direction if they can find a suitable gap (lag). The
A shifted negative exponential distribution is
minor road driver decision as to whether the gap
used to generate vehicle headway (t) from the aver-
(lag) is suitable or not depends on its size compared
age headway (i) calculated from the traffic volume.
The functional form of this distribution is with his or her critical gap.

P(h < t) = 1 - ,-Kt-c)i(i-c)l Factors affecting driver’s critical gap value


The type ofintersection control is important for
for t 2 c (an exogenous fixed time interval) and the drivers’ decision. At Stop-controlled intersec-
where 5 = l/h, and X = vehicle arrival rate. A value tions, drivers usually start from a stop condition,
5% T. SAVED et al.

Table 1. Mean and standard deviation of the gap acceptance number of tunes being crussed in the manoeu-
Yield control Stop control vre affects the critical value for drivers. Based on
the gap acceptance values provided by the Highway
Mean Std. dev. Mean Std. dev. Capacity Manual (National Research Council 1985),
Group (seconds) (seconds) (seconds) (seconds)
a correction factor of 0.25 second is added to the
Young males 4.0 0.75 5.0 0.7s critical gap value for each extra lane being crossed.
Old males 4.5 0.85 5.5 0.85 Heavy vehicles require critical gaps that are
Young females 5.5 1.00 6.5 1.00
Old females 6.0 1.25 7.0 1.25 about 30% larger than those for cars, because they
accelerate and decelerate at slower rates.
The value of the stopped delay significantly
modifies a driver’s gap acceptance behaviour. Driv-
while at Yield-controlled intersections some vehi- ers are usually more relaxed and less sensitive to
cles start from a low speed. It will obviously take gaps when they experience minimal delay, while
the drivers starting from a stop condition longer to they are more alert and more sensitive to gaps when
complete a manoeuvre. This leads to drivers to ac- they suffer greater delay. Wagner (1965) introduced
cept shorter gaps (lags) at Yield-controlled intersec- the term pressure of trafjc demand, which refers
tions than those accepted at intersections controlled to the pressure that the driver is exposed to after
by a stop sign. suffering delay. He found that pressure of traffic
Driver characteristics such as age and sex have demand had a very significant effect on driver behav-
been found to have a significant effect on gap-accep- iour. Ashworth and Bottom (1977) divided the gaps
tance behaviour. The data reviewed by Cooper presented to drivers into two categories: those gaps
f 1976) indicated that younger drivers generally ac- presented to drivers who had been waiting at the
cepted shorter gaps and are more consistent in their head of the queue for less than 8 seconds, and those
driving behaviour than older drivers. Cooper also presented to drivers who had been waiting at the
found that female drivers are often more cautious head of the queue for more than 8 seconds. He found
than male drivers in most traffic situations. Darzen- that the proportion of gaps accepted increased with
tas, McDowell, and Cooper (1980) analyzed unpub- increased waiting time. Another study by Adebisi
lished data by Transport and Road Research Labora- and Sama (1989), concluded that drivers showed
tory on crossing times of different drivers. The data significant changes in their gap acceptance behav-
indicate that young male drivers have the shortest iour when their stopped delay time exceeded the
crossing times. Both young male and female drivers range of 25-30 seconds, and they began to accept
have much smaller standard deviations than old driv- shorter than normal gaps.
ers of the same sex. Darzentas et al. (1980) also Based on information from those authors and on
found that female drivers were involved in fewer the function suggested by Tudge (1988) the following
conflict situations than male drivers. Wennell and stopped delay modification factor is used to alter
Cooper (1981) studied the gap acceptance behaviour the driver critical gap value:
of men and women drivers at four unsignalized inter-
sections. They found that the median accepted gap DL
for women drivers is longer than that for male drivers @‘=Q,,+DLtC
in all situations.
The model considers four types of drivers:
where
young males, young females, old males, and old
females. The model uses a truncated normal distri-
bution gap acceptance function, Based on the work on = stopped delay modification factor
by Darzentas et al. (1980) and Polus (1983). the gap DL = the delay value after which driver be-
acceptance values in Table 1 were chosen for the haviour begins to change (seconds).
mean and standard deviation for each group. The QD = the stopped delay value (seconds)
model allows these values to be changed, with the C= constant value (seconds)
higher mean gap-acceptance value designating the
more cautious driver group. The higher the standard Tudge (1988) used a value of 8 seconds for the
deviation of the gap acceptance value, the less con- delay value after which drivers begin to accept
sistent the driver group. The value of the mean and shorter than normal gaps based on information re-
standard deviation of the gap acceptance is consid- ported by Ashworth (1977). However, Ashworth
ered to be the same for both merging and single-lane considered only the time drivers wait as head of
crossing manoeuvres. queue and not the stopped delay. The values of DL
Simulation of traffic conflicts 597

