Accident Analysis and Prevention 71 (2014) 177–182
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Microsimulation modelling of driver behaviour towards alternative warning devices at railway level crossings Li-Sian Tey a,c,∗ , Sicong Zhu a , Luis Ferreira a , Guy Wallis b a b c
Faculty of Engineering, Architecture and Information Technology, University of Queensland, QLD 4072, Australia School of Human Movement Studies and Queensland Brain Institute, University of Queensland, QLD 4072, Australia Faculty of Civil Engineering, Universiti Teknologi MARA, 13500 Permatang Pauh, Penang, Malaysia
a r t i c l e
i n f o
Article history: Received 17 July 2013 Received in revised form 10 April 2014 Accepted 21 May 2014 Keywords: Railway level crossing Alternative warning systems Driver behaviour Safety evaluation Microsimulation
a b s t r a c t Level crossings are amongst the most complex of road safety issues, due to the addition of rail infrastructure, trains and train operations. The differences in the operational characteristics of different warning devices together with varying crossing, traffic or/and train characteristics, cause different driver behaviour at crossings. This paper compares driver behaviour towards two novel warning devices (rumble strips and in-vehicle audio warning) with two conventional warning devices (flashing light and stop sign) at railway level crossings using microsimulation modelling. Two safety performance indicators directly related to collision risks, violation and time-to-collision, were adopted. Results indicated the active systems were more effective at reducing likely collisions compared to passive devices. With the combined application of driving simulation and traffic microsimulation modelling, traffic safety performance indicators for a level crossing can be estimated. From these, relative safety comparisons for the different traffic devices are derived, or even for absolute safety evaluation with proper calibration from field investigations. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Level crossings are amongst the most complex of road safety issues, due to the addition of rail infrastructure, trains and train operation. Generally, there are several contributory factors to level crossing collisions and these can be difficult to determine. Nevertheless, in Australia, collisions at crossing are reported to be mainly attributed to drivers’ responses to the warning devices (Australian Transport Council, 2003; Wallace et al., 2008; Chartier, 2000, etc.). The differences in the operational characteristics of different warning devices together with varying crossing, traffic and/or train characteristics, cause different driver behaviour at crossings. Several different types of warning devices are used at crossings, which recent research has shown have significantly different effects on driver behaviour (Yeh and Multer, 2007; Caird et al., 2002; Tey et al., 2011). In view of that, considerable research and innovation has occurred in some countries on the development of low-cost
∗ Corresponding author at: Faculty of Civil Engineering, Universiti Teknologi MARA, 13500 Permatang Pauh, Penang, Malaysia. Tel.: +60 43822527; mobile: +60 192300425. E-mail addresses: [email protected], [email protected] (L.-S. Tey). http://dx.doi.org/10.1016/j.aap.2014.05.014 0001-4575/© 2014 Elsevier Ltd. All rights reserved.
warning systems for level crossings. In the present study, rumble strips (a potential passive device) and in-vehicle audio warnings (a potential active in-vehicle device) were investigated. Human factors (driver behaviour) identified were generally driver characteristics, unintended human errors, intentional actions and risk seeking behaviour. Among driver characteristics, age-related (Schoppert and Hoyt, 1968; Yeh and Multer, 2007; Caird et al., 2002) and gender-related factors (Caird et al., 2002; Abraham et al., 1998; Tey et al., 2011) were acknowledged to be risk factors at level crossings. Compensatory and protective factors employed by older drivers were believed to reduce or control their risk. Younger drivers demonstrated a low perceived risk of consequences in relation to level crossing behaviour and subsequently reported the highest levels of risk taking of all the sub groups. A study of driver behaviour at 37 rail-highway crossings in Michigan, US, revealed that the drivers aged between 25 and 40 years was observed to commit more violations than any other age group. Of these, male drivers committed more violations than female drivers (Abraham et al., 1998). However, a previous study (Tey et al., 2011) conducted in Queensland, Australia, observed that more female drivers (24%) than male drivers (14%) drove through a passive crossing without stopping or slowing down. The relationship of drivers’ stopping compliance behaviour (violation) and braking responses at level crossings in particular to influences of speed, age and gender
178
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
has been studied and driver behavioural models developed (Tey et al., 2013a). These models were adopted in this paper for the microsimulation and are detailed in Section 2. Traffic micro-simulation models have a number of restrictions for traffic safety analyses. This limitation is mainly attributed to the high degree of variance in driver perception, reaction and driving behaviour and errors. Nevertheless, recent years have witnessed the appearance of several approaches to traffic safety issues in traffic micro-simulation. For instance, Cunto and Saccomanno (2008) used VISSIM 4.3 to estimate the safety performance for individual vehicles, which is expressed in terms of a crash potential index at the intersection, Piao and Mcdonald (2008) used AIMSUN to assess safety impacts of Variable Speed Limits (VSL) while Ozbay et al. (2008) suggested modified-TTC (MTTC) and a new crash index (CI) in conjunction with use of Paramics. In this paper, two safety performance indicators directly related to collision risks were adopted; namely, violation and ‘time-to-collision’ (TTC). ‘Violation’ refers to non-adherence of the traffic rules or traffic control devices (i.e., running red light, not stopping at a stop sign, etc.). Abraham et al. (1998), in their studies to test the relationships between driver violations and railway level crossing collisions, revealed promising use of violation data in determining the relative hazardousness of level crossings in combination with crash histories. Violation data were related to driver characteristics such as age, gender and to the types of warning devices (Tey et al., 2013a). Violation data may also be used to develop countermeasures that would help alleviate violations and eventually traffic collision problems at railway level crossings. The concept of TTC was defined by the US researcher Hayward (1972) as the time at which two vehicles would possibly collide if they keep their current speed and steering (Hayward, 1972; Hyden, 1996; Lundgren and Tapani, 2006). The TTC value decreases with time to ‘zero’ as the vehicles approach their conflict point and collide. The value of TTC in various situations in which traffic conflicts frequently happen has been studied by many researchers (Van Der Horst, 1991; Hirst and Graham, 1997; Hogema et al., 1996; Van Der Horst, 1990). The minimum and critical TTC values identified for approaches at intersections are 1.1 and 1.6 s (Van Der Horst and Brown, 1989) and 1 and 1.5 s (Van Der Horst, 1991), respectively, while the critical TTC for unintentionally dangerous situations is 4 s (Hirst and Graham, 1997). This paper incorporates and presents an application of the driver behavioural models (Tey et al., 2013a) into a microscopic traffic simulation using MATLAB in order to model driver behaviour for safety performance evaluation in terms of the likely number of collisions and TTC. It shows the potential of combined application of driving simulation and traffic microsimulation modelling for evaluating safety performance of the railway level crossing warning systems. The paper is structured as follows: Section 2 provides background information of data collection and driver behavioural models developed; Section 3 discusses the contributing variables under consideration; Section 4 provides a brief description of model development and the results of the simulation; and Section 5 concludes the main findings.
