Safety surrogate histograms (SSH): A novel real-time safety assessment of dilemma zone related conflicts at signalized intersections.

G Model AAP 3796 No. of Pages 10

Accident Analysis and Prevention xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

Safety surrogate histograms (SSH): A novel real-time safety assessment of dilemma zone related conflicts at signalized intersections Sahar Ghanipoor Machiani a, * , Montasir Abbas b a b

Department of Civil and Environmental Engineering, Virginia Tech, 301-D Patton Hall, Blacksburg, VA 24060, United States Department of Civil and Environmental Engineering, Virginia Tech, 301-A Patton Hall, Blacksburg, VA 24060, United States

A R T I C L E I N F O

A B S T R A C T

Article history: Received 1 November 2014 Received in revised form 22 March 2015 Accepted 20 April 2015 Available online xxx

Drivers’ indecisiveness in dilemma zones (DZ) could result in crash-prone situations at signalized intersections. DZ is to the area ahead of an intersection in which drivers encounter a dilemma regarding whether to stop or proceed through the intersection when the signal turns yellow. An improper decision to stop by the leading driver, combined with the following driver deciding to go, can result in a rear-end collision, unless the following driver recognizes a collision is imminent and adjusts his or her behavior at or shortly after the onset of yellow. Considering the significance of DZ-related crashes, a comprehensive safety measure is needed to characterize the level of safety at signalized intersections. In this study, a novel safety surrogate measure was developed utilizing real-time radar field data. This new measure, called safety surrogate histogram (SSH), captures the degree and frequency of DZ-related conflicts at each intersection approach. SSH includes detailed information regarding the possibility of crashes, because it is calculated based on the vehicles conflicts. An example illustrating the application of the new methodology at two study sites in Virginia is presented and discussed, and a comparison is provided between SSH and other DZ-related safety surrogate measures mentioned in the literature. The results of the study reveal the efficacy of the SSH as complementary to existing surrogate measures. ã 2015 Elsevier Ltd. All rights reserved.

Keywords: Signalized intersections Safety surrogate histogram (SSH) Time to collision (TTC) Dilemma zone (DZ) DZ-protection algorithms

1. Introduction Rear-end crashes at signalized intersections can occur for several reasons, including drivers’ indecisiveness in dilemma zones (Li and Abbas, 2010). A dilemma zone (DZ) is the area ahead of a signalized intersection in which drivers are not sure whether it is safer to continue through the intersection or to stop at the onset of the yellow light. Variation in decisions in dilemma zone might lead to one driver stopping, while the driver to the rear continues. In such cases, drivers in the rear usually recognize the danger of an impending rear-collision, and slow down. The faster those drivers decelerate, the sooner the dangerous situation is mitigated. This “turbulence” in behavioral adjustment occurs during and/or shortly after the onset of yellow and can lead to crashes if drivers do not pay attention, or fail to recognize, the dangerous situation soon enough.

* Corresponding author. Present address: Center for Infrastructure Based Safety Systems, Virginia Tech Transportation Institute, 3500 Transportation Research Plaza, Blacksburg, VA 24061, United States. Tel.: +1 540 2007560. E-mail addresses: [email protected] (S. Ghanipoor Machiani), [email protected] (M. Abbas).

Various DZ-protection algorithms and advanced signal-control strategies have been developed utilizing different technologies such as green extension, clearance time extension, and queue clearance. The majority of these algorithms assume that the major factor in DZ protection is the number of vehicles caught in the DZ, and so the algorithms hold the green signal until the DZ is clear or nearly clear. Examples of those algorithms are D-CS (Zimmerman, 2007; Zimmerman et al., 2003; Zimmerman and Bonneson, 2005), LHOVRA (Engstrom, 1994), and MOVA (Vincent and Peirce, 1988). Other algorithms, such as the platoon identification algorithm (PIA) (Chaudhary et al., 1978), take into account additional factors, like queue clearance. Advancements in the field of DZ protection include continuous tracking and recalculation of vehicle information in real time, used in Wavetronix systems (Wavetronix, 2014). DZ-related safety measures found in the literature include the number of vehicles caught in a DZ, the number of yellow and red light violators, and stop-line encroachment (Gettman and Head, 2003). However, the main input parameter for all DZ protection systems is the number of vehicles caught in a DZ. DZ conflicts can be influenced by several factors, including geometry and visibility issues, improper signal timing and/or phasing, and drivers’ behavioral tendencies. Different mitigation

http://dx.doi.org/10.1016/j.aap.2015.04.024 0001-4575/ ã 2015 Elsevier Ltd. All rights reserved.