and C were chosen as 27 and 0.5 seconds, respec- close time intervals. Location of the conflict is taken
tively. These values were selected after examining to be where the vehicles originated. The model uses
the data provided by Adebisi and Sama (1989). the time to collision (TTC) as a severity measure
for traffic conflicts; the smaller the TTC value the
The gap acceptance process more severe the conflict. The model first estimates
This process takes place when a vehicle has whether or not the vehicles are on a collision course.
to cross or merge with other traffic streams where If the vehicles are on a collision course, the TTC
different traffic streams have different priority levels value is calculated and compared with the threshold
according to the rules of the road. Generally, major value of 1.5 seconds. If the TTC is less than or
road vehicles have a higher priority than minor road equal to the threshold value, the model records the
vehicles, and straight-ahead major road vehicles conflict, its type, location, and the TTC value. Con-
have a higher priority than major road vehicles turn- flicts with TTC value less than 1 second are usually
ing left or right. Each vehicle is assigned a primary considered to be serious conflicts, with nonzero
critical gap value by testing the gap acceptance func- probability of signalling a crash.
tion according to the driver type and the intersection
type of control. The primary critical gap value is Model output
modified according to the vehicle type and the num- The output statistics of the model are included
ber of lanes to be crossed. Vehicles trying to cross in the file OUTPUT.LIS, which is produced with
or merge wait for a gap in the conflicting traffic each simulation run. These include: (i) the inputed
stream (streams) greater than or equal to their criti- and generated traffic volumes for each direction, (ii)
cal gap. The critical gap value is obtained by multi- the average vehicle delay for each queue along with
plying the primary critical gap with the delay modi- the average and maximum queue contents, and (iii)
fication factor 0,. The delay modification factor has the information obtained from importance sampling,
an initial value of 1.5 when the vehicle faces no which comprises the total number of conflicts, con-
delay, and this value decreases as the vehicle’s flict locations and type, and the TTC values.
stopped delay increases with a minimum theoretical
value of 0.5 when the vehicle faces infinite delay.
MODEL GRAPHICS
The model assumes that no driver will accept a gap
that he/she thinks will certainly lead to a collision. The visualization of the simulation at extreme
Therefore, a minimum acceptable gap (Gcmin) is events is an important aspect of this model. Its objec-
used, with a value of 2.0 seconds as a minimum tive is to show how the model is behaving during
allowable critical gap, based on data provided by traffic conflicts so characteristic behaviour, if any,
Wennell and Cooper (1981). If the critical gap value can be observed. This provides understanding of
is less than the minimum acceptable gap, it is set to critical situations and the parameters affecting their
the minimum. occurrence without the necessity of viewing many
Vehicle drivers who decide to enter the inter- average events. The general-purpose system anima-
section are assigned a single lane manoeuvre time. tion software PROOF (Wolverine Software Corpo-
This time is sampled from a truncated normal distri- ration, Annandale, VA, USA) is used to provide the
bution function. The mean and standard deviation graphical representation of the simulation events.
of the function depend on the driver type (Darzentas Two files are needed to run a PROOF anima-
et al. 1980). The sampled manoeuvre time is then tion. The first is the animation trace file which con-
corrected according to the number of lanes to be tains the sequence of timing information and other
crossed and the vehicle type. commands that make the animation happen. The
second is the layout file that has all the background
Conflict resolution text and graphics for an animation. The animation
A conflict occurs when a driver decides to exe- trace file is generally written directly by an executing
cute a manoeuvre that puts him/her at risk of colli- model or program designed to generate syntactically
sion with another vehicle. Conflicts in the simulation correct PROOF commands. The layout file is cre-
model are classified into three conflict types and two ated either by the graphical tool provided by PROOF
location groups: conflicts within the intersection that (as the case of the current model) or other programs,
result from conflicting vehicles entering the intersec- such as CAD programs.
tion at close interval times and that can mainly be Two animation files are produced during the
divided into crossing and merging conflicts, and simulation run. The first is the animation trace file
rear-end conflicts that result when conflicting vehi- (*.ATF) that has the animation commands for the
cles leave the intersection through the same lane at whole simulation time. The second is the presenta-
598 T. SAVED et al.