2. Background of data collection and the driver behavioural models The data used for the driver behavioural models adopted in this paper were collected from a driving simulator experiment (Tey et al., 2013a). Twenty four volunteers ranging in age from 17 to 66 years were recruited from the local community and The University of Queensland for a driving simulation experiment conducted in a fixed-base driving simulator located in 10 m × 5 m laboratory. The simulator comprised an overhead projector, a force-feedback steering wheel, and an accelerator and brake pedals. Three-dimensional
images were projected onto a 3.2 m × 2.7 m flat, white projection screen at a distance of 2 m from the ‘driving seat’. A controlling computer recorded foot pedal and steering-wheel data of each frame. A virtual environment was developed, which consisted of a simulated two-lane two-way road with a level crossing. Four different types of warning devices appeared randomly at the crossing. Two of the conventional warning devices (stop sign and flashing red-lights with bell) were included as ‘baseline’ comparisons with two innovative warning devices (rumble strip with stop sign and in-vehicle audio warning). The stop sign and rumble strip (with stop sign) are passive devices while flashing red-lights and in-vehicle audio warnings are activated by train presence at a single track crossing. Rumble strips alert drivers of a crossing ahead through vibration and sound. In the simulation this was imitated by vibrating the forcefeedback steering wheel. The in-vehicle audio warning triggered verbal cautions: ‘Warning! Train approaching!’, ‘Train crossing! Stop at the stop line!’ and ‘Train departed. Please proceed’. The participant drivers were advised to maintain the fixed maximum speed assigned until they encountered a stimulus or traffic hazard where they were expected to react as they would in the real world. For each test trial, data on vehicle trajectories, including brake and accelerator activation, were recorded. From the vehicle trajectories, the following data were retrieved: (i) Driver stopping compliance at crossings (whether subject stopped or crossed at crossings); (ii) Position at which driver released the accelerator; (iii) Position at which initial brake was applied; and (iv) Position at which final brake (final maximum slope change of time-space curve) was applied before stopping. After data analysis, regression models were developed to reflect driver’s responses towards the four different devices. The contributing variables tested were gender, age, speed and warning devices. Different variables were found significant for different approach stages to the level crossings at different levels of confidence statistically. Driver behavioural models were developed. The regression models included a binary choice model for predicting the probability of a driver stopping or driving through a railway crossing, as well as mixed regression models for predicting the moment at which a driver produced specific behavioural responses before stopping at the crossing (e.g., initiation of accelerator release and application of the brake foot–pedal); namely, initiation of accelerator release (AccR), initial (IniBr) and final (FinalBr) applications of brake foot–pedal, measured in distance from the stop line (m), in the form of Eqs. (1)–(4). Pi (cross) =
1 1 + e−zi
zi = −1.26 − 0.96Xgender + 1.72Xm.age − 3.12XFL − 2.39XIV
(1)
Pi (stop) = 1 − P(cross) AccRij = 182.36 − 11.12XFL − 7.82Xspeed
(2)
IniBrij = 104.97 − 17.36XFL − 13.84XIV − 13.92Xspeed
(3)
FinalBrij = 20.73 − 4.96Xgender + 5.46Xy.age + 16.3XFL + 18.38XIV (4) where Pi (cross = 1): the probability of the ith vehicle crossing (violating warning device); Pi (stop = 0): the probability of the ith vehicle stopping (complying to warning device); Xspeed : speed, comparing 80 km/h (1) to 60 km/h (0); Xgender : gender, comparing female (1) to male (0); Xm.age : age, comparing the age group of 31–50 years (1) to the age group of >50 years (0); Xy.age : age, comparing the age group of 17–30 years (1) to the age group of >50
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
179
Table 1 Distribution of male and female drivers. Number
Male Female
1,612,887 1,519,454
51.5 48.5
Total
3,132,341
100.0
%
>75
70-74
60-69
50-59
30-39
40-49
25-29
20
21-24
19
18
17
25% 20% 15% 10% 5% 0% 16
Percentage
Gender
Age Group Fig. 1. Distribution of driver licences by age group.
Percentage
50% 40%
Fig. 3. A roadway approaching a level crossing created in the simulation model
30% 20% 10% 0%
24% 16-29
38%
38%
30-49 Age Group
>50
Fig. 2. Distribution of driver licences by age group in simulation.
years (0); XFL : the type of devices, comparing flashing light (1) to stop sign (0); XIV : the type of devices, comparing in-vehicle warning (1) to stop sign (0). The models reflect that female drivers were less likely to violate warnings than male drivers while middle age group were more likely to violate the warning compared to the senior age group. The violation behaviour might possibly be due to drivers’ unintentional actions, or their driving style. The small variation in terms of stopping compliance between the two active systems reported is likely due to the fixed design warning activation time used in the laboratory experiments which could be further optimised in the future. As for the behavioural responses, speed affected drivers’ reaction in the early stage when approaching the level crossings (AccR and IniBr) and types of warning devices, age and gender had more influence at the final stages of stopping (FinalBr). The experimental procedure and the development of the driver behavioural models are detailed in previous publications (Tey et al., 2013a,b). There are limitations to directly adopting the behavioural models developed due to aspects such as small sample size and low model sensitivity. Nevertheless, the intention of this research was to explore the potential application of the behavioural models using drivers’ stopping compliance and reaction positions in a micro-simulation model for relative safety evaluation of alternative devices, as detailed in the following sections. 3. Contributing variables and their distribution For simulation purposes in the current paper, the distribution of gender and age for Queensland according to the Department of Transport and Main Roads (2010) was adopted. There was almost equal numbers of male and female drivers in Queensland as shown in Table 1. The distribution of driver licences in Queensland based on age is shown in Fig. 1. Fig. 2 shows the distribution
of driver licences adopted for simulation purposes in the current paper, based on the age group categories in the driving simulator experiment. Two speeds were adopted in the driving simulator experiment, 60 km/h or 80 km/h. 4. Development of simulation model MATLAB 7.11.0 was used for the simulation of driver behaviour. The model was refined to include appropriate driver–vehicle categories such as gender and age, probability to stop at level crossing, foot–pedal activation positions approaching level crossing prior to stopping at stop line, and a range of deceleration rates. A direct link (roadway) approaching a level crossing was created as shown in the schematic diagram (Fig. 3). A simulation of 365 days with two peaks/day was applied. Trains were assigned based on: (1) uniform distribution (with maximum train incoming time of 60 min); and (2) normal distribution (with mean train incoming time of 30 min). Both distributions resulted in a train incoming time of 30 min. A negative exponential distribution was used for generating the inter-arrival time of vehicles (vehicle entry headways) to the simulation model. The simulation began with ten (10) trains/day and eight (8) vehicles/h. There were two parts of the analysis conducted: collision likelihood and TTC. A ‘collision likelihood’ (potential number of conflicts) was recorded when a vehicle entered the ‘hazardous zone’, as denoted in Fig. 3, while a train was approaching the level crossing. TTC was identified as the warning was activated or the train was within 20 s of the crossing. TTC is defined herein as the time between a vehicle arriving at the stop line and a train arriving at the crossing (time vehicle arrives at crossing – time train arrives at crossing). Discrete-event simulation was adopted. The random samples used in the Monte Carlo simulation were generated by MATLAB functions. The reliability of the random number generator approach used in the MATLAB function has been confirmed by Kroese et al. (2011). Comparable random variables defined over successive regenerative cycles are statistically independent and identically distributed (Crane and Lemoine, 1977). Each simulation run consisted of 10,000 iterations. For different incoming traffic volumes (8 and 50 vehicles/h), the mean number of vehicles with TTC less than 30 s was generated. Fig. 4 shows an example of the frequency of the occasion (TTC < 30 s) from each iteration of one run
180
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
4.1. Simulation results
Fig. 4. Frequencies of TTC < 30 s from a test run.
