This article was downloaded by: [USC University of Southern California] On: 13 March 2015, At: 05:38 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Effect of Audio In-vehicle Red Light–Running Warning Message on Driving Behavior Based on a Driving Simulator Experiment ab
ab
Xuedong Yan , Yang Liu
ab
& Yongcun Xu
a
MOE Key Laboratory for Urban Transportation Complex Systems Theory and Technology, School of Traffic and Transportation, Beijing Jiaotong University, Beijing, P. R. China b
Center of Cooperative Innovation for Beijing Metropolitan Transportation, Beijing, P. R. China Accepted author version posted online: 03 Apr 2014.Published online: 26 Sep 2014.
Click for updates To cite this article: Xuedong Yan, Yang Liu & Yongcun Xu (2015) Effect of Audio In-vehicle Red Light–Running Warning Message on Driving Behavior Based on a Driving Simulator Experiment, Traffic Injury Prevention, 16:1, 48-54, DOI: 10.1080/15389588.2014.906038 To link to this article: http://dx.doi.org/10.1080/15389588.2014.906038
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2015) 16, 48–54 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2014.906038
Effect of Audio In-vehicle Red Light–Running Warning Message on Driving Behavior Based on a Driving Simulator Experiment XUEDONG YAN1,2, YANG LIU1,2, and YONGCUN XU1,2
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
1
MOE Key Laboratory for Urban Transportation Complex Systems Theory and Technology, School of Traffic and Transportation, Beijing Jiaotong University, Beijing, P. R. China 2 Center of Cooperative Innovation for Beijing Metropolitan Transportation, Beijing, P. R. China Received 8 December 2013, Accepted 16 March 2014
Objective: Drivers’ incorrect decisions of crossing signalized intersections at the onset of the yellow change may lead to red light running (RLR), and RLR crashes result in substantial numbers of severe injuries and property damage. In recent years, some Intelligent Transport System (ITS) concepts have focused on reducing RLR by alerting drivers that they are about to violate the signal. The objective of this study is to conduct an experimental investigation on the effectiveness of the red light violation warning system using a voice message. Methods: In this study, the prototype concept of the RLR audio warning system was modeled and tested in a high-fidelity driving simulator. According to the concept, when a vehicle is approaching an intersection at the onset of yellow and the time to the intersection is longer than the yellow interval, the in-vehicle warning system can activate the following audio message “The red light is impending. Please decelerate!” The intent of the warning design is to encourage drivers who cannot clear an intersection during the yellow change interval to stop at the intersection. Results: The experimental results showed that the warning message could decrease red light running violations by 84.3 percent. Based on the logistic regression analyses, drivers without a warning were about 86 times more likely to make go decisions at the onset of yellow and about 15 times more likely to run red lights than those with a warning. Additionally, it was found that the audio warning message could significantly reduce RLR severity because the RLR drivers’ red-entry times without a warning were longer than those with a warning. Conclusions: This driving simulator study showed a promising effect of the audio in-vehicle warning message on reducing RLR violations and crashes. It is worthwhile to further develop the proposed technology in field applications. Keywords: signalized intersection, red light running, yellow signal change, audio warning message, driving simulator
Introduction Red light running (RLR) contributes to substantial numbers of motor vehicle crashes and injuries internationally. Red light runners can be classified into 2 categories, intentional violators and unintentional violators (Bonneson et al. 2001). RLR cameras have been widely used as effective deterrence tools for intentional violators and can significantly lower the RLR rate (Retting and Kyrychenko 2001). However, there could be a
Associate Editor Clay Gabler oversaw the review of this article Address correspondence to Xuedong Yan, MOE Key Laboratory for Urban Transportation Complex Systems Theory and Technology, School of Traffic and Transportation, Beijing Jiaotong University, Beijing 100044, P. R. China. E-mail: [email protected] Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/gcpi.
connection between red light cameras and rear-end accidents (Council et al. 2005; Quiroga et al. 2003). In order to improve drivers’ stop–go decisions when encountering a signal change, several engineering methods were introduced to reduce unintentional violators, including pavement markings, advance warning signs, countdown timers, flashing green, etc. The pavement marking method is to place a marking with a word message “Signal Ahead” upstream of a signalized intersection based on stop sight distance to the intersection (U.S. Department of Transportation 2003). Because the marking design provides drivers with a reference to judge whether they should stop at intersections or proceed when encountering the yellow signal change, the marking can reduce RLR rate to some extent according to both driving simulator experiments (Yan et al. 2009) and field observations (Elmitiny et al. 2010). Similarly, advance warning signs or flashers (AWF) can be placed upstream of high-speed signalized intersections to improvE drivers’ stop–go decisions (Messer et al. 2003). Some research results showed that intersections with AWFs appear to
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
In-vehicle Red Light–Running Warning have lower left-turn, right-angle, and, in some instances, rearend accidents (Agent and Pigman 1994; Eck and Sabra 1985; Gibby et al. 1992; Klugman et al. 1992; Sayed et al. 1999). The others indicated that AWFs may have a negative effect on driving behavior (Pant and Huang 1992; Pant and Xie 1995; Smith and Harney 2001) because aggressive drivers would increase their speeds at intersection approaches with AWFs in an attempt to beat the red signal at the end of green phase. Pre-yellow signal indication systems (PSIS) utilize a flashing green or yellow signal during the last few seconds of the green phase to warn drivers. PSIS was found to be associated with an increase in rear-end crashes and negligible changes in right-angle collisions (Quiroga et al. 2003) and perform comparably to an increased yellow change interval (Newton ¨ et al. et al. 1997). However, according to a field study (Koll 2004), PSIS contributed to increasing early stops and reduced likelihood of right-angle collisions. A green-phase countdown timer can show many seconds left until the termination of green phase and serve as a warning for drivers to avoid RLR. Some field studies indicated that countdown timers could reduce the number of red light–running violations (Kidwai et al. 2005; Lum and Halim 2006). The others showed that countdown timers had negative effects on intersection safety (Chen et al. 2007) because the dilemma zone related to signal change was lengthened (Chiou and Chang 2010; Puan and Ismail 2010) and countdown timers may cause drivers to beat the red signal to avoid a delay when stopping at the intersection (Long et al. 2011). Though traditional engineering methods cannot completely resolve the RLR problem, intelligent intersection systems provide a vista for enhancing traffic safety. The developing systems include American Cooperative Intersection Collision Avoidance Systems (U.S. Department of Transportation 2007) and European INTERSAFE Systems (ECIS 2009). Currently, one major research task of both systems is providing advisory/warning messages to drivers who are about to run a red light because of failure to judge the adequate time to safely clear a signalized intersection or recognize the presence and status of the traffic signal. The initial analysis suggested that the in-vehicle warning message is helpful to lower RLR violations and drivers have better performances under warning conditions (Feenstra and Pauwelussen 2009). However, the Intelligent Transport System (ITS) concepts and technologies for reducing RLR violations are still under the stage of research and deployment. Before identification and development of mature vehicle-based and infrastructurebased technologies that will be applied to intersections, driving simulators can serve as a useful tool to test the effectiveness of the systems on driver behavior and intersection safety. The motivation of this proposed simulator study is to conduct a foresight investigation on the effectiveness of the red light violation warning system using in-vehicle voice message. The specific objects of this study are to (1) assess the effect of an audio in-vehicle warning message on reduction in RLR violation; (2) evaluate the potential impacts of the warning message on driver behavior at intersections, such as stop–go decisions, brake response time, and stopping deceleration rate; and (3) investigate the influence of driver vocation and gender on RLR violation and effectiveness of the warning message.