tion file (*.PSF), which contains the commands for intersection geometry and traffic control
the animation at the time at which traffic conflicts The first intersection, 156th Street and 20th Av-
occur. In the presentation file conflicts are viewed enue in Vancouver, has four one-lane approaches
one at a time. The animation lasts one minute for that intersect at approximately right angles. All four
each conflict, 30 seconds before the conflict and 30 approaches permit left-turn, through, and right-turn
seconds after the event. The user may go forward movements. The eastbound and westbound ap-
or go back through these conflicts using the “+I’ proaches on 20th Avenue are controlled by Stop
and “-I’ keys. The information shown with each signs. Crossing-sight distances and stopping-sight
conflict included driver, vehicle and conflict types, distances for all traffic movements are adequate.
the primary and driver-critical gaps, the stopped de- The peak hour traffic volume on 156th Street is about
lay value, and the time to collision value. This infor- f50 vehicles per hour (vph) and on 20th Avenue is
mation is used to develop typical profiles of conflict about 80 vph.
events. The second intersection, Holdom Avenue and
Broadway in Burnaby, has four approaches with
the two roadways intersecting at approximately 80
degrees. The northbound approach leg has two lanes
MODEL VALIDATION for left-turn, through, and right-turn movements,
Any computer simulation model has to show and the northbound exit leg, a single lane. The south-
that it (i) is logically correct and (ii) adequately repre- bound approach leg has a single lane for left-turn,
sents the modeled system in the validation process through, and right-turn movements, while the exit
of “assessing the extent to which a test or instrument leg is two lanes. The eastbound and westbound ap-
measures what it purports to measure” (Grayson proach traffic movements on Broadway are con-
and Hakkert 1987). Two levels were used for valida- trolled by Stop signs. In addition to the Stop signs
tion: face validity, or whether the model seems to on Broadway, southbound left-turn movements on
Holdom Avenue are restricted between 7:00 AM and
behave correctly; and external validity, which tested
9:00 AM, Monday through Friday. The peak hour
the “fit” between the model results and field obser-
traffic volume on Holdom Avenue is about 500 vph
vations of similar systems.
and on Broadway, is about 300 vph.
Face validity was tested by observing TSC-Sim
The third intersection, 116th Street and 75A
animation. Animation allows observation of the be-
Avenue in Delta, consists of four one-lane ap-
haviour of individual vehicles at an intersection to
proaches that intersect at approximately right
decide if the vehicle’s behaviour is logically reason-
angles. Each of the four approaches provides left-
able. Also, different model variables are displayed
turn, through, and right-turn movements. The east-
during the animation, and these show the internal
bound and westbound approaches on the 75A Ave-
interaction between different model entities. Re-
nue are controlled by Stop signs. Both crossing- and
searchers such as McCormick, De Fanti, and Brown stopping-sight distances are adequate. The peak-
(1987) indicated that the visualization process plays hour traffic volume on 116th Street is about 460 vph
a very important part in model validation. Face va- and on 20th Avenue is about 110 vph.
lidity provides trust in the model and provides some The fourth intersection, Chesterfield Avenue
assurance that the model is logically correct. and Keith Road in North Vancouver, consists of four
External validity was tested by comparing traf- one-lane approaches that intersect at approximately
fic conflicts observed at four unsignalized intersec- right angles. All movements are permitted on all
tions with traffic conflicts predicted by the simula- approaches. The eastbound and westbound ap-
tion model for these intersections for the same proaches on Keith Road are controlled by Stop
period of time. The validation data base came from signs. The crossing- and stopping-sight distances are
consultants’ studies of traffic confhcts at several in- adequate. The peak hour traffic volume on Chester-
tersections in the Greater Vancouver, Canada, area field Avenue is about 360 vph and on Keith Road is
(Hamilton Associates, 1991). Although several more about 140 vph.
intersection studies were reviewed, the majority of
the unsignalized intersections covered by the con- Conflict observation method
flict studies were complex layouts, beyond the sim- Traffic conflicts were recorded at the study in-
ple T and four-way intersections selected for this tersections by trained observers. The observation
analysis. All selected intersections were 4-leg inter- and recording method is an on-site, online record of
sections with negligible grades, good visibility, and the incidence and severity of traffic conflicts. Obser-
a simple layout. vations were made for two days at each intersection.
Simulation of traffic conflicts 599