and the mean results are 46 and 304 for incoming traffic volume of 8 and 50 vehicles/h, respectively. It is believed that a TTC between a vehicle and a train greater than 30 s provided adequate time for a driver to react appropriately (stop) or to cross the crossing safely before the arrival of the train, thus, collision was unlikely to happen. It was assumed that a driver would not collide with a train after seeing the train passing the crossing. Vehicles with a TTC of less than 30 s were subsequently further evaluated for the numbers of potential collisions (collision likelihood) and the TTC. For each vehicle entering the system, different conditions were assigned randomly to each of the contributing variables (gender, age and speed). Driver stopping compliance and response patterns were simulated to comply with the regression models developed as given in Section 2, with the corresponding contributing variables assigned. Therefore, the simulation model in the current paper incorporated human factors (driver characteristics of age and gender) and driver behaviour (stopping compliance and foot–pedal activations) into traffic microsimulation by applying the driver behavioural models developed earlier. A degree of variance between individual drivers is introduced by including the driver behavioural models and the inclusion of distributions that can be sampled by randomisation functions as verified in Fig. 5. After the surrogate measures were obtained from the simulation model, the alternatives devices were evaluated for their relative safety performance. The collision likelihood and TTC distribution among the four types of warning devices were compared.
Fig. 5. Verification of random number that controls the input of gender, speed, age and device.
4.1.1. Collision likelihood The stop-or-go decision data can be represented by a binary logistic regression function that can be used to estimate the probability of the driver stopping given his/her age, gender and the warning device. Since there is a high possibility of a conflict if the vehicles do not comply with the warning, the total number of vehicles entered the ‘hazardous zone’, as denoted in Fig. 3, as the train approached was recorded as potential conflicts (collision likelihood). The simulation began with ten (10) trains/day and eight (8) vehicles/h, then increased to fifty (50) vehicles/h. The simulation was regenerated for 15 cycles to obtain the average number of potential conflicts. The number of likely conflicts was compared for the four warning devices: flashing lights, in-vehicle audio warning, rumble strips and stop sign. Table 2 summarises the results based on the number of vehicles. When the traffic flow is 8 vehicles/h, no potential conflicts were recorded for the flashing lights and one potential conflict for the invehicle audio warning; whereas 12 potential conflicts were found at the stop sign and rumble strips for vehicles travelling at both speeds. When the traffic increased to 50 vehicles/h, the potential conflicts increased to greater than 600% for vehicles travelling at both speeds. As expected, the increment of traffic volume resulted in higher numbers of potential conflicts. In relative terms, the potential conflicts at crossings with active systems (flashing lights and in-vehicle audio warning) were much less than at crossings with passive devices (stop sign and rumble strips). The minor conflict effect is only true as models were developed using results from the simulator experiment where very little or no violation was compiled for active devices. If adjustment from field results (Tey et al., 2011) was considered, the results of collision likelihood in the traffic simulation reported would have increased by 26–33% for passive devices and 27–30% for active devices. The calibration can be done when a larger sample size from both the field and experimental results are available. Nevertheless, the simulation results show the potential of integrating the driving simulator and traffic microsimulation for relative safety assessment of various warning devices, or even for absolute safety evaluation when using proper calibrations from field investigations. 4.1.2. Time-to-collision Drivers know that there is at least some length of time before a train’s arrival at a crossing. However, they have been found unable to accurately judge the speed and distance of an oncoming train (Cohn and Nguyen, 2003; Cooper and Ragland, 2008) and their responses following activation of a warning device are highly variable, which produces varying TTC at level crossings. Trains are not able to stop promptly due to their high momentum. Therefore, TTC can be a surrogate measure to observe the potential of a collision. The severity of a collision may be established in the future with further investigation on the relationship between the TTC, collision and the severity of a collision. A range of approximate deceleration rates shown in Table 3 was adopted in order to stop the vehicles at the distance-to-stop line of not greater than 2 m. These values of deceleration rate used in the simulation might not reflect the real life situation on the road. Nevertheless, for the purpose of presenting the approach showing effect of TTC solely from the driver responses, fixed values were adopted. Table 3 also presents the simulation results of the average TTC of vehicles travelling at 80 km/h with respect to the four warning devices. An analysis of the TTC distribution revealed that the average TTC were greater than 5 s for the active devices while both passive devices resulted in TTC close to 5 s. The results of the TTC would be varied once the deceleration rates were changed. This issue can be
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
181
Table 2 Number of potential conflicts. Seed
Traffic volume = 50 vehicles/h Train volume = 10 trains/day
Traffic volume = 8 vehicles/h Train volume = 10 trains/day Flashing lights
In-vehicle audio warning
Stop signs
Rumble strips
Flashing lights
In-vehicle audio warning
Stop signs
Rumble strips
Run 1 Run 2 Run 3 Run 4 Run 5 Run 6 Run 7 Run 8 Run 9 Run 10 Run 11 Run 12 Run 13 Run 14 Run 15
0 0 1 0 1 1 1 1 0 0 1 0 1 1 1
2 2 1 2 2 2 2 2 2 2 2 1 2 2 1
14 9 12 12 14 13 13 12 12 14 13 13 11 14 12
13 11 14 11 10 13 11 14 14 14 12 13 12 15 13
6 7 6 6 6 6 6 6 6 6 7 6 7 7 6
14 12 13 12 14 12 13 14 13 12 15 14 14 12 13
87 87 86 90 86 86 87 86 88 91 85 90 86 88 89
88 85 89 87 88 86 85 85 85 82 85 90 90 86 92
Number of potential conflicts
0.6
1.8
12.5
12.7
6.3
13.1
87.5
86.9
Table 3 Simulation results of time-to-collision. Flashing light
In-vehicle audio warning
Approaching speed (km/h)
Rumble strips
Stop sign
80
Deceleration rate (m/s2 ) From 200 m to AccR From AccR to IniBr From IniBr to FinalBr
0.5 2.5 3.0
0.5 2.5 3.0
0.5 2.35 2.0
0.5 2.35 2.0
Time-to-collision (s)
7.22
7.20
5.69
5.71
resolved with further research and site investigation to include the deceleration rate as one of the variables of the model. Nevertheless, as the main aim of this research was to apply the concept of incorporating human factors (driver behaviour) and surrogate safety measures in traffic simulation for safety evaluation, the application of deceleration rates were simplified as presented. The increase of average TTC from approximately ‘5 s’ to ‘7 s’ when the active devices (flashing lights or in-vehicle audio warning) are in operation, supports the safety improvement tendency of the active systems over the passive devices. This finding is consistent with real-world situations in field investigations (Tey et al., 2011). The two alternatives, in-vehicle audio warning and rumble strips, provided comparable results to the conventional flashing light and stop sign, respectively. Thus, in-depth research into the two alternatives is recommended for cost effectiveness. 5. Summary and discussion Historically, the safety of new and innovative traffic treatments has been difficult to assess since they are not yet widely used and direct safety measures such as accident and fatality frequencies cannot be obtained in many cases. Since empirical collection of accident data is not yet an option, other methods for safety assessment are needed. To this end, microscopic simulation is a useful approach. With the application of this tool a variety of traffic safety indicators can be adopted, such as collision likelihood and TTC distribution. Also, a comparison of TTC can be made to evaluate safety changes. Nevertheless, absolute safety effects are hard to derive with such comparative analyses unless adequate field studies are conducted. The presented safety indicators are especially (but not
exclusively) suited for application in microscopic simulation studies of traffic, in which much more information is available about individual vehicle movements than is usually the case in empirical studies. Traffic microsimulation models offer many benefits in safety analysis techniques. There is, however, a considerable need to develop robust procedures and models through direct comparison, calibration, and validation of the models with real-world data. More research into other potential variables, such as familiarity at level crossing, risk taking behaviour, differences between urban and rural settings and environmental complexity, would also worth investigated. Despite these limitations, the suggested approach does allow for the following conclusions: • The number of likely collisions for different warning devices has been estimated by incorporating experimental results of drivers’ responses in the model; • The difference found in terms of the number of likely collisions at crossings within either active or passive devices is itself relatively small; • The number of likely collisions at stop signs and rumble strips are significantly higher than for flashing lights and in-vehicle audio warning systems; • Evaluation of alternative level crossing safety devices can be conducted using traffic microsimulation with some modifications; • Evaluation on alternative level crossing safety devices can include human factors (driver behaviour); • The outcome can be better established with adjustments using adequate actual field violations and reaction data; • This tool will assist in determining which safety approach is more cost-effective (best value for capital outlay) for specified traffic and train volumes. The approach presented here is able to model the drivers’ behaviour towards alternative devices at the approach to the crossing. The use of micro-simulation modelling has the advantage of enabling the effectiveness of alternative treatment candidates to be assessed before expensive trials are implemented. The effect of the crossing on the nearby road network, including signalised intersections, can also be studied subsequently. The use of microsimulation models may be able to reduce the need for expensive driver simulator experiments when evaluating driver behaviour in the future.
