Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Special Issue on work of the Cooperative Research Centre for Rail Innovation, Australia

Integration of driving simulator and


traffic simulation to analyse behaviour
at railway crossings

Proc IMechE Part F:


J Rail and Rapid Transit
227(5) 427438
! IMechE 2013
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0954409713489117
pif.sagepub.com

Inhi Kim1, Luis Ferreira1, Li Sian Tey1 and Guy Wallis2

Abstract
The use of state-of-the-art technology to collect and analyse data has significantly improved the effectiveness of safety
studies. Currently, despite the fact that there are many safety systems deployed at railway crossings, only limited research
has been conducted to evaluate which of these systems is the most effective in terms of costs and safety. This paper
demonstrates a way to evaluate safety at railway crossings using a twin-pronged approach: a driving simulator and traffic
simulation software. A number of outputs have been observed from a driving simulator, such as driver compliance rate,
vehicle speed profile, acceleration profile, initial braking position and final braking position. The compliance percentage at
passive crossings (67 and 72% for a stop sign and rumble strips, respectively) has lower compliance rates compared with
active crossings (97 and 93% for flashing red light and in-vehicle audible warning, respectively) at an 80 km/h approach
speed. Using a statistical analysis it is shown that speed and acceleration profiles can be used to differentiate the
effectiveness of active and passive crossings. These indicators are interpreted and used as input to a traffic simulation,
which assists in determining which safety device is more efficient. By integrating driving simulator and traffic simulation
models, this approach can be applied to evaluate and compare safety performance without the need to install costly test
beds at real railway crossings.
Keywords
Driving simulator, traffic simulation, railway crossing, safety evaluation
Date received: 31 October 2012; accepted: 8 April 2013

Introduction
Crashes at railway crossings have been analysed in
detail in many studies in order to improve the level
of safety. One of the biggest practical improvements
has been the installation of safety devices. These range
from passive devices to active protection systems.
In some countries, research into in-vehicle-installed
warning systems has been conducted and in some
cases, implemented, to provide a safer environment
near to railway crossings. Passive devices include
stop signs, pavement markings and reective paint
on the sides of railway rolling stock. Active protection
systems include warning sounds, ashing lights and
boom gates. They provide a much higher level of
safety at most railway crossings. This is because
they provide a signicant level of warning to road
users about when a train is approaching a crossing.
However, despite the high levels of safety created at
signalised crossings, collisions still occur.1
Currently, despite the fact that there are many
safety systems implemented at railway crossings,

only limited research has been conducted to evaluate


which system is the safest and most cost-eective.
Traditionally, most trac accident studies have
relied heavily on reported accident statistics2,3 to analyse black spots where trac accidents frequently
occur, compare the eect of installing safety features
and the eect of education programmes.4
To date, there has been no methodology that uses
trac simulation software to identify the safety level
of a specic system. Most accident models for railway
crossings use historical records as input. In addition,
driving simulators have not been widely applied in the
analysis of railway crossing accidents.
1

University of Queensland, Faculty of Engineering, Architecture and


Information Technology, Australia
2
University of Queensland, School of Human Movement Studies and
Queensland Brain Institute, Australia
Corresponding author:
Inhi Kim, Faculty of Engineering, Architecture and Information
Technology, The University of Queensland, QLD 4072, Australia.
Email: inhi.kim@uq.edu.au

428
The ultimate aim of the research reported in this
paper was to develop an innovative crash prediction
model using current transportation simulation
approaches that could be used to estimate the safety
of railway crossings as a function of the trac
volume, the train headway and the type of safety
devices. The proposed technique will allow transportation engineers to suggest the best approach to solve
a safety problem before actual implementation of a
state-of-the-art system. This paper proposes a trac
safety model based on commercially available trac
simulation software which is integrated into a driving
simulator and allows the evaluation of safety
interventions.
This paper is structured as follows. The next section critically reviews what kind of parameters inuence safety at railway crossings and how a driving
simulator and trac simulation can be used to
enhance safety levels. This is followed by the section:
Model development which provides a brief description of the adopted procedure. An analysis of the
main results obtained from an experiment conducted
using a driving simulator is given the section:
Results. The application of the proposed approach
to simulation studies is described in the section
Application to trac simulation. Finally, conclusions are drawn from the discussed analysis of results.