Table 2. Time-to-collision and risk-of-collision scores set at 50 km/h, which is the legal posted speed limit
TTC and ROC Time to collision Risk of collision at the four intersections. The comparisons of ob-
scores (TTC) (ROC) served and estimated conflicts for both intersections
are given in Tables 3 through 6.
I 1.6-2.0 seconds Low risk
2 I .O- I 5 seconds Moderate risk In the case of the 156th Street and 20th Avenue
3 0.0-0.9 seconds High risk intersection (Table 3), the model predicted five con-
flicts compared to eight observed conflicts with a
Source: Brown, G. R. (1991).
similar distribution of the types of conflicts over the
different locations within the intersection. It was
observed in the field that typical vehicles’ speeds
Two observers were used each day for an eight-hour exceeded the legal 50 km/hr limit along 156th Street,
observation period, giving a total of 32 man-hours and this was suspected to be a factor contributing
at each intersection. The severity of traffic conflicts to the difference between number of conflicts and
is determined by the sum of two scores: the TTC those predicted by the model. Increasing the speed
score and the risk-of-collision or ROC score. The limit in the model to 70 km/h caused the model to
ROC score is a subjective measure of the risk of increase the total number of conflicts to 10 conflicts
collision and is dependent on the perceived control with a distribution of predicted conflicts very close
that the driver has over the conflict situation. The to that of observed conflicts.
TTC and ROC scales were given equal weighting The Holdom Avenue and Broadway intersec-
and combined into a 5-point Likert-type scale. The tion model predicted 17 conflicts compared to 19
summation of the TTC and ROC scores gives the observed. Increasing the speed by 10 km/h to
overall severity score, which ranges between 2 and 60 km/hr produced 21 conflicts. In both cases the
6. An overall severity score of 2 signifies a low- predicted conflicts distribution was close to the ob-
risk conflict situation and a score of 6 is a high- served conflicts distribution (Table 4).
risk conflict situation (Table 2). The midpoint of the In the 116th Street and 75A Avenue intersec-
composite scale registers the critical event, corre- tion, six conflicts were predicted compared to 10
sponding to a TTC of 1.5 seconds or less with a observed with the speed limit set at 50 km/h. Raising
“moderate” ROC. the speed limit to 60 km/h resulted in eight predicted
Reliability tests of the observation method gave conflicts with approximately identical conflicts dis-
77% accuracy with 95% confidence, with a high of tribution except for the left-turn crossing conflicts
85% accuracy for assessing the correct TTC. In addi- (Table 5).
tion, in a study of 13 intersections to test the validity The Chesterfield Avenue and Keith Road inter-
of a TTC = 1.5 seconds or less for a measure of section conflict study team observed 15 conflicts.
safety as defined by the number of accidents, it was The study also indicated that the major road vehicle
found that at 8 of I1 intersections conflicts are sig- speed largely exceed the 50 km/h speed limit. The
nificantly correlated with accidents at 95% confi- model predicted nine conflicts out of the 15 observed
dence with R2 Z- .64 with three intersections having at 50 km/h speed limit and 12 conflicts at 70 km/h
R2 z .81 (Brown 1994). speed limit. Although there is some difference in

Comparing results
A comparison between the observed conflicts Table 3. Observed and predicted conflicts distribution, 156th
and the simulated conflicts was carried out. The Street and 20th Avenue intersection
model was modified to allow for the traffic volume Predicted Predicted
changes through the morning, noon, and afternoon Conflict name and Observed conflicts conflicts
periods. There was further modification to allow the description conflicts 50 kmihr 70 km/hr
inclusion of the restricted southbound, left-turn 156th Street northbound, 4 2 4
movement between 7:00 AM and 9:00 AM hours at the rear-end conflicts
Holdom Avenue and Broadway intersection. The 156th Street southbound, 1 1 1
crossing conflicts
simulated conflicts were averaged over a large num- 156th Street northbound, 2 1 2
ber of runs. left-turn crossing con-
Since the simulation model considers only con- flicts
156th Street southbound, 1 0 I
flicts with TTC values less than or equal to 1.5 sec- rear-end conflicts
onds, observed conflicts with TTC values greater 156th Street northbound, 0 I 2
than 1.5 seconds were excluded from the validation crossing conflicts
Total 8 5 10
test data. At first the computational speed limit was
600 T. SAVED et al.

Table ‘4. Observed and predicted conflicts distribution, Holdom Avenue and Broadway
intersection

Predicted Predicted
Observed conflicts- conflicts-
Conflict name and description conflicts 50 km/hr 60 km/hr

Left turn crossing conflicts involving 10 I 9


westbound left turning motorists and
southbound through motorists.
Left turn opposing conflicts involving 1 2 3
eastbound through motorists and
westbound left turning motorists.
Crossing conflicts involving eastbound 2 3 3
through motorists and northbound or
southbound through motorists.
Crossing conflicts involving eastbound
through motorists and northbound or
southbound through motorists.
Right turn (merging) conflicts involving
westbound right turning motorists
and northbound through motorists.
Rear-end conflicts involving south-
bound through motorists.
Left turn opposing conflicts involving 2 1
southbound through motorists and
northbound left turning motorists.
Total 19 17 21