49 Methodology Subjects The experiment was a 2 (Vocation) × 2 (Gender) × 2 (Warning) × 5 (Time to intersection at the onset of yellow) withinsubject repeated measures design. A total of 49 paid participants in 2 vocation groups, 23 professional taxi drivers (14 males vs. 9 females) and 26 nonprofessional drivers (13 males vs. 13 females), were tested for this research. Subjects’ average age was 33.6, ranging from 20 to 52 years old, with a standard deviation of 9.9 years. For males the average age was 33.7 with a standard deviation of 8.9 years; for females the average age was 33.4 with a standard deviation of 11.5 years. The subjects’ average self-reported annual driving mileage was 40,755 km, with a standard deviation of 34,809 km. The taxi drivers’ average annual driving millage was significantly higher than the nonprofessional drivers (M = 72,609 km, SD = 24,349 km vs. M = 12,577 km, SD = 6,736 km). Every participant held a valid Beijing’s driver’s license with at least one year of driving experience. The experiment lasted for about 30 min in total and each participant was compensated 500 Chinese RMB (about $US80). Apparatus/Equipment This study used the Beijing Jiaotong University (BJTU) driving simulator as a tool for the experiment and data collection. The BJTU simulator is a high-performance, high-fidelity driving simulator with a linear motion base capable of operation with 1 degree of freedom. It includes 8 channels (5 forward views and 3 rearview mirrors) of image generation and the simulated environment is projected at 300◦ field of view and at a resolution of 1400 × 1050 pixels. The BJTU simulator is composed of full-size vehicle cabin (Ford Focus) with a real operation interface, environmental noise and shaking simulation system, digital video replay system, and vehicle dynamic simulation system. The data sampling frequency is 60 Hz. Scenario Design The prototype concept of the red light violation audio warning system was modeled and tested in this driving simulator study. The concept is based on the assumptions that intersections’ signal status data are available in a real-time manner and can be accurately sent to the approaching vehicle through infrastructure–vehicle communication systems. In this study, the in-vehicle warning system is designed for individual vehicle drivers. Only when they are located upstream of intersection and their time to intersection is equal to or longer than yellow interval at the onset of yellow will the warning message “The red light is impending. Please decelerate!” be activated to advise the driver to stop at the intersection. Compared to the current practical countermeasures, the warning system is more pertinent to the potential RLR drivers. As illustrated in Figure 1, each subject needs to complete 2 driving scenarios, A and B, to test the effectiveness of the audio warning message on RLR violation. The 2 scenarios in the same urban environment road network composed of
50
Yan et al. 10 min to become familiar with the driving simulator operation. After a 10-min rest, the subjects performed the formal experiment with the 2 driving scenarios in a random sequence to eliminate the experimental order effect.
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
Dependent Measures In the simulator experiment, data collection and analysis were based on the decisions made by the 49 subjects during the signal change at test intersections. Each subject responded to 5 intersection approaches with a warning and 5 regular intersection approaches without a warning for a total of 490 driver–intersection encounters. The dependent measures for data analysis are defined as follows:
Fig. 1. Scenario design for the driving simulation experiment: (a) Scenario A and (b) Scenario B.
a series of signalized intersections and 2-lane road segments were designed to test driving behaviors related to yellow signal change. Each interval between adjacent intersections is 400 m and the speed limit on the road network is 80 km/h. The yellow signal length was designed as 4.5 s based on the recommendation of Institute of Transportation Engineers (1999). Five yellow-onset gaps defined as the time to intersection stop line (TTI) at the onset of yellow were designed for the test intersections, ranging from 4.5 to 6.5 m with a 0.5-s increment. In order to counterbalance the temporal/spatial order effect during the experiment operation, the yellow-onset gaps to intersection stop lines were randomly assigned to 10 test intersections. Subjects would encounter yellow signal change at the test intersections with or without the audio warning message paired in terms of yellow-onset gaps. In addition to the test intersections that displayed signal change, subjects drove through additional signalized intersections that always displayed a continuous green phase in order to keep subjects from expecting a signal change at each intersection. Experiment Procedure Upon arrival, the subjects were asked to fill out and sign an informed consent form (per Institutional Review Board requirements). The subjects were then advised to drive and behave as they normally would and to adhere to traffic laws as in real-life situations. The subjects were also notified that they could quit the experiment at any time in case of simulator sickness or any kind of discomfort. (During the test, one female subject did not finish the experiment due to simulator sickness.) Prior to the formal experiment, subjects were trained for at least
• SPEED (km/h): The vehicle’s operating speed at the onset of the yellow. • ST GO (stop = 0; go = 1): Whether the driver stopped or went through the intersection. • RLR (No = 0; Yes = 1): Whether the driver ran a red light or not. An RLR violation was defined as an event that the vehicle was going through an intersection and was located upstream of the stop line at the onset of red signal. • BRT (s): Brake response time to yellow, which was measured as the time elapsed from the onset of the yellow until the driver started stepping on the brake pedal, if the driver decided to stop at the intersection. • DEC (m/s2): The deceleration rate, which was measured as the speed reduction rate from the speed of the vehicle at the onset of the yellow phase until the speed decreased to 8 km/h (5 mph). Zero kilometers per hour was not used because drivers might maintain a crawling speed until they reached the stop bar, which would bias the experimental results (Yan et al. 2009). • RET (s): Red-entry time to the intersection, which was measured as the time difference from the moment of red signal onset to the moment when the vehicle reached the stop line. The measurement was designed to reflect the RLR severity level.
Experiment Results and Discussion Speed at the Onset of Yellow For all of the subjects, the mean speed at the onset of yellow was 72.74 km/h with a standard deviation of 15.28 km/h, indicating that most of the drivers were approaching the intersections under the speed limit. The analysis of variance result for speed indicates that there was no significant difference between male and female drivers (M = 72.69 km/h, SD = 12.78 km/h vs. M = 72.81 km/h, SD = 17.92 km/h; P = .684) and between with and without a warning (M = 73.06 km/h, SD = 14.68 km/h vs. M = 72.43 km/h, SD = 15.89 km/h; P = .640). In addition, treating TTI as a covariate in the analysis of variance, speed was not significantly associated with TTI (P = .390). However, it was found that the average speed for the nonprofessional drivers was significantly higher than that for taxi drivers (M = 75.61 km/h, SD = 15.43 km/h vs. M = 69.51 km/h, SD = 14.48 km/h; P < .001). A
In-vehicle Red Light–Running Warning
51
Table 1. Descriptive statistics for stop–go decision and RLR by independent factors Stop–go Level
Go
Stop
Yes
No
Gender
Male
76 28.1% 79 35.9% 98 42.6% 57 21.9% 46 46.9% 39 39.8% 23 23.5% 21 21.4% 26 26.5% 135 55.1% 20 8.2% 155 31.6%
194 71.9% 141 64.1% 132 57.4% 203 78.1% 52 53.1% 59 60.2% 75 76.5% 77 78.6% 72 73.5% 110 44.9% 225 91.8% 335 68.4%
69 25.6% 71 32.3% 89 38.7% 51 19.6% 31 31.6% 39 39.8% 23 23.5% 21 21.4% 26 26.5% 121 49.4% 19 7.8% 140 28.6%
201 74.4% 149 67.7% 141 61.3% 209 80.4% 67 68.4% 59 60.2% 75 76.5% 77 78.6% 72 73.5% 124 50.6% 226 92.2% 350 71.4%
Female Vocation
Taxi driver Non-taxi driver
TTI
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
RLR
Factor
4.5 5 5.5 6 6.5
Warning
Without With Total
possible explanation is that due to the effectiveness of the local speed enforcement countermeasure with regard to public vehicle management, the taxi drivers are less likely to speed than nonprofessional drivers. Stop–Go Decision and RLR Violation Analysis The basic statistical descriptions of stop–go decisions and RLR are summarized in Table 1. There were a 31.6% of drivers who went through the intersection when encountering yellow lights and the RLR rate was 28.6%. Compared to female drivers, male drivers had a lower go decision rate (28.1% vs. 35.9%) and RLR rate (25.6% vs. 32.3%). Compared to taxi drivers, nonprofessional drivers had a lower go decision rate (42.6% vs. 21.9%) and RLR rate (38.7% vs. 19.6%). There was a trend that the go decision and RLR rates decreased as TTI increased. Because the yellow-onset TTIs were equal to or greater than the yellow interval, most drivers who made go decisions ran red lights unless they accelerated during the yellow. Therefore, the stop rate (46.9%) at 4.5 s TTI was greater than the RLR rate (31.6%), and the stop rates at 5, 5.5, 6, and 6.5 s were the same as the RLR rates. Interestingly, the go decision and RLR rates of drivers without warning were remarkably higher than those with warning (go decision: 55.1% vs. 8.2%; RLR: 49.4% vs. 7.8%). Further, the logistic regression analyses for stop–go decisions and RLR were applied to investigate differences between factors, as shown in Table 2. The main effect models indicate that the independent variables of gender, vocation, TTI, and warning were significantly associated with both stop–go decision and RLR behavior of drivers based on a .05 significance level. For the effect of gender, the odds ratios estimators
in stop–go decision and RLR models were 0.473 and 0.540, respectively, which means that male drivers were 52.7% less likely to make go decisions at the onset of yellow and 46.0% less likely to result in RLR than female drivers. Compared to the nonprofessional drivers, taxi drivers were 4.787 times and 3.833 times more likely to make go decisions and run red lights. Treating TTI as a continuous variable in the main effect model of stop–go decisions, with a 1-s increment of TTI, the possibility that drivers would make go decisions decreased by 58.0%. This trend is consistent with previous relevant studies ¨ et al. 2003; Long et al. 2011; Papaioannou 2007; Yan (Koll et al. 2009). Similarly, the drivers’ RLR possibility decreased by 32.7% with a 1-s increment of TTI; the warning had a remarkably positive effect on drivers’ stop decisions and RLR violations. Compared to with a warning, drivers without a warning were about 22.0 times more likely to make go decisions at the onset of yellow and about 14.7 times more likely to run red lights. Therefore, the audio warning message can contribute to dramatically decreasing RLR violations. However, from Table 1, the go decision rate and RLR rate without a warning display a decreasing pattern in terms of TTI, whereas they are randomly distributed in the case of with awarning. The results imply that the warning message would be effective for reducing unintentional RLR runners but not intentional violators, who are more like to be influenced by enforcement tools; for example, RLR cameras. In order to further capture interaction effects between the independent variables, the full model analyses for both stop–go decisions and RLR were conducted based on the backward regression method (only the significant interaction variables were kept in the final models), as shown in Table 2. It was found that there were significant interaction effects between gender and vocation, warning and vocation, and warning and TTI. The interaction effect between warning and TTI shows that without a warning, the trend in which the likelihoods of go decisions and RLR decrease as TTI increases is significantly stronger than that with a warning. The finding indicates that under the warning condition, drivers make decisions relying on the warning system more than their own judgment. Interestingly, the interaction effect between vocation and gender shows that the descending order of RLR likelihood is female taxi drivers (odds = 2.500), male taxi drivers (odds = 1.146), female nonprofessional drivers (odds = 0.625), and male nonprofessional drivers (odds = 0.595). However, the interaction effect of warning and vocation shows that without a warning the RLR likelihood of taxi drivers (odds = 5.714) is significantly higher than that of nonprofessional drivers (odds = 1.152), whereas with a warning the RLR likelihood of taxi drivers (odds = 0.212) is remarkably reduced and similar to that of the nonprofessional group (odds = 0.208). The result indicates that the warning system can unify the diversity among different driver groups’ RLR rates. Yellow Brake Response Time and Deceleration for Stop Decision The basic statistical descriptions for yellow brake response time (BRT) and deceleration (DEC) for drivers who made stop decision are shown in Table 3. The multivariate analysis of variance results for BRT and DEC indicate that gender
52
Yan et al.