Please cite this article in press as: S. Ghanipoor Machiani, M. Abbas, Safety surrogate histograms (SSH): A novel real-time safety assessment of dilemma zone related conflicts at signalized intersections, Accid. Anal. Prev. (2015), http://dx.doi.org/10.1016/j.aap.2015.04.024


2

S. Ghanipoor Machiani, M. Abbas / Accident Analysis and Prevention xxx (2015) xxx–xxx

strategies could be more effective depending on which is the most influential factor. There is therefore a need for a new measure that can: Quantitatively

measure behavioral turbulence near the intersection; Provide a basis for prioritizing different sites for crash mitigation and improvement strategies; Independently evaluate the efficacy of DZ-protection algorithms; and Provide clues regarding the main factor behind the safety problem, without having to wait for crashes to occur. The objective of this research was to develop a safety measure complimentary to existing strategies and meeting the needs above. Such a measure could help traffic-safety engineers and signal practitioners identify the causes of crashes and select appropriate DZ-protection strategies addressing these causes. We define a novel safety surrogate histogram (SSH) measure to capture the degree and character of DZ-related conflicts. Our method is based on real-time field measurement of vehicle trajectory data, obtained, in this case, from the commercial Wavetronix system radar, and a histogram plot of time-to-collision frequencies shortly before, during, and shortly after the change intervals at signalized intersections. The SSH provides a better picture of the safety level at intersection approaches than existing measures. The results of this study are relevant to collision-avoidance systems. SSH could be used as a performance measure to evaluate different DZ-protection strategies, and could help prioritize different sites for crash mitigation. Our manuscript is organized as follows: the following section provides a background review of related efforts. Next, the study’s methodology is described, including the research approach and the data-collection system. Then an application of the method is presented, including a description of the study sites and data, followed by results and discussion. The last section draws conclusions on the proposed approach and discusses potential future research efforts. 2. Background Contributing factors to DZ-related safety issues, DZ modeling approaches, and crash mitigation strategies in DZ have been widely studied for decades. A review of these research efforts is provided in research studies such as Abbas and Ghanipoor Machiani (2013); Adam et al. (2009); Jahangiri et al. (2015); Papaioannou (2007); Ghanipoor Machiani and Abbas (2014a,b); Abbas et al. (2014a); Rakha et al. (2008) and Sharma et al. (2011). The most relevant research regarding our study is on collision and safety measures in dilemma zones. A broad range of transportation studies have been dedicated to collision prediction and avoidance models and algorithms (for examples, see (Jansson and Gustafsson, 2008; Kuchar and Yang, 2000; Sato and Ishii, 1998)). The first measure used to evaluate traffic safety was the number and severity of crashes. However, crashes occur infrequently, and transportation engineers must evaluate safety before facilities are built, so surrogate safety measures began to be considered. Surrogate safety measures are “measures of safety not based on a series of actual crashes (Gettman et al., 2008)”. Many surrogate measures have been introduced by researchers, such as vehicle delay, queue lengths, percentage of left turns, and deceleration distribution (Gettman and Head, 2003). One of the recent efforts related to surrogate measures is the development of the surrogate safety assessment model (SSAM) software tool. SSAM uses a specific format of vehicle trajectory data obtained from simulation software to evaluate