Past work
Safety-related parameters
There have been several attempts in the literature to
identify what factors contribute to the risk of accidents in the vicinity of a level crossing. One of the
main factors is the behaviour of drivers and understanding this behaviour is a key to reducing the
chances of an accident.
The compliance rate is regarded as a parameter
that directly corresponds to dierent types of crossings.5 A relationship can be derived that measures
whether vehicles stop, and thus ensure a collision
does not occur, at a crossing when it is equipped
with dierent types of safety devices. The compliance
rate is also related to both driving speed and to the
distance between a driver and the stop line. This phenomenon seems to be close to what happens at signalised intersections on highways which is commonly
referred to as the dilemma zone problem. As dierent warnings at the crossings are likely to inuence
driving speeds in dierent ways, the relationship
between compliance rate and vehicle speed also
needs to be considered.
The relationship between the likelihood of a collision and speed has been widely studied in the literature. At higher speeds, the time to react to changes in
the environment is shorter, the stopping distance is
larger and manoeuvrability is reduced.6 After analysing accident data compiled between 2000 and 2009,

Proc IMechE Part F: J Rail and Rapid Transit 227(5)


Richards and Cuerden7 concluded that higher impact
speeds led to more severe injuries and stated that the
50th percentile speed for seriously injured drivers is 24
miles/h and for fatally injured drivers it is 34 miles/h.
Elvik8 reviewed several studies and found that the
major factors involved in determining the likelihood
of an accident are: a willingness to stick to the speed
limit; weather conditions (fog, rain, darkness), familiarity with the road and visibility. The data on driving
behaviour collected by Mesken et al.9 and Whissell
and Bigelow10 was analysed in terms of vehicle
speed, age and gender. It is clear from the analysis
of laboratory data and accident records that individuals react dierently in terms of speed.
It has been found that a surprising number of accidents occur as a result of the wrong pedal being
depressed. From the total of 43,000 accidents reported
for the period 19791995, 3740 of them were caused
by pedal misapplication.11 That study showed that
pedal errors occurred not only at the start of an
engine but also during driving. Acceleration force
has also been measured on-board city buses in regular
trac in order to nd out its relationship with trac
accident frequency.12,13 In those studies, the author
pointed out that the combination of acceleration
and deceleration played an important role in determining the likelihood of an accident. The limited
power of an engine meant that less variation in the
acceleration behaviour was expected.13 This point
highlights that dierent types of vehicles need to be
considered when safety measurements are performed.
Braking prole is another important factor that
inuences the likelihood of a collision.5,14 Analysing
the time of initial and nal braking can provide a clear
description of how the warning devices inuence a
drivers braking response. While driver distance and
driver time are the most appropriate factors to explain
the risk of a collision15, braking can be directly linked
to the time to collision (TTC). The TTC has come to
play an important role in analysing trac safety.16-18
The TTC value decreases with time as the vehicles
approach the point of impact. A collision is expected
when the TTC equals zero.

Use of driving simulator and traffic simulations


It is important to use state-of-the-art technology to
identify both the causes and solutions to safety
issues at crossings. The use of trac simulations
allows safety evaluations to be performed without
interrupting existing trac ows or placing drivers
at risk. Simulation-based safety analyses identify not
only the probability of collisions but also the severity
of these potential collisions. Using trac simulations,
many dierent scenarios can be tested at low cost.
Some past studies have used trac simulations to
evaluate road system evaluations.7, 8 However, the
use of trac simulations to analyse railway safety
remains largely unexplored. The reason for this