Table 2. Parameter estimates of logistic regression models for stop–go decisions and RLR Model Stop–go main effect model
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
RLR main effect model
Stop–go full model
RLR full model
Variable Gender
Male vs. female
Vocation
Taxi driver vs. non-taxi driver Without vs. with In seconds
Warning TTI Constant Gender Vocation
Male vs. female Taxi driver vs. non-taxi driver Without vs. with In seconds
Warning TTI Constant Gender Vocation Warning TTI Gender ∗ Vocation Warning ∗ Vocation Warning ∗ TTI Constant Gender Vocation Warning TTI Gender ∗ Vocation Warning ∗ Vocation Warning ∗ TTI Constant
Male vs. female Taxi driver vs. non-taxi driver Without vs. with In seconds
Male vs. female Taxi driver vs. non-taxi driver Without vs. with In seconds
(P = .033), vocation (P = .002), and TTI (P < .001) had significant effects on BRT, and vocation (P = .004) and TTI (P = .015) had significant effects on BRT. However, there was no significant difference in either BRT (P = .524) or DEC (P = .404) between with and without a warning. It was observed that female drivers had longer BRTs than male drivers (M = 1.372 s, SD = 0.049 s vs. M = 1.247 s, SD = 0.039 s). This result is consistent with previous experimental observations that driver reaction time is faster for men than women (Lings 1991; Sivak et al. 1981; Yan et al. 2009). From Figure 2, there was a clear trend that BRT increased and DEC decreased as TTI increased. The findings are consistent with
B
SE
−0.749
0.253
1.566
Wald
df
Significance
Exp(B)
8.761
1
.003
0.473
0.264
35.176
1
.000
4.787
3.092 −0.868 1.617 −0.616 1.344
0.306 0.183 0.978 0.243 0.249
101.961 22.526 2.731 6.419 29.162
1 1 1 1 1
.000 .000 .098 .011 .000
22.016 0.420 5.036 0.540 3.833
2.686 −0.395 −0.829 −0.290 0.539
0.290 0.170 0.952 0.341 0.515
86.083 5.404 0.758 0.720 1.096
1 1 1 1 1
.000 .020 .384 .396 .295
14.677 0.673 0.437 0.749 1.714
9.615 0.056 −1.487 2.780 −1.448 −2.553 −0.055 0.666
2.298 0.334 0.598 0.647 0.416 1.878 0.333 0.526
17.511 0.028 6.178 18.476 12.124 1.847 0.028 1.600
1 1 1 1 1 1 1 1
.000 .867 .013 .000 .000 .174 .868 .206
14, 989.250 1.057 0.226 16.112 0.235 0.078 0.946 1.946
6.120 0.175 −1.418 1.981 −0.807 −3.425
2.245 0.342 0.534 0.595 0.401 1.943
7.430 0.261 7.049 11.089 4.062 3.108
1 1 1 1 1 1
.006 .610 .008 .001 .044 .078
454.939 1.191 0.242 7.252 0.446 0.033
the previous simulator study results done by Caird et al. (2007) and field observations by Gates et al. (2007). It was explained that if drivers are relatively far from intersections at the onset of the yellow, BRT actually includes some lag time during which it does not require immediate braking reaction, thus resulting in a longer BRT. In addition, the closer the drivers are to intersections, the higher the deceleration rate required t stop. Compared to taxi drivers, non-taxi drivers had longer response times (M = 1.405 s, SD = 0.037 s vs. M = 1.214 s, SD = 0.052 s) but lower deceleration rates (M = 2.924 m/s2, SD = 0.086 m/s2 vs. M = 3.331 m/s2, SD = 0.117 m/s2). By increasing braking skills, drivers can significantly reduce
Table 3. Descriptive statistical results for BRT and DEC BRT (s)
DEC (m/s/s)
Variable
Mean
SE
Mean
SE
Male Female Taxi driver Non-taxi driver 4.5 s TTI 5 s TTI 5.5 s TTI 6 s TTI 6.5 s TTI Without warning With warning Total
1.247 1.372 1.214 1.405 1.121 0.990 1.258 1.542 1.637 1.330 1.289 1.310
0.039 0.049 0.052 0.037 0.080 0.070 0.061 0.060 0.062 0.055 0.035 0.033
3.026 3.228 3.331 2.924 3.403 3.277 3.264 2.829 2.862 3.066 3.188 3.127
0.090 0.111 0.117 0.086 0.176 0.160 0.142 0.139 0.144 0.125 0.080 0.075
In-vehicle Red Light–Running Warning
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
Fig. 2. Relationships between BRT, DEC, and TTI for the stop decision drivers: (a) BRT and TTI and (b) DEC and TTI.
both stop time and distance at a high deceleration rate without loss of control (Dewar and Olson 2002). Due to better driving skills, taxi drivers are more likely to stop faster and result in a higher deceleration rate than nonprofessional drivers. Red-Entry Time to Intersection For red light runners, RET to intersection is an important measure to analyze the RLR violation severity at an intersection (Bonneson et al. 2001; Bonneson and Zimmerman 2004; Washburn and Courage 2004). The shorter the RET, the lower the RLR severity (the later the red light runner enters the intersection after the red, the less likelihood of a traffic conflict or crash with other moving vehicles). Milazzo et al. (2001) reported that most RLR crashes occurring within the first few seconds of red are related to unprotected left turns because the left turners can legally clear intersections at the onset of red in order to not block the intersections (referred to as a “sneaker” in traffic engineering terminology). The results show that the average RET without a warning is 0.124 s with a standard deviation of 0.206 s and the time with a warning is 0.083 s with a standard deviation of 0.064 s. Figure 3 clearly illustrates that the RET data do not follow a normal distribution (P < .001 according to the Kolmogorov-
53 Smirnov test). The data distribution of RET is consistent with a previous field study (Bonneson et al. 2001) that found that more than half of the red light violations occurred in the first 0.5 s of the red signal. The study also reported that drivers who run only the first 0.5 s of red have a lower likelihood of collision with conflicting vehicles because they typically have not yet begun to move into the intersection. In this study, no subject whose RET was longer than 0.2 s was observed for the warning condition. Therefore, a 0.2-s RET was applied as a threshold value to define RLR severity, and a further analysis was conducted to compare lower RLR severity (less than or equal to 0.2 s) and higher RLR severity (larger than 0.2 s) between with and without a warning. Without a warning, there were 15 drivers who resulted in higher RLR severity (vs. 106 cases of lower RLR severity); however, with a warning, there were no drivers resulting in higher RLR severity (vs. 19 cases of lower RLR severity). It seems that the audio warning message can reduce RLR severity and that red light runners tended to minimize red-entry times to lower RLR violation severity. This trend is statistically significant based on the likelihood ratio test (P = .031). It should be noted that the logistic regression or chi-square test is not proper here because there are zero observations in the higher RLR severity for drivers with a warning. In this study, the effectiveness of the in-vehicle audio RLR warning system was tested in a driving simulation experiment. According to the experimental results, the warning message can dramatically decrease RLR violations by 84.3%. This effect is significantly better than the various engineering RLR countermeasures and close to the enforcement effect of RLR cameras (above 87% reduction according to Retting et al. 2008). It was also found that the audio warning message can significantly reduce RLR severity because 12.4% RLR drivers’ RETs without a warning were longer than 0.2 s but none of the RLR drivers had red-entry times longer than 0.2 s. The experimental results indicated that as the time to intersection increased during the yellow change interval, the drivers’ brake response time increased and the corresponding deceleration rate decreased; female drivers had longer brake response times to yellow than male drivers, and non-taxi drivers had a longer response times but lower deceleration than taxi drivers. Additionally, it was found that female taxi drivers had the highest RLR likelihood and male nonprofessional drivers’ RLR likelihood was lowest. However, the warning system can unify the diversity among different drivers’ RLR rates. In summary, this driving simulator study showed a promising effect of the audio in-vehicle warning message on reducing RLR violations and crash potentials. It is worthwhile to further develop the proposed technology in field applications.