182
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
Acknowledgments The authors would like to thank Queensland Rail and the CRC for Rail Innovation for their support. Li-Sian Tey acknowledges financial assistance from a Malaysia Higher Education Scholarship and the Universiti Teknologi MARA. Guy Wallis was funded by an Australian Research Council Future Fellowship. References Abraham, J., Datta, T.K., Datta, S., 1998. Driver behaviour at rail-highway crossings. Transport. Res. Rec.: J. Transport. Res. Board 1648, 28–34. Australian Transport Council, 2003. National Railway Level Crossing Safety Strategy. Australian Transport Council, Canberra. Caird, J.K., Creaser, J.I., Edwards, C.J., Dewar, R.E., 2002. A Human Factors Analysis of Highway–Railway Grade Crossing Accidents in Canada. University of Calgary, Alberta. Chartier, P., 2000. Level crossing safety in countries of the Asia-Pacific region: comparative analysis – problems identified – possible remedial actions. In: 22nd ASEAN Railway General Managers’ Conference, Yangon, Myanmar. Cohn, T., Nguyen, T., 2003. Sensory cause of railroad grade-crossing collisions: test of the leibowitz hypothesis. Transport. Res. Rec.: J. Transport. Res. Board 1843, 24–30. Cooper, D.L., Ragland, D.R., 2008. Addressing Inappropriate Driver Behaviour at RailHighway Crossings. UC Berkeley Traffic Safety Center, Institute of Transportation Studies. Crane, A., Lemoine, A.J., 1977. An Introduction to the Regenerative Method for Simulation Analysis. Springer-Verlag, New York. Cunto, F., Saccomanno, F., 2008. Calibration and validation of simulated vehicle safety performance at signalized intersections. Accid. Anal. Prev. 40, 1171–1179. Department of Transport and Main Roads, 2010. Queensland Current Driver Licences by age group as at 30 June 1994 to 2010. 2008–2010. Queensland Government. Hayward, J.C., 1972. Near-miss determination through use of a scale of danger. Highw. Res. Rec. 384, 24–34. Hirst, S., Graham, R., 1997. The format and presentation of collision warnings. In: Noy, Y.I. (Ed.), Ergonomics and Safety of Intelligent Driver Interfaces. Lawrence Erlbaum Associates Inc., New Jersey, United States of America.
Hogema, J.H., Janssen, W.H., Coemet, M., 1996. Effects of Intelligent Cruise Control on Driving Behaviour: A Simulator Study., pp. 2081. Hyden, C., 1996. Traffic conflicts technique: state-of-the-art. In: Topp, H.H. (Ed.), Traffic Safety Work with Video Processing. Transportation Department, University Kaiserslautern. Kroese, D.P., Taimre, T., Botev, Z.I., 2011. Handbook of Monte Carlo Methods. John Wiley and Sons Inc., New Jersey. Lundgren, J., Tapani, A., 2006. Evaluation of safety effects of driver assistance systems through traffic simulation. Transport. Res. Rec.: J. Transport. Res. Board 1953, 81–88. Ozbay, K., Yang, H., Bartin, B., Mudigonda, S., 2008. Derivation and validation of new simulation-based surrogate safety measure. Transport. Res. Rec.: J. Transport. Res. Board 2083, 105–113. Piao, J., Mcdonald, M., 2008. Year safety impacts of variable speed limits – a simulation study. In: 11th International IEEE Conference on Intelligent Transportation Systems, 10–12 December 2008. Institute of Electrical and Electronics Engineers Inc., Beijing, China, pp. 833–837. Schoppert, D.W., Hoyt, D.W., 1968. Factors Influencing Safety at HighwayRailway Grade Crossings. Washington: Transportation Research Board. National Research Council. Tey, L.S., Ferreira, L., Wallace, A., 2011. Measuring driver responses at railway level crossings. Accid. Anal. Prev. 43, 2134–2141. Tey, L.S., Wallis, G., Cloete, S., Ferreira, L., 2013a. Modelling driver behaviour towards innovative warning devices at railway level crossings. Accid. Anal. Prev. 51, 104–111. Tey, L.S., Wallis, G., Cloete, S., Ferreira, L., Zhu, S., 2013b. Evaluating driver behaviour towards innovative warning devices at railway level crossings using a driving simulator. J. Transport. Saf. Security 5, 118–130. Van Der Horst, A.R.A., 1990. A Time-based Analysis of Road User Behaviour in Normal and Critical Encounters. Technische Universiteit Delft, Delft, Netherlands. Van Der Horst, A.R.A., Brown, G.R., 1989. Time-to-Collection and Driver Decisionmaking in Braking. TNO Institute for Perception, Soesterberg. Van Der Horst, R., 1991. Time-to-collision as a cue for decision making in braking. Vision Vehicles 3, 19–26. Wallace, A., Mccusker, L., Hirsch, L., 2008. Level Crossings Research Database. CRC for Rail Innovation, Brisbane, Australia. Yeh, M., Multer, J., 2007. Year applying a sociotechnical framework for improving safety at highway-railroad grade crossings. In: Human Factors and Ergonomics Society 51st Annual Meeting, 1–5 October 2007, Baltimore, pp. 1550–1554.