Kim et al.
oversight is that the collection of data about crashes
at or near railway crossings is very dicult due to its
infrequent occurrence, compared with the situation of
road accidents. The use of trac simulation to evaluate safety faces problems because most simulation
models assume that all vehicles move in the network
in a safe manner. This means that all vehicles follow
trac regulations and the driver is not driving in an
aggressive manner. Additionally, simulation itself is
not able to directly consider new systems such as
ITS safety devices, GPS-equipped systems or driving
behaviour, such as the eects of fatigue, drugs and
alcohol. Therefore, in the current paper, data from a
driving simulator is combined with trac microsimulations.
Driving simulators play a pivotal role in conducting trac-related studies, and can be a good alternative to eld-based data collection.9,10 First, in most
cases, eld studies are very expensive to perform
and evaluate. In order to achieve reliable outputs,
environment and trac conditions should be under
control. For example, to determine whether or not
the creation of a rapid transit lane exclusively for
the use of buses is ecient, would require it to be
constructed on an existing road and the existing trac
would have to be controlled which would be costly.
Second, a driving simulator does not encounter physically dangerous situations. For example, studies of
the eects of alcohol, drugs and fatigue, and weather
conditions such as rain, fog and snow are dangerous
to conduct on the road. Third, it is much easier to
collect data. A driving simulator is equipped with several computers, some of which can be used to collect
the required data. Driving simulators can be used to
test potential future scenarios. Thus, emerging safety
technologies can be tested before being implemented.
By obtaining reliable data about driving behaviour
from a driving simulator, trac simulations are able
to replicate driving behaviour and improve the evaluation process.19
Many trac research organizations and universities have adopted driving simulators including the
CARRS-Q advanced driving simulator at the Centre
for Accident Research and Road Safety of the
Queensland University of Technology.

Model development
Driving simulator setup
Twenty-four volunteers ranging in age from 17 to
66 years were recruited from the local community
and the University of Queensland. All the participants
had normal or corrected-to-normal vision and a current drivers licence. Participants were selected on the
basis of having had little or no experience with
action computer games, i.e. driving games, ight
simulators, etc. Ethics approval was obtained from
the Behavioural and Social Sciences Ethical Review

429
Committee at the University of Queensland.
Participants volunteered to take part in the study
and informed consent was gained. The participants
were not informed of the objectives of the research
on testing driver behaviour. The main reason for
this was to avoid the subjects feeling they were
being assessed which might have resulted in them
not exhibiting their normal driving behaviour towards
the varying types of warning devices most especially,
articially high levels of vigilance or compliance. One
trial consisted of four level crossings with dierent
warning devices, namely: stop sign; rumble strips;
ashing red lights; and in-vehicle warnings. One set
of 12 trials was performed by each participant. While
driving, the participant encountered dierent speed
limits: either 60 km/h or 80 km/h randomly.
Therefore, 48 (4  12) data sets were created by each
successive driver.

Development of simulated driving task environment


A simulated virtual environment was developed using
a driver simulator at the University of Queensland.
Once a start button was pressed and the accelerator
pedal activated, the participant was required to drive
in the left-hand lane of a simulated two-lane two-way
road at a pre-set maximum speed by manipulating the
steering wheel. Participants had the visual impression
of driving along a curved road. A digital speedometer
at the central bottom of the display screen showed the
vehicles speed. Figure 1 provides a schematic illustration of the simulated roadway. The distances specied
in Figure 1, pertaining to the level crossing layout and
warning sign placement are based on design specications contained in the Manual of Uniform Trac
Control Devices, Part 7: Railway Crossings.20 The
driver approached a level crossing after approximately 1 km of driving. All level crossings that the
drivers encountered during the scenarios had the
same road characteristics; however, four dierent
types of warning devices appeared randomly at the
crossing as shown in Table 1. Two of the conventional
warning devices (stop sign and ashing red lights)
were included as control samples for another two
innovative warning devices (rumble strips and in-vehicle audible warning). The stop sign and rumble strips
are passive devices while ashing red lights and invehicle audible warnings were activated by a train
that was a minimum of 20 s away from arrival at a
single track crossing this is based on the Manual of
Uniform Trac Control Devices, Part 7: Railway
Crossings.20 Passive crossings provide only stationary
signs that do not convey information about a train.
Drivers have to look for the presence of a train before
proceeding onto the crossing. Rumble strips are transverse strips raised above the pavement that give an
audible and tactile sensation to the motorists passing
over them. In the simulation this was imitated
by vibrating the force-feedback steering wheel.

430

Proc IMechE Part F: J Rail and Rapid Transit 227(5)

Figure 1. Schematic illustration of a typical simulated roadway. (a) Stop Sign, (b) Rumble Strips, (c) Flashing Light, and (d) In Vehicle.

Kim et al.

431

Table 1. Road and crossing characteristics.