Funding
Fig. 3. Histogram comparison between red-entry time with and without a warning.
This work was financially supported by the National Natural Science Foundation (71171014), Ph.D. Programs Foundation of Ministry of Education of China (20110009110013), and Program for New Century Excellent Talents in University (NCET-11-0570).
54
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
References Agent KR, Pigman JG. Evaluation of Change Interval Treatments for Traffic Signals at High-Speed Intersections. Lexington, KY: Kentucky Transportation Center; 1994. Report KTC-94-26. Bonneson J, Brewer M, Zimmerman K. Review and Evaluation of Factors That Affect the Frequency of Red Light-Running. Washington, DC: Federal Highway Administration; 2001. Report FHWA/TX-02/ 4027-1. Bonneson J, Zimmerman K. Effect of yellow-interval timing on the frequency of red-light violations at urban intersections. Transp Res Rec. 2004;1865(1):20–27. Caird JK, Chisholm SL, Edwards CJ, Creaser JI. The effect of yellow light onset time on older and younger drivers’ perception response time (PRT) and intersection behavior. Transportation Research Part F: Traffic Psychology and Behavior. 2007;10:383–396. Chen IC, Chang KK, Chang CC, Lai CH. The Impact Evaluation of Vehicular Signal Countdown Displays. Taiwan: Institute of Transportation, Ministry of Transportation and Communications; 2007. Chiou YC, Chang CH. Driver responses to green and red vehicular signal countdown displays: safety and efficiency aspects. Accid Anal Prev. 2010;42:1057–1065. Council FM, Persaud B, Eccles K, Lyon C, Griffith MS. Safety Evaluation of Red Light Cameras. Washington, DC: Federal Highway Administration; 2005. Report FHWA-HRT-05-048/HRDS-02/04-05(500)E. Dewar RE, Olson PL. Human Factors in Traffic Safety. Tucson, AZ: Lawers & Judges Publishing Company, Inc; 2002. Eck RW, Sabra ZA. Active advance warning signs at high-speed signalized intersections: a survey of practice (abridgement). Transp Res Rec. 1985;1010:62–64. Elmitiny N, Radwan E, Yan X, Jardaneh M. Field testing of a proposed pavement marking as a red light running countermeasure. Transp J. 2010;4:79–86. European Commission Information Society (ECIS). SAFESPOT Integrated Project [EB/OL]. 2009. Available at: http://www.safespoteu.org. Feenstra P, Pauwelussen J. Driving simulator study to support the design of an intersection safety application. Paper presented at: Transportation Research Board 88th Annual Meeting (TRB); January 11–15, 2009; Washington, DC. Gates TJ, Noyce DA, Laracuente L, Nordheim EV. Analysis of driver behavior in dilemma zones at signalized intersections. Transportation Research Record. 2007;2030:29–39. Gibby AR, Washington SP, Ferrara TC. Evaluation of high-speed isolated signalized intersections in California. Transportation Research Record. 1992;1376:45–56. Institute of Transportation Engineers (ITE). Traffic Engineering Handbook. Washington, DC: Institute of Transportation Engineers; 1999. Kidwai FA, Karim MR, Ibrahim MR. Traffic flow analysis of digital countdown signalized urban intersection. Proc EASTS J. 2005;5:1301–1308. Klugman A, Boje B, Belrose M. A study of the use and operation of advanced warning flashers at signalized intersection. Report no. MN/RC-93/01, Minnesota Department of Transportation; 1992. ¨ H, Bader M, Axhausen KW. Driver behavior during flashing green Koll before amber: a comparative study. Accid Anal Prev. 2003;36:273–280. Lings S. Assessing driving capability: a method for individual testing. Appl Ergon. 1991;22(2):75–84.
Yan et al. Long K, Han L, Yang Q. Effects of countdown timers on driver behavior after the yellow onset at Chinese intersections. Traffic Inj Prev. 2011;12:538–544. Lum KM, Halim H. A before-and-after study on green signal countdown device installation. Transp Res Part F Traffic Psychol Behav. 2006;9(1):29–41. Messer CJ, Sunkari SR, Charara HA, Parker RT. Development of Advance Warning Systems for End-of-Green Phase at High Speed Traffic Signals. TX: Texas Transportation Institute, Texas A & M University System; 2003. Report 4260-4. Milazzo JS, Hummer JE, Prothe LM. A Recommended Policy for Automated Electronic Traffic Enforcement of Red Light Running Violations in North Carolina. NC: Institute for Transportation Research and Education; 2001. Newton C, Mussa NR, Sadalla KE, Burns KE, Matthias J. Evaluation of an alternative traffic light change anticipation system. Accid Anal Prev. 1997;29:201–209. Pan PD, Huang XH. Active advance warning signs at high-speed signalized intersections: results of a study in Ohio. Transp Res Rec. 1992;1368:18–26. Pant PD, Xie Y. Comparative study of advance warning signs at highspeed signalized intersections. Transp Res Rec. 1995;1495:28–35. Papaioannou P. Driver behaviour, dilemma zone and safety effects at urban signalised intersections in Greece. Accid Anal Prev. 2007;12:147–158. Puan OC, Ismail CR. Dilemma zone conflicts at isolated intersections controlled with fixed-time and vehicle actuated traffic signal systems. Int J Civ Eng. 2010;10(3):19–25. Quiroga C, Kraus E, Schalkwyk IV, Bonneson J. Red Light Running—A Policy Review. College Station, TX: Texas Transportation Institute; 2003. Report CTS-02/150206-1. Retting RA, Ferguson SA, Farmer CM. Reducing red light running through longer yellow signal timing and red light camera enforcement: results of a field investigation. Accid Anal Prev. 2008;40(1):327–333. Retting RA, Kyrychenko S. Crash Reductions Associated With Red Light Enforcement in Oxnard California. Oxnard, CA: Insurance Institute for Highway Safety; 2001. Sayed T, Vahidi H, Rodriguez F. Advanced warning flashers: do they improve safety. Transp Res Rec. 1999;1692:30–38. Sivak M, Post D, Olson P, Donohue R. Driver responses to highmounted brake lights in actual traffic. Hum Factors. 1981;23(2): 231–235. Smith TJ, Harney RT. In Effects of advanced warning flashers at signalized intersections on simulated driving performance. Paper presented in: Proceedings of the Human Factors and Ergonomics Society Annual Meeting; October 8–12, 2001; Minneapolis/St. Paul, MN. US Department of Transportation. Manual on Uniform Traffic Control Devices (MUTCD). Washington, DC: Federal Highway Administration; 2003. US Department of Transportation. Cooperative intersection collision avoidance systems [press release]. 2007. Available at: http://www.its.dot.gov/cicas. Accessed March 18, 2014. Washburn SS, Courage KG. Investigation of Red Light Running Factors. Knoxville, TN: Southeast Transportation Center, the University of Tennessee; 2004. Yan X, Radwan E, Guo D, Richards SH. Impact of “Signal Ahead” pavement marking on driver behavior at signalized intersections. Transp Res Part F Traffic Psychol Behav. 2009;12(1):50–67.