simulated traffic conflicts. Some simulation software packages can produce the format used by SSAM (Gettman et al., 2008). Surrogate safety measure is a main concept in traffic-conflict analysis. A conflict is defined as a situation in which there is a risk of collision between two or more road users, if neither changes his or her trajectory. The traffic conflicts technique is based on field observations, and considers sudden maneuvers such as slamming on the brakes (Gettman and Head, 2003). Time to collision (TTC) is the primary severity measure for conflicts, and is defined as the expected time for two vehicles to collide, if they maintain their speeds and directions (Gettman and Head, 2003). TTC is widely used as a safety indicator to measure crash risk, evaluate roadway safety, and test the importance of contributing factors to crashes (Meng and Weng, 2011; Kiefer et al., 2006). TTC was suggested as a scale of danger more than four decades ago by Hayward (1972). Hayward defined TTC as the time measured until collision between two vehicles, if the collision situation and speed difference were maintained (Hayward, 1972). This measure is calculated using different sources of data, including videos and trajectories. For examples, Hayward (1972) used the Federal Highway Administration’s (FHWA) traffic sensing and surveillance system to calculate TTC from videos at an urban intersection (Hayward, 1972). Xu and Qu (2014) also used video data from a Beijing expressway to investigate the effect of road environment, traffic conditions, and vehicle types on TTC. Their results showed that vehicle type did not correlate with TTC averages; however, traffic conditions and road environments significantly influenced TTC (Xu and Qu, 2014). Minderhoud and Bovy (2001) used vehicle trajectory from a specific road segment and certain time period to calculate TTC, and to introduce safety measures based on that TTC value. These safety indicators, time exposed, time-to-collision, and time integrated time-to-collision, are applicable to intelligent driver-support systems (Minderhoud and Bovy, 2001). The concept of TTC has been applied to different crash situations and maneuvers, such as passing maneuvers on rural two-lane highways (Farah et al., 2009) and rear-end vehicle crashes in urban road tunnels (Meng and Qu, 2012). Farah et al. (2009) used an interactive driving simulator to develop a model to predict the risk during passing maneuvers on rural two-lane highways. The measure of risk in their study was TTC. Meng and Qu (2012) proposed a model to estimate rear-end vehicle crash frequency in road tunnels based on TTC distributions. TTC was analyzed using data from traffic videos of two road tunnels in Singapore. They concluded that an inverse Gaussian distribution was the best model for relating TTC to its contributing factors. Berthelot et al. (2012) also developed a real-time algorithm to compute the probability distribution of TTC applicable to vehicle design and an advanced driver-assistance system. They evaluated the accuracy of their approach by simulating several crossing scenarios. Kiefer et al. (2006) investigated TTC from a driver’s point of view, or a TTC judgment. In their test-track research, participants were asked to indicate the time to collision to a lead vehicle by pressing a button. The results showed that when a driver’s speed decreased, or his or her relative speed increased, the ratio of perceived TTC to actual TTC increased. Hoffmann and Mortimer (1994) also examined drivers’ estimation of the time to collision in a laboratory simulation. The results showed that drivers underestimated the time to collision when the value of TTC was low (Hoffmann and Mortimer, 1994). Some research efforts have compared TTC with other safety measures, such as the inverse time to collision (Balas and Balas, 2006) and headway (Vogel, 2003). Vogel (2003) indicated that a suggested TTC threshold in the literature ranges from 1.5 to 5 seconds, and TTC larger than 6 s was not considered to be dangerous (Vogel, 2003).




Although TTC has been widely researched, to the best of our knowledge, this measure has not been directly used in DZ-related studies. DZ conflicts are one of the leading cause of crashes at signalized intersections (Li and Abbas, 2010). Surrogate measures such as stop-bar encroachments, red and yellow light violations by phase, and number of vehicles caught in a DZ have been considered in this regard, although no quantitative assessment has been done to relate these measures to the number of crashes (Gettman and Head, 2003). In this research, we take advantage of the TTC concept to develop a novel safety surrogate measure that we call the safety surrogate histogram (SSH) of DZ-related conflicts at signalized intersections. This new measure provides useful information and insightful perspective regarding the level of safety at signalized intersections, and it could be a valuable addition to existing metrics.

3

Fig. 1. TTC values plotted on a time-space diagram. (For interpretation of the references to color in text, the reader is referred to the web version of this article.)

3. Methodology 3.1. Research approach and rationale Our approach relies on TTC, not the number of vehicles caught in a DZ. Utilizing real-time vehicle trajectory data obtained from commercial radar detection systems at signalized intersections, we measure and plot the TTC for every two consecutive vehicles on a time space diagram. If there are more than two vehicles, each pair is analyzed separately. Fig. 1 is an example time-space diagram for vehicles A and B, where vehicle A is following vehicle B. At every point in time, the TTC can be calculated based on each vehicle’s speed. In the figure, the slope of each line corresponds to the speed of the vehicle. For example, the slope of the curves at point c, shown in red, represents the speed of two vehicles at this point in time. If two lines with those slopes are drawn, the point where they intersect is the time of collision. Thus, the time between the intersecting point of two lines with the slopes of the curves at point c, and point c itself, is the TTC from point c, and is shown as TTCc in the figure. The TTC can be calculated for any other point in time. In the figure, two other TTC values are illustrated at points d and f, and are named TTCd and TTCf, respectively. At point c, vehicle A has started to decelerate to avoid the collision, so as time passes between point c and point f, vehicle A’s speed decreases, and the