Characteristic

Simulated environment

Road type

Two-lane two-way sealed


pavement road in a
rural setting
One
Straight road approaching
level crossing
Level
1.0 km
3.5 m
No other vehicles on the road.
Constant at either 60 km/h
or 80 km/h
50 km/h
Three cars x 69.5 m/car
90

Number of train tracks


Horizontal alignment
of road
Vertical alignment of road
Road length
Lane width
Vehicles
Road speed limits
Train speed
Train length
Rail/road angle

Its design broadly followed the design outlined by the


Transport and Main Roads Department of the
Queensland Government.21 The ashing red light
signal consists of twin red round lights arranged horizontally and equipped to ash alternately. The invehicle audible warning triggered verbal warnings:
Warning! Train approaching!, Train crossing! Stop
at the stop line! and Train departed. Please proceed.
The initial warning was played 21 s prior to the train
arriving at the crossing, the second was played as the
train passed and the third was played after the train
had departed.
The four warning devices and the associated
advance warning signs were included on the left shoulder of the roadway, as shown in Figure 2. The rail
track crossed the roadway at an angle of 90 . A simulated Queensland Rail train consisting of an engine
and three cars was used. The train was introduced
20 s after the warning was activated for active crossings or when within 40 m of the track for passive

Figure 2. The four warning devices used in the simulator. (a) Stop Sign, (b) Rumble Strips, (c) Flashing Light, and (d) In Vehicle.

432
crossings. The train produced a squealing sound of
wheels on tracks 10 s before the train crossed the
road, but only if the driver slowed to a stop at the
crossing, preventing the driver reacting to the train
per se. The initial speed of the car was chosen in
pseudorandom order to be 60 or 80 km/h. The participants were advised to maintain this xed maximum speed until they encountered a stimulus or
trac hazard which they should react to as they
would if driving in the real world. Potential reactions
included releasing the accelerator to slow down gradually, releasing the accelerator and pressing the brake
to slow down further or more abruptly, stop completely, slow down then accelerate to proceed, proceed
without slowing down.

Proc IMechE Part F: J Rail and Rapid Transit 227(5)

Figure 3. Data structure.

Table 2. Comparison of compliance behaviours (in percent)


for different warning devices for an approach speed of 60 km/h.

Procedure of main experiment


The data collected on the behaviour of all the drivers
was utilised in subsequent analyses. As shown in
Figure 3, in the rst stage of analysis data was
extracted 250 m away from the stop line. It was
programmed that 95 m after the starting point of the
sorted data was a threshold where warnings were triggered. At this point, drivers would receive warnings.
Just after this point, the data on when drivers started
to change their speed in response to the safety devices
and/or existence of a train were used.
For each test trial, data on vehicle trajectories were
recorded. These data were generated from the accelerator and brake pedal as a result of the drivers
responses/behaviour to the stimulus encountered in
the virtual environment. From the vehicle trajectories,
the following data were retrieved.
1. Driver stopping compliance at crossings (whether
subject stopped/crossed at crossings).
2. Approaching speed prole, from 250 to 0 m from
stop line.
3. Acceleration prole from 250 to 0 m from stop
line.
4. Position at which initial braking action occurred.
5. Position at which nal braking action occurred
before stopping.
After completing the test session, subjects were
asked to answer a post-experiment questionnaire,
focusing on the eectiveness of the warning devices.
A short interview was also conducted to discuss the
subjects responses in the experiment.

Results
Compliance
Drivers approaching a crossing with a stop sign are
expected to obey the regulatory sign and stop the
vehicle before the stop line, to look left and right
for train trac, regardless of the presence of a train.

Non-compliance
Devices

Comply

Slowdown

Drive
through

Stop sign
Rumble strips
Flashing light
In-vehicle audible warnings

74
72
100
99

16
17
0
0

10
11
0
1

Table 3. Comparison of compliance behaviours (in percent)


for different warning devices for an approach speed of 80 km/h.
Non-compliance
Devices

Comply

Slowdown

Drive
through

Stop sign
Rumble strips
Flashing light
In-vehicle audible warnings

67
72
97
93

22
19
1
3

11
8
1
4

For active systems, either the ashing red light or


the in-vehicle audible warning, drivers are required
by the Highway Code to stop when the red light or
audible warning is activated by an approaching
train. Thus, in the driving simulation experiment,
driver stopping compliance for passive devices (stop
sign and rumble strips) was investigated both with
and without a train present, while only the scenario
in which a train was present was used for active
devices (ashing red light and in-vehicle audible
warning).
Non-compliance behaviours were categorized into
two parts: slowdown and drive through. The denition of slowdown in this work is that vehicles reduced
speed then continued to cross. Drive through means
that vehicles neither fully stopped nor reduced
speed. As shown in Table 2, at the approach speed
of 60 km/h most drivers responded to the active

Kim et al.