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Effect of Audio In-vehicle Red Light–Running Warning Message on Driving Behavior Based on a Driving Simulator Experiment ab
ab
Xuedong Yan , Yang Liu
ab
& Yongcun Xu
a
MOE Key Laboratory for Urban Transportation Complex Systems Theory and Technology, School of Traffic and Transportation, Beijing Jiaotong University, Beijing, P. R. China b
Center of Cooperative Innovation for Beijing Metropolitan Transportation, Beijing, P. R. China Accepted author version posted online: 03 Apr 2014.Published online: 26 Sep 2014.
Click for updates To cite this article: Xuedong Yan, Yang Liu & Yongcun Xu (2015) Effect of Audio In-vehicle Red Light–Running Warning Message on Driving Behavior Based on a Driving Simulator Experiment, Traffic Injury Prevention, 16:1, 48-54, DOI: 10.1080/15389588.2014.906038 To link to this article: http://dx.doi.org/10.1080/15389588.2014.906038
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2015) 16, 48–54 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2014.906038
Effect of Audio In-vehicle Red Light–Running Warning Message on Driving Behavior Based on a Driving Simulator Experiment XUEDONG YAN1,2, YANG LIU1,2, and YONGCUN XU1,2
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
1
MOE Key Laboratory for Urban Transportation Complex Systems Theory and Technology, School of Traffic and Transportation, Beijing Jiaotong University, Beijing, P. R. China 2 Center of Cooperative Innovation for Beijing Metropolitan Transportation, Beijing, P. R. China Received 8 December 2013, Accepted 16 March 2014
Objective: Drivers’ incorrect decisions of crossing signalized intersections at the onset of the yellow change may lead to red light running (RLR), and RLR crashes result in substantial numbers of severe injuries and property damage. In recent years, some Intelligent Transport System (ITS) concepts have focused on reducing RLR by alerting drivers that they are about to violate the signal. The objective of this study is to conduct an experimental investigation on the effectiveness of the red light violation warning system using a voice message. Methods: In this study, the prototype concept of the RLR audio warning system was modeled and tested in a high-fidelity driving simulator. According to the concept, when a vehicle is approaching an intersection at the onset of yellow and the time to the intersection is longer than the yellow interval, the in-vehicle warning system can activate the following audio message “The red light is impending. Please decelerate!” The intent of the warning design is to encourage drivers who cannot clear an intersection during the yellow change interval to stop at the intersection. Results: The experimental results showed that the warning message could decrease red light running violations by 84.3 percent. Based on the logistic regression analyses, drivers without a warning were about 86 times more likely to make go decisions at the onset of yellow and about 15 times more likely to run red lights than those with a warning. Additionally, it was found that the audio warning message could significantly reduce RLR severity because the RLR drivers’ red-entry times without a warning were longer than those with a warning. Conclusions: This driving simulator study showed a promising effect of the audio in-vehicle warning message on reducing RLR violations and crashes. It is worthwhile to further develop the proposed technology in field applications. Keywords: signalized intersection, red light running, yellow signal change, audio warning message, driving simulator
Introduction Red light running (RLR) contributes to substantial numbers of motor vehicle crashes and injuries internationally. Red light runners can be classified into 2 categories, intentional violators and unintentional violators (Bonneson et al. 2001). RLR cameras have been widely used as effective deterrence tools for intentional violators and can significantly lower the RLR rate (Retting and Kyrychenko 2001). However, there could be a
Associate Editor Clay Gabler oversaw the review of this article Address correspondence to Xuedong Yan, MOE Key Laboratory for Urban Transportation Complex Systems Theory and Technology, School of Traffic and Transportation, Beijing Jiaotong University, Beijing 100044, P. R. China. E-mail: [email protected] Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/gcpi.
connection between red light cameras and rear-end accidents (Council et al. 2005; Quiroga et al. 2003). In order to improve drivers’ stop–go decisions when encountering a signal change, several engineering methods were introduced to reduce unintentional violators, including pavement markings, advance warning signs, countdown timers, flashing green, etc. The pavement marking method is to place a marking with a word message “Signal Ahead” upstream of a signalized intersection based on stop sight distance to the intersection (U.S. Department of Transportation 2003). Because the marking design provides drivers with a reference to judge whether they should stop at intersections or proceed when encountering the yellow signal change, the marking can reduce RLR rate to some extent according to both driving simulator experiments (Yan et al. 2009) and field observations (Elmitiny et al. 2010). Similarly, advance warning signs or flashers (AWF) can be placed upstream of high-speed signalized intersections to improvE drivers’ stop–go decisions (Messer et al. 2003). Some research results showed that intersections with AWFs appear to
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
In-vehicle Red Light–Running Warning have lower left-turn, right-angle, and, in some instances, rearend accidents (Agent and Pigman 1994; Eck and Sabra 1985; Gibby et al. 1992; Klugman et al. 1992; Sayed et al. 1999). The others indicated that AWFs may have a negative effect on driving behavior (Pant and Huang 1992; Pant and Xie 1995; Smith and Harney 2001) because aggressive drivers would increase their speeds at intersection approaches with AWFs in an attempt to beat the red signal at the end of green phase. Pre-yellow signal indication systems (PSIS) utilize a flashing green or yellow signal during the last few seconds of the green phase to warn drivers. PSIS was found to be associated with an increase in rear-end crashes and negligible changes in right-angle collisions (Quiroga et al. 2003) and perform comparably to an increased yellow change interval (Newton ¨ et al. et al. 1997). However, according to a field study (Koll 2004), PSIS contributed to increasing early stops and reduced likelihood of right-angle collisions. A green-phase countdown timer can show many seconds left until the termination of green phase and serve as a warning for drivers to avoid RLR. Some field studies indicated that countdown timers could reduce the number of red light–running violations (Kidwai et al. 2005; Lum and Halim 2006). The others showed that countdown timers had negative effects on intersection safety (Chen et al. 2007) because the dilemma zone related to signal change was lengthened (Chiou and Chang 2010; Puan and Ismail 2010) and countdown timers may cause drivers to beat the red signal to avoid a delay when stopping at the intersection (Long et al. 2011). Though traditional engineering methods cannot completely resolve the RLR problem, intelligent intersection systems provide a vista for enhancing traffic safety. The developing systems include American Cooperative Intersection Collision Avoidance Systems (U.S. Department of Transportation 2007) and European INTERSAFE Systems (ECIS 2009). Currently, one major research task of both systems is providing advisory/warning messages to drivers who are about to run a red light because of failure to judge the adequate time to safely clear a signalized intersection or recognize the presence and status of the traffic signal. The initial analysis suggested that the in-vehicle warning message is helpful to lower RLR violations and drivers have better performances under warning conditions (Feenstra and Pauwelussen 2009). However, the Intelligent Transport System (ITS) concepts and technologies for reducing RLR violations are still under the stage of research and deployment. Before identification and development of mature vehicle-based and infrastructurebased technologies that will be applied to intersections, driving simulators can serve as a useful tool to test the effectiveness of the systems on driver behavior and intersection safety. The motivation of this proposed simulator study is to conduct a foresight investigation on the effectiveness of the red light violation warning system using in-vehicle voice message. The specific objects of this study are to (1) assess the effect of an audio in-vehicle warning message on reduction in RLR violation; (2) evaluate the potential impacts of the warning message on driver behavior at intersections, such as stop–go decisions, brake response time, and stopping deceleration rate; and (3) investigate the influence of driver vocation and gender on RLR violation and effectiveness of the warning message.