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Microsimulation modelling of driver behaviour towards alternative warning devices at railway level crossings Li-Sian Tey a,c,∗ , Sicong Zhu a , Luis Ferreira a , Guy Wallis b a b c
Faculty of Engineering, Architecture and Information Technology, University of Queensland, QLD 4072, Australia School of Human Movement Studies and Queensland Brain Institute, University of Queensland, QLD 4072, Australia Faculty of Civil Engineering, Universiti Teknologi MARA, 13500 Permatang Pauh, Penang, Malaysia
a r t i c l e
i n f o
Article history: Received 17 July 2013 Received in revised form 10 April 2014 Accepted 21 May 2014 Keywords: Railway level crossing Alternative warning systems Driver behaviour Safety evaluation Microsimulation
a b s t r a c t Level crossings are amongst the most complex of road safety issues, due to the addition of rail infrastructure, trains and train operations. The differences in the operational characteristics of different warning devices together with varying crossing, traffic or/and train characteristics, cause different driver behaviour at crossings. This paper compares driver behaviour towards two novel warning devices (rumble strips and in-vehicle audio warning) with two conventional warning devices (flashing light and stop sign) at railway level crossings using microsimulation modelling. Two safety performance indicators directly related to collision risks, violation and time-to-collision, were adopted. Results indicated the active systems were more effective at reducing likely collisions compared to passive devices. With the combined application of driving simulation and traffic microsimulation modelling, traffic safety performance indicators for a level crossing can be estimated. From these, relative safety comparisons for the different traffic devices are derived, or even for absolute safety evaluation with proper calibration from field investigations. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Level crossings are amongst the most complex of road safety issues, due to the addition of rail infrastructure, trains and train operation. Generally, there are several contributory factors to level crossing collisions and these can be difficult to determine. Nevertheless, in Australia, collisions at crossing are reported to be mainly attributed to drivers’ responses to the warning devices (Australian Transport Council, 2003; Wallace et al., 2008; Chartier, 2000, etc.). The differences in the operational characteristics of different warning devices together with varying crossing, traffic and/or train characteristics, cause different driver behaviour at crossings. Several different types of warning devices are used at crossings, which recent research has shown have significantly different effects on driver behaviour (Yeh and Multer, 2007; Caird et al., 2002; Tey et al., 2011). In view of that, considerable research and innovation has occurred in some countries on the development of low-cost
∗ Corresponding author at: Faculty of Civil Engineering, Universiti Teknologi MARA, 13500 Permatang Pauh, Penang, Malaysia. Tel.: +60 43822527; mobile: +60 192300425. E-mail addresses: [email protected], [email protected] (L.-S. Tey). http://dx.doi.org/10.1016/j.aap.2014.05.014 0001-4575/© 2014 Elsevier Ltd. All rights reserved.
warning systems for level crossings. In the present study, rumble strips (a potential passive device) and in-vehicle audio warnings (a potential active in-vehicle device) were investigated. Human factors (driver behaviour) identified were generally driver characteristics, unintended human errors, intentional actions and risk seeking behaviour. Among driver characteristics, age-related (Schoppert and Hoyt, 1968; Yeh and Multer, 2007; Caird et al., 2002) and gender-related factors (Caird et al., 2002; Abraham et al., 1998; Tey et al., 2011) were acknowledged to be risk factors at level crossings. Compensatory and protective factors employed by older drivers were believed to reduce or control their risk. Younger drivers demonstrated a low perceived risk of consequences in relation to level crossing behaviour and subsequently reported the highest levels of risk taking of all the sub groups. A study of driver behaviour at 37 rail-highway crossings in Michigan, US, revealed that the drivers aged between 25 and 40 years was observed to commit more violations than any other age group. Of these, male drivers committed more violations than female drivers (Abraham et al., 1998). However, a previous study (Tey et al., 2011) conducted in Queensland, Australia, observed that more female drivers (24%) than male drivers (14%) drove through a passive crossing without stopping or slowing down. The relationship of drivers’ stopping compliance behaviour (violation) and braking responses at level crossings in particular to influences of speed, age and gender
178
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
has been studied and driver behavioural models developed (Tey et al., 2013a). These models were adopted in this paper for the microsimulation and are detailed in Section 2. Traffic micro-simulation models have a number of restrictions for traffic safety analyses. This limitation is mainly attributed to the high degree of variance in driver perception, reaction and driving behaviour and errors. Nevertheless, recent years have witnessed the appearance of several approaches to traffic safety issues in traffic micro-simulation. For instance, Cunto and Saccomanno (2008) used VISSIM 4.3 to estimate the safety performance for individual vehicles, which is expressed in terms of a crash potential index at the intersection, Piao and Mcdonald (2008) used AIMSUN to assess safety impacts of Variable Speed Limits (VSL) while Ozbay et al. (2008) suggested modified-TTC (MTTC) and a new crash index (CI) in conjunction with use of Paramics. In this paper, two safety performance indicators directly related to collision risks were adopted; namely, violation and ‘time-to-collision’ (TTC). ‘Violation’ refers to non-adherence of the traffic rules or traffic control devices (i.e., running red light, not stopping at a stop sign, etc.). Abraham et al. (1998), in their studies to test the relationships between driver violations and railway level crossing collisions, revealed promising use of violation data in determining the relative hazardousness of level crossings in combination with crash histories. Violation data were related to driver characteristics such as age, gender and to the types of warning devices (Tey et al., 2013a). Violation data may also be used to develop countermeasures that would help alleviate violations and eventually traffic collision problems at railway level crossings. The concept of TTC was defined by the US researcher Hayward (1972) as the time at which two vehicles would possibly collide if they keep their current speed and steering (Hayward, 1972; Hyden, 1996; Lundgren and Tapani, 2006). The TTC value decreases with time to ‘zero’ as the vehicles approach their conflict point and collide. The value of TTC in various situations in which traffic conflicts frequently happen has been studied by many researchers (Van Der Horst, 1991; Hirst and Graham, 1997; Hogema et al., 1996; Van Der Horst, 1990). The minimum and critical TTC values identified for approaches at intersections are 1.1 and 1.6 s (Van Der Horst and Brown, 1989) and 1 and 1.5 s (Van Der Horst, 1991), respectively, while the critical TTC for unintentionally dangerous situations is 4 s (Hirst and Graham, 1997). This paper incorporates and presents an application of the driver behavioural models (Tey et al., 2013a) into a microscopic traffic simulation using MATLAB in order to model driver behaviour for safety performance evaluation in terms of the likely number of collisions and TTC. It shows the potential of combined application of driving simulation and traffic microsimulation modelling for evaluating safety performance of the railway level crossing warning systems. The paper is structured as follows: Section 2 provides background information of data collection and driver behavioural models developed; Section 3 discusses the contributing variables under consideration; Section 4 provides a brief description of model development and the results of the simulation; and Section 5 concludes the main findings.