433

Figure 4. Comparison of speed profile patterns for the four warning devices for vehicle approach speeds of 60 and 80 km/h.
(a) 60km/hr with Train, (b) 60km/hr without Train, (c) 80km/hr with Train, and (d) 80km/hr without Train.

systems, i.e. 100% for ashing red light and 99% for
in-vehicle audible warnings. As intuitively expected,
lower compliance rates were observed at passive
crossings, i.e. 74% for the stop sign and 72% for
rumble strips. Although the results for the approach
speed of 80 km/h showed a similar behaviour as for
60 km/h, the overall compliance rates reduced as
shown in Table 3. Active crossings showed higher
compliance rates: 97% for ashing red light and
93% for in-vehicle audible warning whereas passive
crossings had rates of 67% for stop sign and 72% for
rumble strips. In the non-complied categories, 19
22% reduced speed before violation whereas 811%
drove through without reducing speed at passive
crossings. The compliance rates for active crossings
for an approach speed of 80 km/h remained similar
to those for 60 km/h, 13% slowed down whereas
14% did not stop.22

Speed profiles
Approach speed proles for drivers who stopped at
the crossing were plotted. Speed was plotted against
distance to stop line. In these gures, the Y-axis represents the speed 150 m away from the stop line.
Vehicle trajectories represents the approaching
speed patterns and drivers responses towards varying
types of warning devices. The comparison of speed

proles for the four warning devices at the approach


speeds of 60 and 80 km/h are presented in Figure 4.
Overall, vehicle speed reduced as the drivers
approached the crossing.
As shown in Figure 4(a) and (c), speed proles of
active crossings (ashing red lights and in-vehicle
audible warnings) are similar to each other regardless of the approach speed (60 or 80 km/h). Also
speed proles at passive crossings (stop sign and
rumble strips) are similar to each other. When drivers approached a railway crossing equipped with
active crossings, they tended to rapidly reduce their
speed at approximately 72 m away from the stop line
whereas they did not abruptly reduce their speed at
passive crossings. The dierence in behaviour is due
to the lack of advance information to drivers about
needing to slow down. When it comes to a comparison of dierent approach speeds (60 and 80 km/h)
with the presence of a train, the speed proles
showed similar patterns for each device. However,
the reactions to warnings were closer to the stop
line for the 80 km/h case. Drivers tended to reduce
their speed at approximately 60 m away from the
stop line. This is a result of the initial speed of the
vehicle inuencing the drivers reaction times.
Figure 4(b) and (c) show a similar pattern of behaviour when there was no train. In both gures, the
speed of drivers for the ashing red lights remained

434

Proc IMechE Part F: J Rail and Rapid Transit 227(5)

Figure 5. Comparison of acceleration profile patterns for the four warning devices with vehicle approach speed of 60 or 80 km/h.
(a) 60km/hr with Train, (b) 60km/hr without Train, (c) 80km/hr with Train, and (d) 80km/hr without Train.

higher than other devices followed by in-vehicle audible warnings, stop sign and rumble strips. This is
because for the cases of ashing red lights and invehicle audible warnings the drivers know that there
is no train approaching the crossing and thus they
did not tend to reduce their speed. In comparison,
for the stop sign and rumble strip cases drivers need
to reduce speed and visually check if a train was
approaching the crossing.

Acceleration (deceleration) profile


Measuring acceleration (deceleration) is important in
order to understand how frequently and by how much
drivers change their speed. As for the speed results,
acceleration showed a dierent pattern between active
and passive crossings. The extent of deceleration was
also dierent based on the approach speed. For example, when the approach speed was 60 km/h and a train
was crossing, the maximum deceleration was
1.7  1.5 m/s2 at a distance of 65 m to the stop
line as shown in Figure 5(a). When the approach
speed was 80 km/h with the existence of a train, the
maximum deceleration was 3.1  3 m/s2 at a distance of 55 m as shown in Figure 5(c). When the railway crossing was equipped with passive devices, the
deceleration was gradual. The maximum deceleration
from the 80 km/h approach speed is double that of the

60 km/h approach speed. When there was no train,


graphs in Figure 5(b) and (d) showed that drivers
still slowed down until they were sure no train was
approaching at 60km/hr and 80 km/hr. In both
graphs, the acceleration at ashing-red-lights was
applied earlier than other devices.