TTC value increases. At point e, the two slopes become almost parallel and the TTC value approaches infinity. If the front vehicle continues through the intersection and the back vehicle comes to a stop, the value of TTC could become negative. Thus, if TTC values are positive and decreasing, the vehicles are increasingly exposed to dangerous situations, and a small positive TTC value indicates an imminent collision. Negative TTC values mean the two subject vehicles are moving away from a possible collision. Instead of using the slope of the continuous time-space diagram to calculate the TTC, we used speed data collected discretely over time. In cases where speed data were not available for the lead vehicle, the speed at the moment of interest was interpolated. Using speed data is more realistic and accurate than calculating the slope of a curve fitted to data points in a discrete time-space diagram. To illustrate this concept, consider Fig. 2 below, where an imminent collision and a safety conflict between vehicles A and B are illustrated. Vehicle trajectory field data are shown as dots on the time-space diagrams. The color of each dot indicates the measured speed as shown on the color map to the right of the chart. The x-axis represents time in seconds overlaid by the signal indication colors (green, yellow, and red). The y-axis shows the range (distance) in feet, where zero is the intersection’s stop bar. Vehicle A is following vehicle B as they approach the intersection.

Fig. 2. Time-space diagram color coded by speed for approaching vehicles. (For interpretation of the references to color in text and figure legend, the reader is referred to the web version of this article.)



4


The portions of the vehicles’ trajectories showing the discussed behavioral changes are indicated by dotted polygons. During yellow, as vehicle B decides to stop, to avoid collision, vehicle A needs to change its decision and also stop. As vehicle A starts to decelerate, it increases its distance to the leading vehicle, and the TTC changes to greater positive values. Therefore, monitoring changes in the value of TTC for approaching vehicles, as well as the frequency of each TTC value, would provide a characteristic assessment of the safety level at the intersection. Based on the literature, TTC values that are greater than six seconds are considered safe (Vogel, 2003). In this research, a histogram is generated per intersection approach to show the frequency of TTC less than six seconds with a bin size of one unit. This measure is called the safety surrogate histogram (SSH). The concept behind the SSH is related to the behavior of traffic passing through the red and green shockwaves at the intersection. Fig. 3 illustrates this relationship. The stop bar is at the top of Fig. 3a, color coded based on the signal indication. The x-axis and y-axis represent time and distance, respectively. The start and end of the red phase form the two shockwaves, shown in purple. Consider a case where traffic is approaching the signal. Vehicle 2 is following vehicle 1, and they are going to encounter a red signal. Vehicle 1 brakes and slows down. At this moment, the TTC for these

two vehicles decreases to a very low value. This is shown in Fig. 3b where the blue line, representing vehicles 1 and 2, has a steep negative slope. The value of the TTC remains low until vehicle 2 also starts to slow down. Afterward, the TTC increases as the vehicles reach the same speed and pass through the intersection. Consider the same situation for vehicles 3 and 4, who encounter the far end of the shockwaves. Vehicle 3 reduces its speed to accommodate the shockwave, so the TTC for these two vehicles drops slightly as shown in Fig. 3b on the brown graph. The TTC values for these two vehicles return to large values faster than the TTC values for vehicles 1 and 2. In essence, only a few vehicles in each cycle are exposed to sever turbulence at or near the onset of yellow and near the start of the shockwave. The rest experience low turbulence. Therefore, as the red light duration increases, so does the number of vehicles facing the red light shockwave, and the percentage of vehicles with severe turbulence (i.e., lower TTC values), becomes lower. This would in turn skew the SSH distribution towards higher values. 3.2. VT-SCORES real-time intersection data collection system The data used in this study were collected using the secondgeneration intersection safety data collection and evaluation

Fig. 3. TTC concept in relation to shockwave. (For interpretation of the references to color in text, the reader is referred to the web version of this article.)




5

Fig. 4. VT-SCORES safety data collection and evaluation system (Abbas et al., 2014b).

Fig. 5. Wavetronix setup (Wavetronix, 2014).

system developed by the Virginia Tech Signal Control & Operations Research and Education System (VT-SCORES) lab. The system, illustrated in Fig. 4, uses the commercial Wavetronix system integrated with signal phase, detector, and video data. It includes a hardened field computer, two advanced bus interface units (BIU), a Sierra wireless modem, two Wavetronix Click 304 units, and up to four camera streams. The controller and other components stream data using a terminal and facilities (TF1) BIU including signal phase status, and a detector BIU with detector status, to the computer, where the data is stored. The Sierra wireless modem provides wireless communication and allows data to be transferred to the lab. The two Wavetronix Click 304 systems (Fig. 5) continuously monitor approaching vehicles and collect radar data for every intersection approach. The Wavetronix can simultaneously track up to 25 vehicles per approach.

at this site. The second site, the US460 site, shown in Fig. 7b, was located at US 460 and Southgate Drive. A PEEK video detection

4. Application of the methodology 4.1. Study sites The data collection system was installed at two T-intersections in Virginia, both featuring TS-2 cabinets (shown in Fig. 6). The first intersection, the US220 site, shown in Fig. 7a, is located at US 220 and Route 87. An Autoscope video detection system was used

Fig. 6. Installed system at a TS-2 cabinet.