49 Methodology Subjects The experiment was a 2 (Vocation) × 2 (Gender) × 2 (Warning) × 5 (Time to intersection at the onset of yellow) withinsubject repeated measures design. A total of 49 paid participants in 2 vocation groups, 23 professional taxi drivers (14 males vs. 9 females) and 26 nonprofessional drivers (13 males vs. 13 females), were tested for this research. Subjects’ average age was 33.6, ranging from 20 to 52 years old, with a standard deviation of 9.9 years. For males the average age was 33.7 with a standard deviation of 8.9 years; for females the average age was 33.4 with a standard deviation of 11.5 years. The subjects’ average self-reported annual driving mileage was 40,755 km, with a standard deviation of 34,809 km. The taxi drivers’ average annual driving millage was significantly higher than the nonprofessional drivers (M = 72,609 km, SD = 24,349 km vs. M = 12,577 km, SD = 6,736 km). Every participant held a valid Beijing’s driver’s license with at least one year of driving experience. The experiment lasted for about 30 min in total and each participant was compensated 500 Chinese RMB (about $US80). Apparatus/Equipment This study used the Beijing Jiaotong University (BJTU) driving simulator as a tool for the experiment and data collection. The BJTU simulator is a high-performance, high-fidelity driving simulator with a linear motion base capable of operation with 1 degree of freedom. It includes 8 channels (5 forward views and 3 rearview mirrors) of image generation and the simulated environment is projected at 300◦ field of view and at a resolution of 1400 × 1050 pixels. The BJTU simulator is composed of full-size vehicle cabin (Ford Focus) with a real operation interface, environmental noise and shaking simulation system, digital video replay system, and vehicle dynamic simulation system. The data sampling frequency is 60 Hz. Scenario Design The prototype concept of the red light violation audio warning system was modeled and tested in this driving simulator study. The concept is based on the assumptions that intersections’ signal status data are available in a real-time manner and can be accurately sent to the approaching vehicle through infrastructure–vehicle communication systems. In this study, the in-vehicle warning system is designed for individual vehicle drivers. Only when they are located upstream of intersection and their time to intersection is equal to or longer than yellow interval at the onset of yellow will the warning message “The red light is impending. Please decelerate!” be activated to advise the driver to stop at the intersection. Compared to the current practical countermeasures, the warning system is more pertinent to the potential RLR drivers. As illustrated in Figure 1, each subject needs to complete 2 driving scenarios, A and B, to test the effectiveness of the audio warning message on RLR violation. The 2 scenarios in the same urban environment road network composed of
50
Yan et al. 10 min to become familiar with the driving simulator operation. After a 10-min rest, the subjects performed the formal experiment with the 2 driving scenarios in a random sequence to eliminate the experimental order effect.
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
Dependent Measures In the simulator experiment, data collection and analysis were based on the decisions made by the 49 subjects during the signal change at test intersections. Each subject responded to 5 intersection approaches with a warning and 5 regular intersection approaches without a warning for a total of 490 driver–intersection encounters. The dependent measures for data analysis are defined as follows:
Fig. 1. Scenario design for the driving simulation experiment: (a) Scenario A and (b) Scenario B.
a series of signalized intersections and 2-lane road segments were designed to test driving behaviors related to yellow signal change. Each interval between adjacent intersections is 400 m and the speed limit on the road network is 80 km/h. The yellow signal length was designed as 4.5 s based on the recommendation of Institute of Transportation Engineers (1999). Five yellow-onset gaps defined as the time to intersection stop line (TTI) at the onset of yellow were designed for the test intersections, ranging from 4.5 to 6.5 m with a 0.5-s increment. In order to counterbalance the temporal/spatial order effect during the experiment operation, the yellow-onset gaps to intersection stop lines were randomly assigned to 10 test intersections. Subjects would encounter yellow signal change at the test intersections with or without the audio warning message paired in terms of yellow-onset gaps. In addition to the test intersections that displayed signal change, subjects drove through additional signalized intersections that always displayed a continuous green phase in order to keep subjects from expecting a signal change at each intersection. Experiment Procedure Upon arrival, the subjects were asked to fill out and sign an informed consent form (per Institutional Review Board requirements). The subjects were then advised to drive and behave as they normally would and to adhere to traffic laws as in real-life situations. The subjects were also notified that they could quit the experiment at any time in case of simulator sickness or any kind of discomfort. (During the test, one female subject did not finish the experiment due to simulator sickness.) Prior to the formal experiment, subjects were trained for at least
• SPEED (km/h): The vehicle’s operating speed at the onset of the yellow. • ST GO (stop = 0; go = 1): Whether the driver stopped or went through the intersection. • RLR (No = 0; Yes = 1): Whether the driver ran a red light or not. An RLR violation was defined as an event that the vehicle was going through an intersection and was located upstream of the stop line at the onset of red signal. • BRT (s): Brake response time to yellow, which was measured as the time elapsed from the onset of the yellow until the driver started stepping on the brake pedal, if the driver decided to stop at the intersection. • DEC (m/s2): The deceleration rate, which was measured as the speed reduction rate from the speed of the vehicle at the onset of the yellow phase until the speed decreased to 8 km/h (5 mph). Zero kilometers per hour was not used because drivers might maintain a crawling speed until they reached the stop bar, which would bias the experimental results (Yan et al. 2009). • RET (s): Red-entry time to the intersection, which was measured as the time difference from the moment of red signal onset to the moment when the vehicle reached the stop line. The measurement was designed to reflect the RLR severity level.
Experiment Results and Discussion Speed at the Onset of Yellow For all of the subjects, the mean speed at the onset of yellow was 72.74 km/h with a standard deviation of 15.28 km/h, indicating that most of the drivers were approaching the intersections under the speed limit. The analysis of variance result for speed indicates that there was no significant difference between male and female drivers (M = 72.69 km/h, SD = 12.78 km/h vs. M = 72.81 km/h, SD = 17.92 km/h; P = .684) and between with and without a warning (M = 73.06 km/h, SD = 14.68 km/h vs. M = 72.43 km/h, SD = 15.89 km/h; P = .640). In addition, treating TTI as a covariate in the analysis of variance, speed was not significantly associated with TTI (P = .390). However, it was found that the average speed for the nonprofessional drivers was significantly higher than that for taxi drivers (M = 75.61 km/h, SD = 15.43 km/h vs. M = 69.51 km/h, SD = 14.48 km/h; P < .001). A
In-vehicle Red Light–Running Warning
51
Table 1. Descriptive statistics for stop–go decision and RLR by independent factors Stop–go Level
Go
Stop
Yes
No
Gender
Male
76 28.1% 79 35.9% 98 42.6% 57 21.9% 46 46.9% 39 39.8% 23 23.5% 21 21.4% 26 26.5% 135 55.1% 20 8.2% 155 31.6%
194 71.9% 141 64.1% 132 57.4% 203 78.1% 52 53.1% 59 60.2% 75 76.5% 77 78.6% 72 73.5% 110 44.9% 225 91.8% 335 68.4%
69 25.6% 71 32.3% 89 38.7% 51 19.6% 31 31.6% 39 39.8% 23 23.5% 21 21.4% 26 26.5% 121 49.4% 19 7.8% 140 28.6%
201 74.4% 149 67.7% 141 61.3% 209 80.4% 67 68.4% 59 60.2% 75 76.5% 77 78.6% 72 73.5% 124 50.6% 226 92.2% 350 71.4%
Female Vocation
Taxi driver Non-taxi driver
TTI
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
RLR
Factor
4.5 5 5.5 6 6.5
Warning
Without With Total
possible explanation is that due to the effectiveness of the local speed enforcement countermeasure with regard to public vehicle management, the taxi drivers are less likely to speed than nonprofessional drivers. Stop–Go Decision and RLR Violation Analysis The basic statistical descriptions of stop–go decisions and RLR are summarized in Table 1. There were a 31.6% of drivers who went through the intersection when encountering yellow lights and the RLR rate was 28.6%. Compared to female drivers, male drivers had a lower go decision rate (28.1% vs. 35.9%) and RLR rate (25.6% vs. 32.3%). Compared to taxi drivers, nonprofessional drivers had a lower go decision rate (42.6% vs. 21.9%) and RLR rate (38.7% vs. 19.6%). There was a trend that the go decision and RLR rates decreased as TTI increased. Because the yellow-onset TTIs were equal to or greater than the yellow interval, most drivers who made go decisions ran red lights unless they accelerated during the yellow. Therefore, the stop rate (46.9%) at 4.5 s TTI was greater than the RLR rate (31.6%), and the stop rates at 5, 5.5, 6, and 6.5 s were the same as the RLR rates. Interestingly, the go decision and RLR rates of drivers without warning were remarkably higher than those with warning (go decision: 55.1% vs. 8.2%; RLR: 49.4% vs. 7.8%). Further, the logistic regression analyses for stop–go decisions and RLR were applied to investigate differences between factors, as shown in Table 2. The main effect models indicate that the independent variables of gender, vocation, TTI, and warning were significantly associated with both stop–go decision and RLR behavior of drivers based on a .05 significance level. For the effect of gender, the odds ratios estimators
in stop–go decision and RLR models were 0.473 and 0.540, respectively, which means that male drivers were 52.7% less likely to make go decisions at the onset of yellow and 46.0% less likely to result in RLR than female drivers. Compared to the nonprofessional drivers, taxi drivers were 4.787 times and 3.833 times more likely to make go decisions and run red lights. Treating TTI as a continuous variable in the main effect model of stop–go decisions, with a 1-s increment of TTI, the possibility that drivers would make go decisions decreased by 58.0%. This trend is consistent with previous relevant studies ¨ et al. 2003; Long et al. 2011; Papaioannou 2007; Yan (Koll et al. 2009). Similarly, the drivers’ RLR possibility decreased by 32.7% with a 1-s increment of TTI; the warning had a remarkably positive effect on drivers’ stop decisions and RLR violations. Compared to with a warning, drivers without a warning were about 22.0 times more likely to make go decisions at the onset of yellow and about 14.7 times more likely to run red lights. Therefore, the audio warning message can contribute to dramatically decreasing RLR violations. However, from Table 1, the go decision rate and RLR rate without a warning display a decreasing pattern in terms of TTI, whereas they are randomly distributed in the case of with awarning. The results imply that the warning message would be effective for reducing unintentional RLR runners but not intentional violators, who are more like to be influenced by enforcement tools; for example, RLR cameras. In order to further capture interaction effects between the independent variables, the full model analyses for both stop–go decisions and RLR were conducted based on the backward regression method (only the significant interaction variables were kept in the final models), as shown in Table 2. It was found that there were significant interaction effects between gender and vocation, warning and vocation, and warning and TTI. The interaction effect between warning and TTI shows that without a warning, the trend in which the likelihoods of go decisions and RLR decrease as TTI increases is significantly stronger than that with a warning. The finding indicates that under the warning condition, drivers make decisions relying on the warning system more than their own judgment. Interestingly, the interaction effect between vocation and gender shows that the descending order of RLR likelihood is female taxi drivers (odds = 2.500), male taxi drivers (odds = 1.146), female nonprofessional drivers (odds = 0.625), and male nonprofessional drivers (odds = 0.595). However, the interaction effect of warning and vocation shows that without a warning the RLR likelihood of taxi drivers (odds = 5.714) is significantly higher than that of nonprofessional drivers (odds = 1.152), whereas with a warning the RLR likelihood of taxi drivers (odds = 0.212) is remarkably reduced and similar to that of the nonprofessional group (odds = 0.208). The result indicates that the warning system can unify the diversity among different driver groups’ RLR rates. Yellow Brake Response Time and Deceleration for Stop Decision The basic statistical descriptions for yellow brake response time (BRT) and deceleration (DEC) for drivers who made stop decision are shown in Table 3. The multivariate analysis of variance results for BRT and DEC indicate that gender
52
Yan et al.