2. Background of data collection and the driver behavioural models The data used for the driver behavioural models adopted in this paper were collected from a driving simulator experiment (Tey et al., 2013a). Twenty four volunteers ranging in age from 17 to 66 years were recruited from the local community and The University of Queensland for a driving simulation experiment conducted in a fixed-base driving simulator located in 10 m × 5 m laboratory. The simulator comprised an overhead projector, a force-feedback steering wheel, and an accelerator and brake pedals. Three-dimensional
images were projected onto a 3.2 m × 2.7 m flat, white projection screen at a distance of 2 m from the ‘driving seat’. A controlling computer recorded foot pedal and steering-wheel data of each frame. A virtual environment was developed, which consisted of a simulated two-lane two-way road with a level crossing. Four different types of warning devices appeared randomly at the crossing. Two of the conventional warning devices (stop sign and flashing red-lights with bell) were included as ‘baseline’ comparisons with two innovative warning devices (rumble strip with stop sign and in-vehicle audio warning). The stop sign and rumble strip (with stop sign) are passive devices while flashing red-lights and in-vehicle audio warnings are activated by train presence at a single track crossing. Rumble strips alert drivers of a crossing ahead through vibration and sound. In the simulation this was imitated by vibrating the forcefeedback steering wheel. The in-vehicle audio warning triggered verbal cautions: ‘Warning! Train approaching!’, ‘Train crossing! Stop at the stop line!’ and ‘Train departed. Please proceed’. The participant drivers were advised to maintain the fixed maximum speed assigned until they encountered a stimulus or traffic hazard where they were expected to react as they would in the real world. For each test trial, data on vehicle trajectories, including brake and accelerator activation, were recorded. From the vehicle trajectories, the following data were retrieved: (i) Driver stopping compliance at crossings (whether subject stopped or crossed at crossings); (ii) Position at which driver released the accelerator; (iii) Position at which initial brake was applied; and (iv) Position at which final brake (final maximum slope change of time-space curve) was applied before stopping. After data analysis, regression models were developed to reflect driver’s responses towards the four different devices. The contributing variables tested were gender, age, speed and warning devices. Different variables were found significant for different approach stages to the level crossings at different levels of confidence statistically. Driver behavioural models were developed. The regression models included a binary choice model for predicting the probability of a driver stopping or driving through a railway crossing, as well as mixed regression models for predicting the moment at which a driver produced specific behavioural responses before stopping at the crossing (e.g., initiation of accelerator release and application of the brake foot–pedal); namely, initiation of accelerator release (AccR), initial (IniBr) and final (FinalBr) applications of brake foot–pedal, measured in distance from the stop line (m), in the form of Eqs. (1)–(4). Pi (cross) =
1 1 + e−zi
zi = −1.26 − 0.96Xgender + 1.72Xm.age − 3.12XFL − 2.39XIV
(1)
Pi (stop) = 1 − P(cross) AccRij = 182.36 − 11.12XFL − 7.82Xspeed
(2)
IniBrij = 104.97 − 17.36XFL − 13.84XIV − 13.92Xspeed
(3)
FinalBrij = 20.73 − 4.96Xgender + 5.46Xy.age + 16.3XFL + 18.38XIV (4) where Pi (cross = 1): the probability of the ith vehicle crossing (violating warning device); Pi (stop = 0): the probability of the ith vehicle stopping (complying to warning device); Xspeed : speed, comparing 80 km/h (1) to 60 km/h (0); Xgender : gender, comparing female (1) to male (0); Xm.age : age, comparing the age group of 31–50 years (1) to the age group of >50 years (0); Xy.age : age, comparing the age group of 17–30 years (1) to the age group of >50
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
179
Table 1 Distribution of male and female drivers. Number
Male Female
1,612,887 1,519,454
51.5 48.5
Total
3,132,341
100.0
%
>75
70-74
60-69
50-59
30-39
40-49
25-29
20
21-24
19
18
17
25% 20% 15% 10% 5% 0% 16
Percentage
Gender
Age Group Fig. 1. Distribution of driver licences by age group.
Percentage
50% 40%
Fig. 3. A roadway approaching a level crossing created in the simulation model
30% 20% 10% 0%
24% 16-29
38%
38%
30-49 Age Group
>50
Fig. 2. Distribution of driver licences by age group in simulation.
years (0); XFL : the type of devices, comparing flashing light (1) to stop sign (0); XIV : the type of devices, comparing in-vehicle warning (1) to stop sign (0). The models reflect that female drivers were less likely to violate warnings than male drivers while middle age group were more likely to violate the warning compared to the senior age group. The violation behaviour might possibly be due to drivers’ unintentional actions, or their driving style. The small variation in terms of stopping compliance between the two active systems reported is likely due to the fixed design warning activation time used in the laboratory experiments which could be further optimised in the future. As for the behavioural responses, speed affected drivers’ reaction in the early stage when approaching the level crossings (AccR and IniBr) and types of warning devices, age and gender had more influence at the final stages of stopping (FinalBr). The experimental procedure and the development of the driver behavioural models are detailed in previous publications (Tey et al., 2013a,b). There are limitations to directly adopting the behavioural models developed due to aspects such as small sample size and low model sensitivity. Nevertheless, the intention of this research was to explore the potential application of the behavioural models using drivers’ stopping compliance and reaction positions in a micro-simulation model for relative safety evaluation of alternative devices, as detailed in the following sections. 3. Contributing variables and their distribution For simulation purposes in the current paper, the distribution of gender and age for Queensland according to the Department of Transport and Main Roads (2010) was adopted. There was almost equal numbers of male and female drivers in Queensland as shown in Table 1. The distribution of driver licences in Queensland based on age is shown in Fig. 1. Fig. 2 shows the distribution
of driver licences adopted for simulation purposes in the current paper, based on the age group categories in the driving simulator experiment. Two speeds were adopted in the driving simulator experiment, 60 km/h or 80 km/h. 4. Development of simulation model MATLAB 7.11.0 was used for the simulation of driver behaviour. The model was refined to include appropriate driver–vehicle categories such as gender and age, probability to stop at level crossing, foot–pedal activation positions approaching level crossing prior to stopping at stop line, and a range of deceleration rates. A direct link (roadway) approaching a level crossing was created as shown in the schematic diagram (Fig. 3). A simulation of 365 days with two peaks/day was applied. Trains were assigned based on: (1) uniform distribution (with maximum train incoming time of 60 min); and (2) normal distribution (with mean train incoming time of 30 min). Both distributions resulted in a train incoming time of 30 min. A negative exponential distribution was used for generating the inter-arrival time of vehicles (vehicle entry headways) to the simulation model. The simulation began with ten (10) trains/day and eight (8) vehicles/h. There were two parts of the analysis conducted: collision likelihood and TTC. A ‘collision likelihood’ (potential number of conflicts) was recorded when a vehicle entered the ‘hazardous zone’, as denoted in Fig. 3, while a train was approaching the level crossing. TTC was identified as the warning was activated or the train was within 20 s of the crossing. TTC is defined herein as the time between a vehicle arriving at the stop line and a train arriving at the crossing (time vehicle arrives at crossing – time train arrives at crossing). Discrete-event simulation was adopted. The random samples used in the Monte Carlo simulation were generated by MATLAB functions. The reliability of the random number generator approach used in the MATLAB function has been confirmed by Kroese et al. (2011). Comparable random variables defined over successive regenerative cycles are statistically independent and identically distributed (Crane and Lemoine, 1977). Each simulation run consisted of 10,000 iterations. For different incoming traffic volumes (8 and 50 vehicles/h), the mean number of vehicles with TTC less than 30 s was generated. Fig. 4 shows an example of the frequency of the occasion (TTC < 30 s) from each iteration of one run
180
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
4.1. Simulation results
Fig. 4. Frequencies of TTC < 30 s from a test run.