6


Fig. 7. Study sites: (a) US220 site and (b) US460 site.

system was used at this site. The AADT for the US460 site is 32,000, and that at the US220 is 16,000. The speed limit for US460 site is 55 mph (but most traffic travels at 65 mph), and for the US220 site is 45 mph. Data were collected on Southbound and Northbound for the US220 site, and Eastbound and Westbound for the US460 site. 4.2. Data description The data collection process started in May 2014 and continued for over 5 months. For every hour, the raw data included a csv file and some short avi (video) files that captured red light runners. The csv file contained vehicle tracks, speed, and range

data obtained from the radar system. It also stored phase information and counts from detectors located at the intersections. The reduction and manipulation of the data was performed using a MATLAB script. Fig. 8 shows an example of the processed data for one day. The x-axis shows time slots for 24 h with data named according to site_date_time. The y-axis illustrates the number of vehicles for every cycle, color coded based on the number of vehicles caught in the DZ during each cycle. A data analysis tool was developed to generate integrated data plots and to analyze the data for TTC calculations. An example of a time-space diagram extracted from the data is shown in Fig. 9. Vehicle trajectories are shown in as dots colored

Fig. 8. An example of collected data; number of vehicles per cycle color coded by the number of vehicles caught in the DZ over 24 h. (For interpretation of the references to color in text and figure legend, the reader is referred to the web version of this article.)




7

Fig. 9. An example of the interactive time-space plots. (For interpretation of the references to color in text, the reader is referred to the web version of this article.)

based on the vehicle’s speed, and according to the color map to the right of the plots. The color on the x-axis indicates the signal indication (green, yellow, and red). 4.3. Results and discussion Vehicle speed data and their corresponding TTC values were extracted from the time-space diagram for each vehicle pair. In this analysis, we only included positive TTC values under six

seconds, because TTC values over six seconds were considered safe. The final SSH values were then normalized by dividing them by the number of vehicles in their corresponding cycles. Therefore the reported SSH is per vehicle per cycle. First, we compared two SSH examples from the US460 site. Figs. 10 and 11 show the two SSH graphs for a shorter red light duration and a longer red light duration, respectively. The figures also show trend lines superimposed on the SSH values. It can be seen from the two figures that:

Fig. 10. SSH for shorter red light duration for US460 EB and WB. (For interpretation of the references to color in text and figure legend, the reader is referred to the web version of this article.)



8


Fig. 11. SSH for longer red light duration for US460 EB and WB. (For interpretation of the references to color in text and figure legend, the reader is referred to the web version of this article.)

With a shorter red light duration, the low TTC values are close in

value to the trend lines. This can be interpreted as a consistent SSH trend. With a longer red light duration, the low TTC values are significantly lower than the trend line, which means that high TTC values are more frequent in the distributions (pushing the whole trend line up). This means that a long red duration is safer (only first vehicles in the platoon approaching a red light are experiencing low TTC values). However, for the westbound approach, we do not see this effect. It should be noted that the webs bound approach at the US460 site has a horizontal curve near the intersection. Therefore, the longer red light duration causes the queue to lengthen, reducing the sight distance of drivers passing the curve. This leads to further turbulence experienced by vehicles at the back of the platoon, increasing the frequency of low TTC values and making the SSH distribution consistent with the trend line again. So, in essence, the positive effects of a longer red light are countered by negative effects due to queue formation (only at approaches with potential sight distance problems).