Table 5. Observed and predicted conflicts distribution, 116th Street and 75A Avenue intersection

Predicted Predicted
Observed conflicts conflicts
Conflict name and description conflicts 50 kmlhr 60 km/hr

Left turn crossing conflicts involving 3 2 2


through northbound vehicles on
116th Street and left turning vehicles
on 7SA Avenue.
Left turn crossing conflicts involving 2 1 1
through southbound vehicles on
116th Street and left turning vehicles
on 75A Avenue.
Rear-end conflicts involving through 2 1 2
northbound vehicles on 116th Street.
Right turn conflicts involving west-
bound right turning vehicles from
7SA Avenue and through northbound
traffic on 116th Street.
Rear-end conflicts involving through
southbound vehicles on 116th Street
Left turn opposing conflicts involving
through southbound vehicles and
northbound left turning vehicles on
116th Street.
Total IO 6 8
Simulation of traffic conflicts 601

Table 6. Observed and predicted conflicts distribution, Chesterfield Avenue and Keith Road
intersection

Predicted Predicted
Observed conflicts conflicts
Conflict name and description conflicts 50 km/hr 70 km/hr

Crossing conflicts involving northbound 8 4 6


through vehicles on Chesterfield Ave-
nue and eastbound through vehicles
on Keith Road.
Crossing conflicts involving southbound 4 3 3
through vehicles on Chesterfield Ave-
nue and westbound through vehicles
on Keith Road.
Crossing conflicts involving southbound 1 1 1
through vehicles on Chesterfield Ave-
nue and eastbound through vehicles
on Keith Road.
Rear-end confiicts involving south- 1 0 1
bound through vehicles on Chester-
field Avenue.
Left turn crossing conflicts involving 1 1 1
southbound vehicles on Chesterfield
Avenue and left turning vehicles on
Keith Road.
Total 15 9 12

the total number of conflicts between predicted and Ferguson (1976) state that the number of conflicts
observed conflicts, the distribution of predicted con- occurring at a certain location is proportional to the
flicts is very close to the observed conflicts espe- product of conflicting volumes. The same result was
cially for the 70 km/h speed limit (Table 6). recorded by Hodge and Richardson (1978). Spicer,
A comparison, while limited to a small number Wheeler, and Older (1979) indicated that at relatively
of intersections, shows that the model results com- low traffic volumes, the total number of observed
pare very well with the observed field data, espe- conflicts is proportional to the square root of the
cially the internal distributions of predicted and ob- product of the conflicting volumes. Darzentas et al.
served conflicts. The comparison implies a (1980) propose that conflicts linearly increase as a
reasonably successful external validation process, function of the traffic volume. None of the above
which leads to the conclusion that the model can researchers considered intersections with high traf-
be used in similar situations with some degree of fic volumes.
confidence. Further validation is necessary, if the Figures 1 and 2 are examples of the relationship
model is applied to other types of intersections, but between traffic volume and the simulated conflict
for this model structure and conflicts definition, the rate obtained from the model. These figures indicate
results are encouraging. that over a wide range of traffic volumes, including
congested conditions, an exponential function
seems to give a good fit. However, if only low traffic
DISCUSSION
volumes are considered (volumes less than the war-
Since traffic conflicts offer a potentially rich rants for a traffic signal), conflicts may appear to be
source of behavioural information, the possibility of proportional to the square root of the conflicting
successful traffic-conflicts simulation opens up areas volumes as suggested by Spicer et al. (1979). The
of practical research into the manner in which differ- large increase in simulated conflicts at high traffic
ent road users interact with various traffic situations. volumes is probably the result of allowing drivers
Some research implications, using TSC-Sim, are in the model to accept shorter gaps as their delay
given below. time increases. However, usually unsignalized inter-
sections operate at lower volumes, reducing this
Volume and conflicts effect.
Several researchers suggest that traffic volume The curves representing conflicts for Yield- and
flow has a significant effect on conflicts. Cooper and Stop-controlled intersections (Figures 1 and 2) were
T. SAVEDet al.

Fig. f. Relationship between traffic volume and conflicts for 8 T-intersection.

very close at low traffic volumes and the difference Speed and conjicls
between them increases rapidly as traffic volume Early research by Cooper, Starr, and Wennell
increases. This outcome tends to confirm the engi- (1977) indicates that risk of an accident increases as
neering practice of setting a volume warrant that the approach speed of major road vehicles increases,
limits the volume after which Yield signs should not which would suggest that traffic conflicts also in-
be used to control unsignalized intersections. crease under the same speed conditions. Darzentas

100 150 200 250 300 350 400


Volume (VEHIHR)
~ I__-
ql - kq k- 0.3
Approroh rpood 40 km/hr
1 tan@for bathm@or and minor road*

Fig. 2. Relationship between traffic vofume and conflicts for a it-leg intersection.
Simulation of traffic conflicts 603

120

100

.Q_ *O
1
. P *O
/
B 40
_e______---’
___- __o____------

____.---
_o______--__o________---;----
20

............. ...... ........ ............ .... .....................