Statistical analysis
Speed prole data were subjected to statistical analysis. A set of speeds from each participant was averaged for every single metre starting 150 m away from
the stop line. After that, the series of speeds was
checked to see if they were normally distributed by
using a quantile-quantile (Q-Q) plot. A Q-Q plot is
a graphical technique for determining if two data sets
come from populations with a common distribution.
The Q-Q plot shows the shape and type of departure
from normality. The Q-Q plot in Figure 6 shows evidence of an underlying distribution that has heavier
tails compared with those of a normal distribution.
Q-Q plots coupled with histograms can be used to
check the assumption that the underlying population
is normally distributed.
If the underlying population could not be determined to be normally distributed using a Q-Q plot,
a Wilcoxon signed-rank test was performed using
MINITAB 16 in order to determine if there was a

Kim et al.

435

Figure 6. Q-Q plots for determining the distribution of data.

dierence between paired samples. This test can also


be used to consider the magnitude of dierence
between the paired samples. The Friedman test that
is used to identify dierences in treatments across
multiple test attempts was not performed in this work.
The case of an approach speed of 80 km/h with an
approaching train was analysed as follows. The null
hypothesis for this test was that there was no dierence between the paired samples (here, one pair is
between active crossing and passive crossing; the
other pair is between two dierent active crossings).
The alternative hypothesis stated that there was a difference between the paired samples:

Table 4. Data comparison between flashing red lights and


stop signs.

H0: there is no dierence between the paired samples;


Ha: there is a dierence between the paired samples.

Table 5. Data comparison between flashing lights and in-vehicle warnings.

By referring to the speed and acceleration proles,


two dierent sets of two data were compared.
1. Flashing red lights versus stop signs (active crossing versus passive crossing).
2. Flashing red lights versus in-vehicle warnings
(active crossing versus active crossing).
The summary of ranks table provided some interesting results on the comparison of participants
behaviour on two dierent sets of data. Table 4
shows that 127(N) points have a higher value than
the other set whereas Table 5 shows that 86 (N)
points have a higher value than the other set. This
table explains how dierent active crossings are
from passive crossings. By examining the nal twotailed asymptotic test, Tables 4 and 5 demonstrate
whether the application of dierent safety treatments

Summary of ranks

Median

Sum

Positive ranks
Negative ranks
Zero differences

79.000
17.000
0.000

10178.000
1147.000
0.000

127
23
0

Two-tailed asymptotic test


Z

8.472

150

0.000

Summary of ranks

Median

Sum

Positive ranks
Negative ranks
Zero differences

56.500
118.500
0.000

4765.000
6560.000
0.000

86
64
0

Two-tailed asymptotic test


Z
1.684

N
150

P
0.092

led overall to a statistically signicant dierence in


speed. An asymptotic sigma (two-tailed) value (P
value), of <0.001 between ashing red lights and
stop signs is shown in Table 4 whereas it has a value
of 0.092 between ashing red lights and in-vehicle
warnings as shown in Table 5.
In Table 4, the Wilcoxon signed ranks test shows
that the speed in response to using ashing lights is

436

Proc IMechE Part F: J Rail and Rapid Transit 227(5)

signicantly dierent from that for stop signs with


(Z 8.472, P <0.001). As the P value is smaller
than the 5% threshold (0.005), the null hypothesis is
rejected. On the other hand, Table 5 shows the speed
resulting from using ashing lights was not signicantly dierent from in-vehicle warnings with
Z 1.684, P 0.092. As the P value is greater than
the 5% threshold (0.005), the null hypothesis is not
rejected.

Application to traffic simulation


The Vissim trac simulation software23 was enhanced
by using the results from the driving simulator in

Table 6. Relationships between terms in the driving simulator


and traffic simulation.
Data

Driving simulator

Traffic simulation (VISSIM)