Second, we investigated the results obtained from one hour of data for each of the study sites. The US460 site had higher volume and speed than the US220 site. Figs. 12 and 13 show the SSH graphs during an hour of data collection for the two sites at the same date and time of the day and for every approach. The SSH has been normalized per vehicle and number of cycles. According to the figures, the SSH values are generally higher for US460 due to higher speed, making the US460 site a higher priority for safety improvement. For both sites, approaches that have a left turning movement, shown by blue bars in the figures, have higher values of SSH due to left turning turbulence. This is more apparent at the US460 site. Finally, we provide a contrast between the results of the SSH and three widely-used safety measures, namely the number of vehicles caught in DZ, the number of yellow light runners (YLR), and the number of red light runners (RLR). Table 1 summarizes these measures normalized per vehicle and number of cycles. The numbers of red light and yellow light runners are usually used as surrogate measures for right-angle crashes, while the number of vehicles caught in a DZ is used as a surrogate measure for rear-end crashes. In our case, the number of red light runners for both sites is

US220 site 0.6 0.5 0.4 0.3 0.2 0.1 0 1

2

3 Southbound

4

5

6

Northbound

Fig. 12. Safety surrogate histogram (SSH) for US220 site. (For interpretation of the references to color in text, the reader is referred to the web version of this article.)




9

US460 site 0.6 0.5 0.4 0.3 0.2 0.1 0 1

2

3 Eastbound

4

5

6

Westbound

Fig. 13. Safety surrogate histogram (SSH) for US460 site. (For interpretation of the references to color in text, the reader is referred to the web version of this article.)

Table 1 Safety surrogate measures other than SSH for the two study sites on both approaches. Site

Direction

Number of vehicles caught in a DZ

Yellow light runners (YLR)

Red light runners (RLR)

US220 US220 US460 US460

Southbound Northbound Westbound Eastbound

0.008696 0.020833 0.011777 0.011002

0.006522 0.010417 0.013374 0.04401

0 0 0 0

0, so it does not provide any detailed safety information. While the number of vehicles caught in a DZ and the SSH share some similarity with regard to the type of accidents for which they are surrogates, SSH includes more crash-related information, because it is directly extracted from the time-to-collision concept. Consider a situation where the number of vehicles caught in a DZ is 1. However, if this vehicle is the only vehicle in the intersection, or other vehicles are far enough away, there is no potential for a rear-end crash. Unlike number of vehicles caught in a DZ, the SSH only reports values when there is definitely a conflicting pair of vehicles. The detailed information embedded in the SSH is also informative when queues are present. In such cases, the number of vehicles caught in the DZ does not relate to the existing queue, whereas the SSH takes queues into account. The SSH measure therefore provides more information for DZ-protection algorithms, because it can ignore situations that are not as dangerous, such as lone vehicles in a DZ, but can also raise alarms for truly dangerous situations. When used in addition to other safety measures, SSH could play an important role in evaluating the level of safety at intersections. A high SSH value is a red flag for practitioners, and indicates they should further investigate the safety problems causing such high SSH values. These investigations could range from examining the signal phasing and control settings to examining the geometry of the intersection. SSH could also be used in connected-vehicles applications to send appropriate speed adjustment messages to vehicles approaching an intersection to smooth out the shockwave. This new measure of safety paves the road for new DZ-protection algorithms, such as queuing-model-based DZ protection algorithms, because SSH contains more information regarding queues. 5. Conclusions Introducing proper traffic safety measures has been the subject of much research. Safety measures are needed to determine the

level of safety at roadways and to plan appropriate crashmitigation strategies, and a comprehensive safety measure calculated from road data is in high demand. The focus of this study was on traffic safety at signalized intersections, and the evaluation of DZ-related conflicts. In this paper, we presented a new methodology for characterizing safety level at each approach of a signalized intersection. Moreover, a new safety surrogate measure was developed and is called the safety surrogate histogram (SSH). Time-space diagrams and speed data were used to produce a SSH for every approach at signalized intersections, and examples applying the methodology at two test sites in Virginia were provided. This new safety measure was compared to other safety measures from the intersection-safety literature, such as the number of vehicles caught in DZ, number of yellow light runners, and number of red light runners. The results showed that SSH did not highly correlate with the other safety surrogate measures. According to our analysis, the SSH would be more informative to DZ-protection algorithms than other measures. For example, safety concerns regarding changes in signal timing are measurable using the SSH measure, but not with the widely-used number of vehicles caught in DZ measure. Therefore, the measure introduced here could be a useful addition to existing metrics, and could help stakeholders better evaluate the level of safety at signalized intersections. The SSH measure in this research was calculated using the VT-SCORES data-collection system; however, any data-collection system reporting vehicle trajectory information could be used to calculate the SSH, with the level of accuracy depending on the precision of the system. For comparisons to be made between intersections under study, data should be collected at the same level of accuracy. In the future, we intend to address some of the limitations of this newly developed approach by studying the relationship between each SSH component in an attempt to develop a single value regarding safety for each approach. We will also investigate how the SSH changes between and within different sites, and