................. D.. ..................

n- ..... .......... I

=40 45 50 55 60 65 70
Speed (Km/hr)
q1 - kq k-O.3
2 Ianw approach for both major and minor rods
nti wntr~ikd
I-

Fig. 3. Relation between approaching speed and conflicts at a Stop-controlled 4-leg intersection.

et al. (1980) suggests a rapid increase of traffic con- model is shown in Figures 3 and 4. The figures indi-
flicts rate (at a fixed flow) as the mean speed on the cate an increase in the number of conflicts as the
major road increases. mean approach speed increases for a fixed volume,
The relationship between the mean approach with the increase proportional to the traffic volume.
speed and number of conflicts from the current A slight increase was obtained at low traffic volume

60

50
g
1 40

*[ 30
3
20
_______?_.--------
________r------- ---ii---
10 ____----__Q_____-----*--
,..,......
P ..*_ .. .. . . . . . . . . . . .

n I I J
“40 45 50 55 60 65 70
SDeed /Km/hr)
ql = kq k=O.3 No turningvohklr
1 lane approach for both major and minor road8
Stop controlled intemaction

Fig. 4. Relation between approach speed and conflicts at a Stop-controlled 4-leg intersection
604 T. SAVEDet al.

1.5 r

o.._ --_.
--*.._
_. qG=fqfF
G 1. 45 ---‘__a
A..
Q..,
‘..,,, ‘Q.
“‘...*,,, -.
““.-- . . . ..__.* *.
‘+______--O..__
I --__
-0
; 1. 4
P

E 1. 35
H
%
1. 3

ql =kq k-0.3
2 lane8 approachfor both major and minor roads
Stan onntroiiad kltnr9nctian

Fig. 5. Relation betw-een approaching speed and severe conflicts for a Yield-controlled T-intersection.

and a significant increase at high traffic volumes. the tendency of drivers to accept shorter gaps at
There was no significant effect of the intersection Yield-controlled intersections. This increased con-
type of traffic control (Yield, Stop) on the value of flict severity may put a constraint on using Yield
this increase. signs to control intersections with relatively high
The number ofconflicts estimated by the model major road speeds, even though other warrants for
increased as the dispersion of speeds on the main using a Yield sign such as traffic volume and sight
road increased. This agrees with the published re- distance are satisfied.
search that indicates a driver’s error of judgement The number of severe conflicts approximately
associated with vehicles having speeds different increased exponentially with speed (Fig. 7) com-
from the mean speed (Brian 1962; Cooper et al. pared with a linear increase in the case of all con-
1976). These results suggest that a decrease in major flicts. This signifies that there is an effect of speed
road vehicles speeds and the dispersion of these on severe conflicts and agrees with research into
speeds can result in adecreasing number of conflicts. accidents in which a nonlinear relationship between
A decrease may be achieved by police activity at the speed and severe accidents is found.
intersection as reported by Cooper and McDowell
(1977). Conflicts and driver type
The values for the mean and standard deviation
of the gap acceptance function and manoeuvre time
The overall severity of traffic conflicts repre- for different driver types were chosen based on the
sented by the average time-to-collision (TTC) value data provided by Cooper (1976), Darzentas et al.
increased as the mean approach speed of the major (1980), and Polus (1983). The model results agreed
road vehicles increased in the model (see Figures 5 with their findings, which indicated that female driv-
and 6). The same results were reported by Cooper ers are generally less involved in conflict situations
et al. (1976, 1977) and Darzentas et al. (1980). This than male drivers and that older drivers are less
indicates that the percentage of injury accidents at consistent in their driving behaviour. This contri-
unsignalized intersections may increase as the major butes to the face validity of the model and indicates
road mean speed increases. that these variables give a good description of driver
The overall severity of conflicts was higher in behaviour, at least as replicated by the model. The
the case of Yield-controlled intersections than Stop- model can perhaps be useful in investigating the
controlled intersections. This may result because of effect of changing these variable values. For exam-
Simulation of traffic con!?icts 605

1. 25

Speed WM/HRl
ql -4 k=O.S
2 kwia~appwch for both major and minor roadr
yield controll8d Intersection I