1
2

Compliance rate
Speed profile

3
4
5

Accelerator profile
Initial braking position
Final braking position

Probability of stopping
Reduced speed area/
speed decision
Acceleration functions
Time to stop

developing a model. Due to dierent terminology


and dierent concepts, interpretation of the outputs
of the driving simulator and the inputs of the trac
simulation was needed. As shown in Table 6 compliance rates, speed proles, accelerator release positions
and initial/nal braking position in the driving simulator were interpreted as the probability of stopping,
reduced speed area/speed decision and time to stop,
respectively.
Compliance can be interpreted as the probability of
stopping. It can be used to mimic a real situation in
terms of the probability of stopping by changing the
values of speed and distance. During the amber period
in signal controls, vehicles make a decision to either
go or stop depending on their speed and distance to
a stop line. Since there are only two choices (stop or
cross), a binary logistic regression model was used to
estimate the probability of a driver stopping or crossing, as a function of the variables representing
the contributing factors under consideration.
Table 7 summarises the results of drivers stopping
compliance from the driving simulator experiments used for model development. The table shows
that the impact of dierent types of safety devices on
the compliance rate is greater than drivers speed and
distance between the drivers and the stop line during
warnings.
The probability model, p, is given by equations (1)
and (2)
p

1
1 eui

Table 7. Coefficient values for each safety device.


Coefficients

1

2

Stop sign
Rumble strips
Flashing light
In-vehicle audible warnings

0.236,956
0.143,968
0.395,141
0.544,755

0.0001
0.0001
0.0001
0.0001

0.013,101
0.016,765
0.015,423
0.016,105

Figure 7. The layout of the model in VISSIM.

ui  1 v 2 dx

where, v is speed, dx is distance, , 1 and 2 are


constants.
A signal at the railway crossing was controlled by
the component object model (COM) functionality in
VISSIM. The COM is an external programming

Kim et al.
environment that enabled the control of VISSIM
components. At every simulation second (here,
0.1 s), detectors in VISSIM send information about
how many vehicles have passed the detectors, how
quickly each vehicle has passed and how long a
vehicle is stationary at the detectors. As shown in
Figure 7, when a train passes impulse detector 1
(D1), the signal controller is activated which commands the signal to change from green to amber so
that vehicles heading into the intersection obtain a
long amber time (21 s in this work). Until the train
passes impulse detector 2 (D2), vehicles make a decision whether to go or stop based on equations (1) and
(2). Immediately after clearing the trap zone (D3), the
signal controller operates the phases as normal.
After calibrating parameters in VISSIM, the speed
proles obtained using the driving simulator and
VISSIM show similar patterns, for example, the crossing equipped with ashing lights shown in Figure 8.
This highlights that the driving behaviour in trac
simulation can closely represent the outputs of the
driving simulator.
Most trac simulation models use distributions
and functions to mimic a drivers behaviour based
on numerous factors such as the type of vehicle,
road condition, etc. By limiting the boundary of maximum acceleration or deceleration, trac simulations
do not consider abnormal movements of a vehicle.

437
Since the driving simulator provides speed proles
and acceleration proles at a time and position, it
can be used in trac simulation by editing the distributions and functions individually. Other indicators
such as initial and nal braking positions can be compared with the time to stop derived from trac simulations as a post-processing procedure. These outputs
can also be found by analysing the acceleration
proles.

Conclusions
There are several studies in the literature that have
considered the eects of driving behaviour and environment on creating road accidents; however, relatively little is known about collisions at railway
crossings. Following on from previously published
results5,22, the current paper delivers promising
insights into how to integrate the outputs of a driving
simulator and trac simulations. The aim of this
paper was to present a method to replicate ndings
from an expensive driving simulator using relatively
cheap trac simulations. The eects of a number of
parameters such as compliance rate, speed prole,
accelerator release position, initial and nal braking
positions were observed using a driving simulator.
These outputs were interpreted in the trac simulation model. This application can be extended to assess
dierent vehicles approaching railway crossings and
dierent train headways. The proposed approach can
be used to evaluate and compare safety performance
without the need to create test beds.
However, this research requires further investigation into the verication of parameters that aect the
results of driving simulators and trac simulation.
The scenarios in the driving simulator were based
on Australian trac rules and regulations. More
observations about driving behaviour need to be conducted to improve verication and usability under different circumstances. Additionally, the used xed
position driving simulator is not as sophisticated as
a state-of-the-art advanced driving simulator where
candidates actually sit inside a 360 movable vehicle
and the use of such a simulator would be of interest to
create more realistic situations. Also, since only four
scenarios were tested, more experiments are recommended to ensure better validity. Additional vehicle
types, such as heavy vehicles, buses and other vehicles
can also be simulated using the proposed technique.