and the mean results are 46 and 304 for incoming traffic volume of 8 and 50 vehicles/h, respectively. It is believed that a TTC between a vehicle and a train greater than 30 s provided adequate time for a driver to react appropriately (stop) or to cross the crossing safely before the arrival of the train, thus, collision was unlikely to happen. It was assumed that a driver would not collide with a train after seeing the train passing the crossing. Vehicles with a TTC of less than 30 s were subsequently further evaluated for the numbers of potential collisions (collision likelihood) and the TTC. For each vehicle entering the system, different conditions were assigned randomly to each of the contributing variables (gender, age and speed). Driver stopping compliance and response patterns were simulated to comply with the regression models developed as given in Section 2, with the corresponding contributing variables assigned. Therefore, the simulation model in the current paper incorporated human factors (driver characteristics of age and gender) and driver behaviour (stopping compliance and foot–pedal activations) into traffic microsimulation by applying the driver behavioural models developed earlier. A degree of variance between individual drivers is introduced by including the driver behavioural models and the inclusion of distributions that can be sampled by randomisation functions as verified in Fig. 5. After the surrogate measures were obtained from the simulation model, the alternatives devices were evaluated for their relative safety performance. The collision likelihood and TTC distribution among the four types of warning devices were compared.
Fig. 5. Verification of random number that controls the input of gender, speed, age and device.
4.1.1. Collision likelihood The stop-or-go decision data can be represented by a binary logistic regression function that can be used to estimate the probability of the driver stopping given his/her age, gender and the warning device. Since there is a high possibility of a conflict if the vehicles do not comply with the warning, the total number of vehicles entered the ‘hazardous zone’, as denoted in Fig. 3, as the train approached was recorded as potential conflicts (collision likelihood). The simulation began with ten (10) trains/day and eight (8) vehicles/h, then increased to fifty (50) vehicles/h. The simulation was regenerated for 15 cycles to obtain the average number of potential conflicts. The number of likely conflicts was compared for the four warning devices: flashing lights, in-vehicle audio warning, rumble strips and stop sign. Table 2 summarises the results based on the number of vehicles. When the traffic flow is 8 vehicles/h, no potential conflicts were recorded for the flashing lights and one potential conflict for the invehicle audio warning; whereas 12 potential conflicts were found at the stop sign and rumble strips for vehicles travelling at both speeds. When the traffic increased to 50 vehicles/h, the potential conflicts increased to greater than 600% for vehicles travelling at both speeds. As expected, the increment of traffic volume resulted in higher numbers of potential conflicts. In relative terms, the potential conflicts at crossings with active systems (flashing lights and in-vehicle audio warning) were much less than at crossings with passive devices (stop sign and rumble strips). The minor conflict effect is only true as models were developed using results from the simulator experiment where very little or no violation was compiled for active devices. If adjustment from field results (Tey et al., 2011) was considered, the results of collision likelihood in the traffic simulation reported would have increased by 26–33% for passive devices and 27–30% for active devices. The calibration can be done when a larger sample size from both the field and experimental results are available. Nevertheless, the simulation results show the potential of integrating the driving simulator and traffic microsimulation for relative safety assessment of various warning devices, or even for absolute safety evaluation when using proper calibrations from field investigations. 4.1.2. Time-to-collision Drivers know that there is at least some length of time before a train’s arrival at a crossing. However, they have been found unable to accurately judge the speed and distance of an oncoming train (Cohn and Nguyen, 2003; Cooper and Ragland, 2008) and their responses following activation of a warning device are highly variable, which produces varying TTC at level crossings. Trains are not able to stop promptly due to their high momentum. Therefore, TTC can be a surrogate measure to observe the potential of a collision. The severity of a collision may be established in the future with further investigation on the relationship between the TTC, collision and the severity of a collision. A range of approximate deceleration rates shown in Table 3 was adopted in order to stop the vehicles at the distance-to-stop line of not greater than 2 m. These values of deceleration rate used in the simulation might not reflect the real life situation on the road. Nevertheless, for the purpose of presenting the approach showing effect of TTC solely from the driver responses, fixed values were adopted. Table 3 also presents the simulation results of the average TTC of vehicles travelling at 80 km/h with respect to the four warning devices. An analysis of the TTC distribution revealed that the average TTC were greater than 5 s for the active devices while both passive devices resulted in TTC close to 5 s. The results of the TTC would be varied once the deceleration rates were changed. This issue can be
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
181
Table 2 Number of potential conflicts. Seed
Traffic volume = 50 vehicles/h Train volume = 10 trains/day
Traffic volume = 8 vehicles/h Train volume = 10 trains/day Flashing lights
In-vehicle audio warning
Stop signs
Rumble strips
Flashing lights
In-vehicle audio warning
Stop signs
Rumble strips
Run 1 Run 2 Run 3 Run 4 Run 5 Run 6 Run 7 Run 8 Run 9 Run 10 Run 11 Run 12 Run 13 Run 14 Run 15
0 0 1 0 1 1 1 1 0 0 1 0 1 1 1
2 2 1 2 2 2 2 2 2 2 2 1 2 2 1
14 9 12 12 14 13 13 12 12 14 13 13 11 14 12
13 11 14 11 10 13 11 14 14 14 12 13 12 15 13
6 7 6 6 6 6 6 6 6 6 7 6 7 7 6
14 12 13 12 14 12 13 14 13 12 15 14 14 12 13
87 87 86 90 86 86 87 86 88 91 85 90 86 88 89
88 85 89 87 88 86 85 85 85 82 85 90 90 86 92
Number of potential conflicts
0.6
1.8
12.5
12.7
6.3
13.1
87.5
86.9
Table 3 Simulation results of time-to-collision. Flashing light
In-vehicle audio warning
Approaching speed (km/h)
Rumble strips
Stop sign
80
Deceleration rate (m/s2 ) From 200 m to AccR From AccR to IniBr From IniBr to FinalBr
0.5 2.5 3.0
0.5 2.5 3.0
0.5 2.35 2.0
0.5 2.35 2.0
Time-to-collision (s)
7.22
7.20
5.69
5.71
resolved with further research and site investigation to include the deceleration rate as one of the variables of the model. Nevertheless, as the main aim of this research was to apply the concept of incorporating human factors (driver behaviour) and surrogate safety measures in traffic simulation for safety evaluation, the application of deceleration rates were simplified as presented. The increase of average TTC from approximately ‘5 s’ to ‘7 s’ when the active devices (flashing lights or in-vehicle audio warning) are in operation, supports the safety improvement tendency of the active systems over the passive devices. This finding is consistent with real-world situations in field investigations (Tey et al., 2011). The two alternatives, in-vehicle audio warning and rumble strips, provided comparable results to the conventional flashing light and stop sign, respectively. Thus, in-depth research into the two alternatives is recommended for cost effectiveness. 5. Summary and discussion Historically, the safety of new and innovative traffic treatments has been difficult to assess since they are not yet widely used and direct safety measures such as accident and fatality frequencies cannot be obtained in many cases. Since empirical collection of accident data is not yet an option, other methods for safety assessment are needed. To this end, microscopic simulation is a useful approach. With the application of this tool a variety of traffic safety indicators can be adopted, such as collision likelihood and TTC distribution. Also, a comparison of TTC can be made to evaluate safety changes. Nevertheless, absolute safety effects are hard to derive with such comparative analyses unless adequate field studies are conducted. The presented safety indicators are especially (but not
exclusively) suited for application in microscopic simulation studies of traffic, in which much more information is available about individual vehicle movements than is usually the case in empirical studies. Traffic microsimulation models offer many benefits in safety analysis techniques. There is, however, a considerable need to develop robust procedures and models through direct comparison, calibration, and validation of the models with real-world data. More research into other potential variables, such as familiarity at level crossing, risk taking behaviour, differences between urban and rural settings and environmental complexity, would also worth investigated. Despite these limitations, the suggested approach does allow for the following conclusions: • The number of likely collisions for different warning devices has been estimated by incorporating experimental results of drivers’ responses in the model; • The difference found in terms of the number of likely collisions at crossings within either active or passive devices is itself relatively small; • The number of likely collisions at stop signs and rumble strips are significantly higher than for flashing lights and in-vehicle audio warning systems; • Evaluation of alternative level crossing safety devices can be conducted using traffic microsimulation with some modifications; • Evaluation on alternative level crossing safety devices can include human factors (driver behaviour); • The outcome can be better established with adjustments using adequate actual field violations and reaction data; • This tool will assist in determining which safety approach is more cost-effective (best value for capital outlay) for specified traffic and train volumes. The approach presented here is able to model the drivers’ behaviour towards alternative devices at the approach to the crossing. The use of micro-simulation modelling has the advantage of enabling the effectiveness of alternative treatment candidates to be assessed before expensive trials are implemented. The effect of the crossing on the nearby road network, including signalised intersections, can also be studied subsequently. The use of microsimulation models may be able to reduce the need for expensive driver simulator experiments when evaluating driver behaviour in the future.