Table 2. Parameter estimates of logistic regression models for stop–go decisions and RLR Model Stop–go main effect model
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
RLR main effect model
Stop–go full model
RLR full model
Variable Gender
Male vs. female
Vocation
Taxi driver vs. non-taxi driver Without vs. with In seconds
Warning TTI Constant Gender Vocation
Male vs. female Taxi driver vs. non-taxi driver Without vs. with In seconds
Warning TTI Constant Gender Vocation Warning TTI Gender ∗ Vocation Warning ∗ Vocation Warning ∗ TTI Constant Gender Vocation Warning TTI Gender ∗ Vocation Warning ∗ Vocation Warning ∗ TTI Constant
Male vs. female Taxi driver vs. non-taxi driver Without vs. with In seconds
Male vs. female Taxi driver vs. non-taxi driver Without vs. with In seconds
(P = .033), vocation (P = .002), and TTI (P < .001) had significant effects on BRT, and vocation (P = .004) and TTI (P = .015) had significant effects on BRT. However, there was no significant difference in either BRT (P = .524) or DEC (P = .404) between with and without a warning. It was observed that female drivers had longer BRTs than male drivers (M = 1.372 s, SD = 0.049 s vs. M = 1.247 s, SD = 0.039 s). This result is consistent with previous experimental observations that driver reaction time is faster for men than women (Lings 1991; Sivak et al. 1981; Yan et al. 2009). From Figure 2, there was a clear trend that BRT increased and DEC decreased as TTI increased. The findings are consistent with
B
SE
−0.749
0.253
1.566
Wald
df
Significance
Exp(B)
8.761
1
.003
0.473
0.264
35.176
1
.000
4.787
3.092 −0.868 1.617 −0.616 1.344
0.306 0.183 0.978 0.243 0.249
101.961 22.526 2.731 6.419 29.162
1 1 1 1 1
.000 .000 .098 .011 .000
22.016 0.420 5.036 0.540 3.833
2.686 −0.395 −0.829 −0.290 0.539
0.290 0.170 0.952 0.341 0.515
86.083 5.404 0.758 0.720 1.096
1 1 1 1 1
.000 .020 .384 .396 .295
14.677 0.673 0.437 0.749 1.714
9.615 0.056 −1.487 2.780 −1.448 −2.553 −0.055 0.666
2.298 0.334 0.598 0.647 0.416 1.878 0.333 0.526
17.511 0.028 6.178 18.476 12.124 1.847 0.028 1.600
1 1 1 1 1 1 1 1
.000 .867 .013 .000 .000 .174 .868 .206
14, 989.250 1.057 0.226 16.112 0.235 0.078 0.946 1.946
6.120 0.175 −1.418 1.981 −0.807 −3.425
2.245 0.342 0.534 0.595 0.401 1.943
7.430 0.261 7.049 11.089 4.062 3.108
1 1 1 1 1 1
.006 .610 .008 .001 .044 .078
454.939 1.191 0.242 7.252 0.446 0.033
the previous simulator study results done by Caird et al. (2007) and field observations by Gates et al. (2007). It was explained that if drivers are relatively far from intersections at the onset of the yellow, BRT actually includes some lag time during which it does not require immediate braking reaction, thus resulting in a longer BRT. In addition, the closer the drivers are to intersections, the higher the deceleration rate required t stop. Compared to taxi drivers, non-taxi drivers had longer response times (M = 1.405 s, SD = 0.037 s vs. M = 1.214 s, SD = 0.052 s) but lower deceleration rates (M = 2.924 m/s2, SD = 0.086 m/s2 vs. M = 3.331 m/s2, SD = 0.117 m/s2). By increasing braking skills, drivers can significantly reduce
Table 3. Descriptive statistical results for BRT and DEC BRT (s)
DEC (m/s/s)
Variable
Mean
SE
Mean
SE
Male Female Taxi driver Non-taxi driver 4.5 s TTI 5 s TTI 5.5 s TTI 6 s TTI 6.5 s TTI Without warning With warning Total
1.247 1.372 1.214 1.405 1.121 0.990 1.258 1.542 1.637 1.330 1.289 1.310
0.039 0.049 0.052 0.037 0.080 0.070 0.061 0.060 0.062 0.055 0.035 0.033
3.026 3.228 3.331 2.924 3.403 3.277 3.264 2.829 2.862 3.066 3.188 3.127
0.090 0.111 0.117 0.086 0.176 0.160 0.142 0.139 0.144 0.125 0.080 0.075
In-vehicle Red Light–Running Warning
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
Fig. 2. Relationships between BRT, DEC, and TTI for the stop decision drivers: (a) BRT and TTI and (b) DEC and TTI.
both stop time and distance at a high deceleration rate without loss of control (Dewar and Olson 2002). Due to better driving skills, taxi drivers are more likely to stop faster and result in a higher deceleration rate than nonprofessional drivers. Red-Entry Time to Intersection For red light runners, RET to intersection is an important measure to analyze the RLR violation severity at an intersection (Bonneson et al. 2001; Bonneson and Zimmerman 2004; Washburn and Courage 2004). The shorter the RET, the lower the RLR severity (the later the red light runner enters the intersection after the red, the less likelihood of a traffic conflict or crash with other moving vehicles). Milazzo et al. (2001) reported that most RLR crashes occurring within the first few seconds of red are related to unprotected left turns because the left turners can legally clear intersections at the onset of red in order to not block the intersections (referred to as a “sneaker” in traffic engineering terminology). The results show that the average RET without a warning is 0.124 s with a standard deviation of 0.206 s and the time with a warning is 0.083 s with a standard deviation of 0.064 s. Figure 3 clearly illustrates that the RET data do not follow a normal distribution (P < .001 according to the Kolmogorov-
53 Smirnov test). The data distribution of RET is consistent with a previous field study (Bonneson et al. 2001) that found that more than half of the red light violations occurred in the first 0.5 s of the red signal. The study also reported that drivers who run only the first 0.5 s of red have a lower likelihood of collision with conflicting vehicles because they typically have not yet begun to move into the intersection. In this study, no subject whose RET was longer than 0.2 s was observed for the warning condition. Therefore, a 0.2-s RET was applied as a threshold value to define RLR severity, and a further analysis was conducted to compare lower RLR severity (less than or equal to 0.2 s) and higher RLR severity (larger than 0.2 s) between with and without a warning. Without a warning, there were 15 drivers who resulted in higher RLR severity (vs. 106 cases of lower RLR severity); however, with a warning, there were no drivers resulting in higher RLR severity (vs. 19 cases of lower RLR severity). It seems that the audio warning message can reduce RLR severity and that red light runners tended to minimize red-entry times to lower RLR violation severity. This trend is statistically significant based on the likelihood ratio test (P = .031). It should be noted that the logistic regression or chi-square test is not proper here because there are zero observations in the higher RLR severity for drivers with a warning. In this study, the effectiveness of the in-vehicle audio RLR warning system was tested in a driving simulation experiment. According to the experimental results, the warning message can dramatically decrease RLR violations by 84.3%. This effect is significantly better than the various engineering RLR countermeasures and close to the enforcement effect of RLR cameras (above 87% reduction according to Retting et al. 2008). It was also found that the audio warning message can significantly reduce RLR severity because 12.4% RLR drivers’ RETs without a warning were longer than 0.2 s but none of the RLR drivers had red-entry times longer than 0.2 s. The experimental results indicated that as the time to intersection increased during the yellow change interval, the drivers’ brake response time increased and the corresponding deceleration rate decreased; female drivers had longer brake response times to yellow than male drivers, and non-taxi drivers had a longer response times but lower deceleration than taxi drivers. Additionally, it was found that female taxi drivers had the highest RLR likelihood and male nonprofessional drivers’ RLR likelihood was lowest. However, the warning system can unify the diversity among different drivers’ RLR rates. In summary, this driving simulator study showed a promising effect of the audio in-vehicle warning message on reducing RLR violations and crash potentials. It is worthwhile to further develop the proposed technology in field applications.