Funding
This work was supported by the CRC for Rail Innovation
(established and supported under the Australian
Governments Cooperative Research Centres program).
Figure 8. Speed profile between driving simulator (DS) and
traffic simulation (TS) for the case (a) without train and (b) with
a train.

Acknowledgements
We thank Queensland Rail for assistance with data
collection.

438
References
1. Australian Transport Safety Bureau. Australian rail
safety occurrence data, http://www.atsb.gov.au/publications/2009/australian-rail-safety-occurrence-data-1january-2001-to-31-december-2009.aspx (2010, accessed
13 January 2011).
2. Lord D, Washington S and Ivan J. Further notes on the
application of zero-inflated models in highway safety.
Accid Analy Prevent 2007; 39(1): 5357.
3. Carrivick PJW, Lee AH and Yau KKW. Zero-inflated
Poisson modeling to evaluate occupational safety interventions. Saf Sci 2003; 41(1): 5363.
4. Anderson TK. Using geodemographics to measure and
explain social and environment differences in road traffic accident risk. Environ Plan A 2010; 42(9): 21862200.
5. Tey LS, Ferreira L and Wallace A. Measuring driver
responses at railway level crossings. Accid Analy
Prevent 2011; 43(6): 21342141.
6. Aarts L and Van Schagen I. Driving speed and the risk
of road crashes: a review. Accid Analy Prevent 2006;
38(2): 215224.
7. Richards D and Cuerden R. The relationship between
speed and car driver injury severity, Report,
Department for Transport, Canberra, Australia, 2009.
8. Elvik R. A restatement of the case for speed limits.
Transp Pol 2010; 17(3): 196204.
9. Mesken J, Lajunen T and Summala H. Interpersonal
violations, speeding violations and their relation to accident involvement in Finland. Ergonomics 2002; 45(7):
469483.
10. Whissell RW and Bigelow BJ. The speeding attitude
scale and the role of sensation seeking in profiling
young drivers at risk. Risk Anal 2003; 23(4): 811820.
11. Schmidt RA and Young DE. Cars gone wild: the major
contributor to unintended acceleration in automobiles
is pedal error. Front Psychol 2010; 1(209): 14.
12. Af Wahlberg A. The relation of acceleration force to
traffic accident frequency: a pilot study. Transp Res
Part F: Traffic Psychol Behav 2000; 3(1): 2938.

Proc IMechE Part F: J Rail and Rapid Transit 227(5)


13. Af Wahlberg A. The stability of driver acceleration
behavior, and a replication of its relation to bus accidents. Accid Analy Prevent 2004; 36(1): 8392.
14. Lee JD, McGehee DV, Brown TL and Reyes ML.
Collision warning timing, driver distraction, and
driver response to imminent rear-end collisions in a
high-fidelity driving simulator. Hum Fact: J Hum Fact
Ergonom Soc 2002; 44(2): 314334.
15. Chipman ML. Risk factors for injury: similarities and
differences for traffic crashes and other causes. Accid
Analy Prevent 1995; 27(5): 699706.
16. Archer J. Indicators for traffic safety assessment and
prediction and their application in micro-simulation modelling: a study of urban and suburban intersections,
Stockholm: Kungliga Tekniska Hogskolan, 2005; 5,
Doctoral thesis in Technology.
17. Gettman D and Head L. Surrogate safety measures
from traffic simulation models. Transp Res Record: J
Transp Res Board 2003; 1840(1): 104115.
18. Minderhoud MM and Bovy PHL. Extended time-tocollision measures for road traffic safety assessment.
Accid Analy Prevent 2001; 33(1): 8997.
19. Vladisavljevic I, Cooper JM, Martin PT and Strayer
DL (eds) Importance of integrating driving and traffic
simulations: case study of impact of cell phone drivers
on traffic flow, Transportation Research Board 88th
Annual Meeting, 2009; 09-0224.
20. Standards Australia. Manual of uniform traffic control
devices, 2009.
21. Transport and Main Roads Department, Queensland
Government. Traffic and road use management
manual, 2010.
22. Tey LS, Wallis G, Cloete S and Ferreira L. Evaluating
driver behaviour towards innovative warning devices at
railway level crossings using a driving simulator.
J Transport Saf Secur 2003; 5: 118130.
23. Planung Transport Verkehr AG. VISSIM version 5.3
user manual, 2011.

You might also like