182
L.-S. Tey et al. / Accident Analysis and Prevention 71 (2014) 177–182
Acknowledgments The authors would like to thank Queensland Rail and the CRC for Rail Innovation for their support. Li-Sian Tey acknowledges financial assistance from a Malaysia Higher Education Scholarship and the Universiti Teknologi MARA. Guy Wallis was funded by an Australian Research Council Future Fellowship. References Abraham, J., Datta, T.K., Datta, S., 1998. Driver behaviour at rail-highway crossings. Transport. Res. Rec.: J. Transport. Res. Board 1648, 28–34. Australian Transport Council, 2003. National Railway Level Crossing Safety Strategy. Australian Transport Council, Canberra. Caird, J.K., Creaser, J.I., Edwards, C.J., Dewar, R.E., 2002. A Human Factors Analysis of Highway–Railway Grade Crossing Accidents in Canada. University of Calgary, Alberta. Chartier, P., 2000. Level crossing safety in countries of the Asia-Pacific region: comparative analysis – problems identified – possible remedial actions. In: 22nd ASEAN Railway General Managers’ Conference, Yangon, Myanmar. Cohn, T., Nguyen, T., 2003. Sensory cause of railroad grade-crossing collisions: test of the leibowitz hypothesis. Transport. Res. Rec.: J. Transport. Res. Board 1843, 24–30. Cooper, D.L., Ragland, D.R., 2008. Addressing Inappropriate Driver Behaviour at RailHighway Crossings. UC Berkeley Traffic Safety Center, Institute of Transportation Studies. Crane, A., Lemoine, A.J., 1977. An Introduction to the Regenerative Method for Simulation Analysis. Springer-Verlag, New York. Cunto, F., Saccomanno, F., 2008. Calibration and validation of simulated vehicle safety performance at signalized intersections. Accid. Anal. Prev. 40, 1171–1179. Department of Transport and Main Roads, 2010. Queensland Current Driver Licences by age group as at 30 June 1994 to 2010. 2008–2010. Queensland Government. Hayward, J.C., 1972. Near-miss determination through use of a scale of danger. Highw. Res. Rec. 384, 24–34. Hirst, S., Graham, R., 1997. The format and presentation of collision warnings. In: Noy, Y.I. (Ed.), Ergonomics and Safety of Intelligent Driver Interfaces. Lawrence Erlbaum Associates Inc., New Jersey, United States of America.
Hogema, J.H., Janssen, W.H., Coemet, M., 1996. Effects of Intelligent Cruise Control on Driving Behaviour: A Simulator Study., pp. 2081. Hyden, C., 1996. Traffic conflicts technique: state-of-the-art. In: Topp, H.H. (Ed.), Traffic Safety Work with Video Processing. Transportation Department, University Kaiserslautern. Kroese, D.P., Taimre, T., Botev, Z.I., 2011. Handbook of Monte Carlo Methods. John Wiley and Sons Inc., New Jersey. Lundgren, J., Tapani, A., 2006. Evaluation of safety effects of driver assistance systems through traffic simulation. Transport. Res. Rec.: J. Transport. Res. Board 1953, 81–88. Ozbay, K., Yang, H., Bartin, B., Mudigonda, S., 2008. Derivation and validation of new simulation-based surrogate safety measure. Transport. Res. Rec.: J. Transport. Res. Board 2083, 105–113. Piao, J., Mcdonald, M., 2008. Year safety impacts of variable speed limits – a simulation study. In: 11th International IEEE Conference on Intelligent Transportation Systems, 10–12 December 2008. Institute of Electrical and Electronics Engineers Inc., Beijing, China, pp. 833–837. Schoppert, D.W., Hoyt, D.W., 1968. Factors Influencing Safety at HighwayRailway Grade Crossings. Washington: Transportation Research Board. National Research Council. Tey, L.S., Ferreira, L., Wallace, A., 2011. Measuring driver responses at railway level crossings. Accid. Anal. Prev. 43, 2134–2141. Tey, L.S., Wallis, G., Cloete, S., Ferreira, L., 2013a. Modelling driver behaviour towards innovative warning devices at railway level crossings. Accid. Anal. Prev. 51, 104–111. Tey, L.S., Wallis, G., Cloete, S., Ferreira, L., Zhu, S., 2013b. Evaluating driver behaviour towards innovative warning devices at railway level crossings using a driving simulator. J. Transport. Saf. Security 5, 118–130. Van Der Horst, A.R.A., 1990. A Time-based Analysis of Road User Behaviour in Normal and Critical Encounters. Technische Universiteit Delft, Delft, Netherlands. Van Der Horst, A.R.A., Brown, G.R., 1989. Time-to-Collection and Driver Decisionmaking in Braking. TNO Institute for Perception, Soesterberg. Van Der Horst, R., 1991. Time-to-collision as a cue for decision making in braking. Vision Vehicles 3, 19–26. Wallace, A., Mccusker, L., Hirsch, L., 2008. Level Crossings Research Database. CRC for Rail Innovation, Brisbane, Australia. Yeh, M., Multer, J., 2007. Year applying a sociotechnical framework for improving safety at highway-railroad grade crossings. In: Human Factors and Ergonomics Society 51st Annual Meeting, 1–5 October 2007, Baltimore, pp. 1550–1554.