Funding
Fig. 3. Histogram comparison between red-entry time with and without a warning.
This work was financially supported by the National Natural Science Foundation (71171014), Ph.D. Programs Foundation of Ministry of Education of China (20110009110013), and Program for New Century Excellent Talents in University (NCET-11-0570).
54
Downloaded by [USC University of Southern California] at 05:38 13 March 2015
References Agent KR, Pigman JG. Evaluation of Change Interval Treatments for Traffic Signals at High-Speed Intersections. Lexington, KY: Kentucky Transportation Center; 1994. Report KTC-94-26. Bonneson J, Brewer M, Zimmerman K. Review and Evaluation of Factors That Affect the Frequency of Red Light-Running. Washington, DC: Federal Highway Administration; 2001. Report FHWA/TX-02/ 4027-1. Bonneson J, Zimmerman K. Effect of yellow-interval timing on the frequency of red-light violations at urban intersections. Transp Res Rec. 2004;1865(1):20–27. Caird JK, Chisholm SL, Edwards CJ, Creaser JI. The effect of yellow light onset time on older and younger drivers’ perception response time (PRT) and intersection behavior. Transportation Research Part F: Traffic Psychology and Behavior. 2007;10:383–396. Chen IC, Chang KK, Chang CC, Lai CH. The Impact Evaluation of Vehicular Signal Countdown Displays. Taiwan: Institute of Transportation, Ministry of Transportation and Communications; 2007. Chiou YC, Chang CH. Driver responses to green and red vehicular signal countdown displays: safety and efficiency aspects. Accid Anal Prev. 2010;42:1057–1065. Council FM, Persaud B, Eccles K, Lyon C, Griffith MS. Safety Evaluation of Red Light Cameras. Washington, DC: Federal Highway Administration; 2005. Report FHWA-HRT-05-048/HRDS-02/04-05(500)E. Dewar RE, Olson PL. Human Factors in Traffic Safety. Tucson, AZ: Lawers & Judges Publishing Company, Inc; 2002. Eck RW, Sabra ZA. Active advance warning signs at high-speed signalized intersections: a survey of practice (abridgement). Transp Res Rec. 1985;1010:62–64. Elmitiny N, Radwan E, Yan X, Jardaneh M. Field testing of a proposed pavement marking as a red light running countermeasure. Transp J. 2010;4:79–86. European Commission Information Society (ECIS). SAFESPOT Integrated Project [EB/OL]. 2009. Available at: http://www.safespoteu.org. Feenstra P, Pauwelussen J. Driving simulator study to support the design of an intersection safety application. Paper presented at: Transportation Research Board 88th Annual Meeting (TRB); January 11–15, 2009; Washington, DC. Gates TJ, Noyce DA, Laracuente L, Nordheim EV. Analysis of driver behavior in dilemma zones at signalized intersections. Transportation Research Record. 2007;2030:29–39. Gibby AR, Washington SP, Ferrara TC. Evaluation of high-speed isolated signalized intersections in California. Transportation Research Record. 1992;1376:45–56. Institute of Transportation Engineers (ITE). Traffic Engineering Handbook. Washington, DC: Institute of Transportation Engineers; 1999. Kidwai FA, Karim MR, Ibrahim MR. Traffic flow analysis of digital countdown signalized urban intersection. Proc EASTS J. 2005;5:1301–1308. Klugman A, Boje B, Belrose M. A study of the use and operation of advanced warning flashers at signalized intersection. Report no. MN/RC-93/01, Minnesota Department of Transportation; 1992. ¨ H, Bader M, Axhausen KW. Driver behavior during flashing green Koll before amber: a comparative study. Accid Anal Prev. 2003;36:273–280. Lings S. Assessing driving capability: a method for individual testing. Appl Ergon. 1991;22(2):75–84.
Yan et al. Long K, Han L, Yang Q. Effects of countdown timers on driver behavior after the yellow onset at Chinese intersections. Traffic Inj Prev. 2011;12:538–544. Lum KM, Halim H. A before-and-after study on green signal countdown device installation. Transp Res Part F Traffic Psychol Behav. 2006;9(1):29–41. Messer CJ, Sunkari SR, Charara HA, Parker RT. Development of Advance Warning Systems for End-of-Green Phase at High Speed Traffic Signals. TX: Texas Transportation Institute, Texas A & M University System; 2003. Report 4260-4. Milazzo JS, Hummer JE, Prothe LM. A Recommended Policy for Automated Electronic Traffic Enforcement of Red Light Running Violations in North Carolina. NC: Institute for Transportation Research and Education; 2001. Newton C, Mussa NR, Sadalla KE, Burns KE, Matthias J. Evaluation of an alternative traffic light change anticipation system. Accid Anal Prev. 1997;29:201–209. Pan PD, Huang XH. Active advance warning signs at high-speed signalized intersections: results of a study in Ohio. Transp Res Rec. 1992;1368:18–26. Pant PD, Xie Y. Comparative study of advance warning signs at highspeed signalized intersections. Transp Res Rec. 1995;1495:28–35. Papaioannou P. Driver behaviour, dilemma zone and safety effects at urban signalised intersections in Greece. Accid Anal Prev. 2007;12:147–158. Puan OC, Ismail CR. Dilemma zone conflicts at isolated intersections controlled with fixed-time and vehicle actuated traffic signal systems. Int J Civ Eng. 2010;10(3):19–25. Quiroga C, Kraus E, Schalkwyk IV, Bonneson J. Red Light Running—A Policy Review. College Station, TX: Texas Transportation Institute; 2003. Report CTS-02/150206-1. Retting RA, Ferguson SA, Farmer CM. Reducing red light running through longer yellow signal timing and red light camera enforcement: results of a field investigation. Accid Anal Prev. 2008;40(1):327–333. Retting RA, Kyrychenko S. Crash Reductions Associated With Red Light Enforcement in Oxnard California. Oxnard, CA: Insurance Institute for Highway Safety; 2001. Sayed T, Vahidi H, Rodriguez F. Advanced warning flashers: do they improve safety. Transp Res Rec. 1999;1692:30–38. Sivak M, Post D, Olson P, Donohue R. Driver responses to highmounted brake lights in actual traffic. Hum Factors. 1981;23(2): 231–235. Smith TJ, Harney RT. In Effects of advanced warning flashers at signalized intersections on simulated driving performance. Paper presented in: Proceedings of the Human Factors and Ergonomics Society Annual Meeting; October 8–12, 2001; Minneapolis/St. Paul, MN. US Department of Transportation. Manual on Uniform Traffic Control Devices (MUTCD). Washington, DC: Federal Highway Administration; 2003. US Department of Transportation. Cooperative intersection collision avoidance systems [press release]. 2007. Available at: http://www.its.dot.gov/cicas. Accessed March 18, 2014. Washburn SS, Courage KG. Investigation of Red Light Running Factors. Knoxville, TN: Southeast Transportation Center, the University of Tennessee; 2004. Yan X, Radwan E, Guo D, Richards SH. Impact of “Signal Ahead” pavement marking on driver behavior at signalized intersections. Transp Res Part F Traffic Psychol Behav. 2009;12(1):50–67.