10


whether some conclusions could be made regarding potential SSH threshold values. The developed SSH will also be evaluated for a longer data-collection period to determine the critical hours when vehicles are more prone to accidents. Intersection traffic and signal characteristics will be assessed to determine the positive and negative factors affecting the SSH distribution, so better crash mitigation strategies and algorithms can be developed. Also, a new DZ-protection system will be installed at the subject intersections. The SSH safety surrogate measure will be applied to quantify the benefits of the new system. Acknowledgements The research undertaken in this paper was sponsored by Virginia Center for Transportation Innovation and Research (VCTIR). The authors are solely responsible for the material in this paper and the views are not necessarily those of the supporting agency. The authors would like to thank Ms. Cathy McGhee, Mr. Ben Cottrell, and the oversight research panel for their support. We would like to especially thank Mr. Joe Smith and his team for their dedication and great assistance and support during the field installation process. References Abbas, M., Ghanipoor Machiani, S., 2013. Modeling the dynamics of driver’s dilemma zone perception using agent based modeling techniques. Conference on Agent-Based Modeling in Transportation Planning and Operations, Blacksburg, VA (Special issue). Abbas, M., Ghanipoor Machiani, S., Garvey, P.M., Farkas, A., Lord-Attivor, R., 2014a. Modeling the Dynamics of Driver’s Dilemma Zone Perception Using Machine Learning Methods for Safer Intersection Control. US Department of Transportation, Research & Innovative Technology Administration, pp. 84. Abbas, M., Higgs, B., Ghanipoor Machiani, S., 2014b. Integrated real-time data collection and safety improvement system at signalized intersections. The 17th International IEEE Conference on Intelligent Transportation Systems, Qingdao, China. Adam, Z.M., Abbas, M.M., Li, P., 2009. Modeling the complexity of driving behavior during signal yellow interval using reinforcement learning accession number. Transportation Research Board 88th Annual Meeting, Washington, D.C. Balas, V.E., Balas, M.M., 2006. Driver assisting by inverse time to collision. Automation Congress, WAC’06 World. Berthelot, A., Tamke, A., Dang, T., Breuel, G., 2012. A novel approach for the probabilistic computation of time-to-collision. Intelligent Vehicles Symposium (IV), IEEE. Chaudhary, N.A., Abbas, M.M., Charara, H.A., 1978. Development and field testing of platoon identification and accommodation system. Transp. Res. Rec.: J. Transp. Res. Board 1, 141–148. Engstrom, A., 1994. 10 years with LHOVRA – what are the experiences? Road Traffic Monitoring and Control, Seventh International Conference. Farah, H., Bekhor, S., Polus, A., 2009. Risk evaluation by modeling of passing behavior on two-lane rural highways. Acc. Anal. Prev. 41 (4), 887–894.