Fig. 6. Effect of approaching speed on conflicts severity for a Stop-controlled T-intersection

ple, the number of conflicts predicted by the model In practice, observing traffic conflicts is expen-
can be reduced by increasing the median acceptance sive, since it requires trained people. An alternative
gap (i.e. make drivers more cautious). This would to direct observation is to utilize the results provided
follow research by Cooper and McDowell (1977) by a simulation model, such as a first estimate of
which indicated that police activity near an intersec- driver behaviour, to better understand the traffic
tion makes drivers turning from a minor road more process in incipient crash conditions. This model
cautious and consequently decreases the number of offers a way of estimating traffic conflicts and can
conflicts. also provide an in-depth study of conflicts as critical
traffic events, The resuhs of the simulation may
eventually reduce the cost for direct observation of
CONCLUSIONS AND FURTHER
all except the most complex intersections.
RESEARCH
The model is intended to be used to meet two
This paper describes a microscopic traffic con- traffic analysis objectives. The first is to assess the
flicts simulation model, TSC-Sim, for both T- and safety and performance of unsignalized intersec-
4-leg intersections using the commercially available tions. The safety measure may be expressed by the
GPSS/H. The model considers different aspects of number and the location of conflicts predicted by
driver behaviour and several other parameters on a the model, while the performance can be expressed
driver’s entry or crossing decision at unsignalized by the intersection level of service as a function of
intersections. The model employs importance sam- the average delay to minor road vehicles. The sec-
pling to save characteristic information for only ond use of the model is as a tool for investigating how
those conflicts that are predefmed to be “signifi- to reduce driver risk. Several researchers emphasize
cant” events. Graphics is a very important part of the value of behavioural studies and driver education
the model as it shows how the model behaves at in reducing road user risk and the model may be a
extreme values: that is, during traffic conflicts, and useful tool in identifying the parameters for these
gives a clearer understanding of these critical situa- kind of studies.
tions and the parameters affecting their occurrence. The research attempts to contribute to road
The validation process showed that the model re- safety studies in two ways. The measure used in the
sults correlate reasonably well with the field ob- research to delineate the conflict has been quantified
servations from the area of Vancouver, British as a time proximity value, allowing road safety to
Columbia. be studied in the context ofgap acceptance research;
606 T. SAVED eta1

40
. Confllot looatlonr
e

q-yTf=F

n I,
“40 45 50 55 60 65 70
Speed (Km/hr)
ql - kq k-0.3
2 bnes approachfor both major and minorroada
Weld controlledinterwction

Fig. 7. Effect of approaching speed on conflicts severity for a Yield-controlled intersection.

and the resultant stimulation model using this mea- Acknowledgements-The financial support for this study comes
from F. Navin’s Natural Science and Engineering Research
sure appears to replicate, with an acceptable degree Council of Canada Operating Grant. P. deLeur provided details
of accuracy, field observations of traffic conflict be- of the study reports used in the research. Hamilton Associates
haviour using time proximity to collision. However, kindly supplied their study reports for selection and analysis.
the study has pointed to several areas of further
research. First, the severity definition used is spe- REFERENCES
cific to the field procedure, and this constrains the Adebisi, 0.; Sama, G. N. Influence of stopped delay on
wider evaluation of the model, particularly the rela- driver gap acceptance behaviour. Journal of Transpor-
tionship among conflicts, flow volume, and speed. tation Engineering 115:305-3 15; 1989.
A larger inventory of conflict studies, using the time Allen, B. L.; Shin, B. T.; Cooper, P. J. Analysis of traffic
proximity definition, is needed to further investigate conflicts and collisions. Hamilton, Ontario, Canada:
Department of Civil Engineering, McMaster Univer-
these relationships. Secondly, more research is sity; 1977.
needed to confirm the conceptual linkage of traffic Amundsen, F.; Hyden, C. Proceedings of first workshop
conflicts to safety and risk. The model simulates on traffic conflicts, Oslo, Institute of Transport Eco-
conflicts (as defined here) and cannot be evaluated nomics, 1977.
against accident statistics nor other measures of road Ashworth, R.; Bottom, C. G. Some observations ofdriver
gap acceptance behaviour at a priority intersection.
user risk. The authors are reasonably confident that Traffic Engineering and Control. 18569-571; 1977.
the time proximity definition of traffic conflict will Brian, R. Vehicle speed estimation at right-angled at-grade
prove to be a measure of driver risk, but to date, junction. Surveyor and Municipal and County Engineer
this contention is still tentative. 21:789-791; 1962.
Lastly, simulation has been suggested as a use- Brown, G. R. Use of traffic conflicts for near-miss re-
porting. In: Near-miss registration as a safety tool.
ful tool for the study of road safety, but there is no Oxford, Butterworth-Heinemann; U.K.: 11 l-125,
attempt to compare simulation with analytical or 1991.
statistical methods. Until the road safety problem Brown, G. R. Traffic conflicts for road user safety studies.
can be described unambiguously, and the complex Canadian Journal of Civil Engineering 21(1):1-15;
relationship between driver behaviour and roadway 1994.
Cooper, D. F. Observation of gap acceptance using televi-
and traffic parameters delineating driver risk is sion. Paper presented at the second European Confer-
known, perhaps computer simulation will prove ence on Operations Research EURO II, Stockholm,
useful. Sweden, November 29-December 1, 1976.
Simulation of traffic confiicts 607

Cooper, D. F.; Smith, W.; Broadie, V. The effect of the National Research Council. Highway capacity manual.
approach speed on merging gap acceptance. Traffic Special Report 209. Washington, DC: Transportation
Engineering and Control 17:256-257; 1976. Research Board; 1985: Chapter 10.
Cooper, D. F.; Ferguson, N. A conflict simulation model. Navin, F. P. D. Computational techniques in transporta-
Traffic Engineering and Control 17:306-309; 1976. tion engineering. Proceedings, 4th International Con-
Cooper, D. F.; Storr, P. A.; Wennell, J. The effect of ference on Computing in Civil and Building Engi-
speed on gap acceptance and conflict rate. Traffic Engi- neering, Tokyo, Japan, June 1991.
neering and Control 18: 110-l 12; 1977. Perkins, S. R.; Harris, J. L. Criteria for traffic conffict
Cooper, D. F. and McDowell, M. R. C. Police effects on characteristics. Report GMR 632. Warren, MI: General
accidents risk at T-junctions. Traffic Engineering and Motors Corporation; 1967.
Control 18:484-487~ 1977. Polus, A. Gap acceptance characteristics at unsignalized
Darzentas, J.; McDoweIl, M. R. C.; Cooper, D. F. Mini- urban intersections. Traffic Engineering and Controi
mum acceptance gaps and conffict involvement in a 24(5):255-2.58: 1983.
single crossing manoeuvre. Traffic Engineering and Schriber, T. J. Simulation using GPSS. New York: John
Control 21:58-61; 1980. Wiley & Sons; 1974.
Grayson, G. B.; Hakkert, A. S. Accident analysis and Spicer, B. R.; Wheeler, A. H.; Older, S. J. Variations
conflict behaviour. In: Rothengatter, T.; deBruin, R. in vehicle conflicts at a T-junction and comparison
(editors). Road users and traffic safety. Groningen, The with recorded collisions. Second International
Netherlands: Van Gorcum; 1987. Workshop on Traffic Conflict Techniques, Paris,
Hamilton Associates. Traffic operations review. Studies 1979.
of several intersections in the Vancouver, Canada, Tudge, R. T. INSECT-the calibration and validation of
area, 1991. an intersection simulation model. Proceedings
Hauer, E. Traffic conflict surveys: Some study design of internationa1 workshop on intersections without
considerations. TRRL Supplementary Report 352. traffic signals, Bochum, West Germany, 16-18
Crowthorne, Berkshire, U.K. Transport and Road Re- March 1988. Berlin: Sponger-Verlag; 1988:214-
search Laboratory; 1978. 235.
Hayward, J. C. H. Near miss determination through use van der Worst, A. R. A. A time-based analysis of road
ofa scale ofdanger. Report No. TTSC 7115. University user behaviour in normal and critical encounters.
Park, The Pennsylvania State University; 1972. Proefschrift, Technische Universiteit Delft, 1990.
Hodge, G. A.; Richardson, B. E. A study of intersection van der Horst, A. R. A.; Brown, G. R. Time to collision
accident exposure. Australian Road Kesearch Journal and driver decision making in braking. Report IZF,
9(5):7-16; 1978. 1989 C-23. Soesterberg, The Netherlands: TN0 Insti-
McCormick, B.; Defanti, A.; Brown, D. Visualization in tute for Perception; 1989.
scientific computing. Computer Graphics 21(6):6-8; Wagner, F. A. An evaluation of fundamental driver deci-
1987. sions and reactions at an intersection. HRB, Highway
McDowell, M. R. C.; Wennell, J.; Storr, P. A.; Darzentas, Research Record 11868-84; 1965.
J. Gap acceptance and traffic conflict simulation as a Wennell, J.; Cooper, D. F. Vehicles and driver effect
measure of risk. TRRL Report No. SR 776. Crow- on junction gap acceptance. Traffic Engineering and
thorne, U.K. Transport and Road Research Labora- Control 628-635 ; I98 1.
tory; 1983.

Potrebbero piacerti anche