Gettman, D., Head, L., 2003. Surrogate Safety Measures From Traffic Simulation Models, Final Report. Office of Safety Research and Development, Federal Highway Administration, McLean, VA. Gettman, D., Pu, L., Sayed, T., Shelby, S., 2008. Surrogate Safety Assessment Model (SSAM) and Validation: Final Report. Office of Safety RD&T, Federal Highway Administration. Ghanipoor Machiani, S., Abbas, M., 2014a. Predicting drivers decision in dilemma zone in a driving simulator environment using canonical discriminant analysis. The 93rd Annual Meeting of the Transportation Research Board, Washington, D.C. Ghanipoor Machiani, S., Abbas, M., 2014b. Dynamic driver’s perception of dilemma zone: experimental design and analysis of driver’s learning in a simulator study. The 93rd Annual Meeting of the Transportation Research Board, Washington, D.C. Hayward, J., 1972. Near-miss determination through use of a scale of danger. Highway Res. Rec. 384, 22–34. Hoffmann, E.R., Mortimer, R.G., 1994. Drivers’ estimates of time to collision. Acc. Anal. Prev. 26 (4), 511–520. Jahangiri, A., Rakha, H., Dingus, T.A., 2015. Predicting red-light running violations at signalized intersections using machine learning techniques. Transportation Research Board 93rd Annual Meeting, Washington, D.C. Jansson, J., Gustafsson, F., 2008. A framework and automotive application of collision avoidance decision making. Automatica 44 (9), 2347–2351. Kiefer, R.J., Flannagan, C.A., Jerome, C.J., 2006. Time-to-collision judgments under realistic driving conditions. Hum. Factors 48 (2), 334–345. Kuchar, J.K., Yang, L.C., 2000. A review of conflict detection and resolution modeling methods. Trans. Intell. Transp. Syst. 1 (4), 179–189. Li, P., Abbas, M., 2010. Stochastic dilemma hazard model at high-speed signalized intersections. J. Transp. Eng. 136 (5), 448–456. Meng, Q., Qu, X., 2012. Estimation of rear-end vehicle crash frequencies in urban road tunnels. Acc. Anal. Prev. 48 (0), 254–263. Meng, Q., Weng, J., 2011. Evaluation of rear-end crash risk at work zone using work zone traffic data. Acc. Anal. Prev. 43 (4), 1291–1300. Minderhoud, M.M., Bovy, P.H.L., 2001. Extended time-to-collision measures for road traffic safety assessment. Acc. Anal. Prev. 33 (1), 89–97. Papaioannou, P., 2007. Driver behaviour, dilemma zone and safety effects at urban signalised intersections in Greece. Acc. Anal. Prev. 39 (1), 147–158. Rakha, H., Amer, A., El-Shawarby, I., 2008. Modeling driver behavior within a signalized intersection approach decision-dilemma zone. Transp. Res. Rec.: J. Transp. Res. Board 2069 (1), 16–25. Sato, Y., Ishii, H., 1998. Study of a collision-avoidance system for ships. Control Eng. Pract. 6 (9), 1141–1149. Sharma, A., Bullock, D., Peeta, S., 2011. Estimating dilemma zone hazard function at high speed isolated intersection. Transp. Res. Part C: Emerging Technol. 19 (3), 400–412. Vincent, R.A., Peirce, J.R., 1988. MOVA: traffic responsive, self-optimising signal control for isolated intersections. Research report (Transport and Road Research Laboratory). Traffic Management Division, Traffic Group, Transport and Road Research Laboratory, Crowthorne, Berkshire p. 170. Vogel, K., 2003. A comparison of headway and time to collision as safety indicators. Acc. Anal. Prev. 35 (3), 427–433. Wavetronix LLC, 2014. http://www.wavetronix.com/en. Xu, C., Qu, Z.W., 2014. Empirical analysis on time to collision at urban expressway. Appl. Mech. Mater. 505–506, 1127–1132. Zimmerman, K., Bonneson, J., 2005. In-Service Evaluation of a Detection-Control System for High-Speed Signalized Intersections, vol. 5. Texas Transportation Institute, Texas A&M University System. Zimmerman, K., Bonneson, J.A., Middleton, D., Abbas, M.M., 2003. Improved detection and control system for isolated high-speed signalized intersections. Transp. Res. Rec.: J. Transp. Res. Board 1856 (1), 212–219. Zimmerman, K., 2007. Additional dilemma zone protection for trucks at high-speed signalized intersections. Transp. Res. Rec.: J. Transp. Res. Board 2009 (1), 82–88.


Safety propensity index for signalized and unsignalized intersections: exploration and assessment.

Modeling the frequency of opposing left-turn conflicts at signalized intersections using generalized linear regression models.

A Heckman selection model for the safety analysis of signalized intersections.

Multimodal injury risk analysis of road users at signalized and non-signalized intersections.

Red light violations by adult pedestrians and other safety-related behaviors at signalized crosswalks.

Estimating rear-end accident probabilities at signalized intersections: a comparison study of intersections with and without green signal countdown devices.

Comparative analysis of risky behaviors of electric bicycles at signalized intersections.

Evaluation of Deployment Challenges of Wireless Sensor Networks at Signalized Intersections.

An accelerated failure time model for investigating pedestrian crossing behavior and waiting times at signalized intersections.

Estimating safety performance trends over time for treatments at intersections in Florida.

A novel safety assessment strategy applied to non-selective extracts.

Contraceptive counselor's dilemma: safety or effectiveness?

Safety performance models for urban intersections in Brazil.

A hazard-based duration model for analyzing crossing behavior of cyclists and electric bike riders at signalized intersections.

Crowded skies: Conflicts between expanding goose populations and aviation safety.

Comprehensive safety management and assessment at rugby football competitions.

Capacity Estimation Model for Signalized Intersections under the Impact of Access Point.

Patient safety at home.

Animal-Free Chemical Safety Assessment.

Readmissions at a public safety net hospital.

PPT-E1A], a novel oncolytic adenovirus for prostate cancer.

A safety culture assessment by mixed methods at a public maternity and infant hospital in China.

Multi-level hot zone identification for pedestrian safety.

Safety zone for posterosuperior shoulder access: study on cadavers.