You are on page 1of 18

MEASURING SAFETY PERFORMANCE AT ROUNDABOUTS USING VIDEOTAPED VEHICLE TRACKING DATA Giuseppe Guido1*, Researcher, Ph.D.

E-mail: guido@unical.it

Frank F. Saccomanno2, Professor, Ph.D., P. Eng.


E-mail: saccoman@uwaterloo.ca

Vittorio Astarita1, Associate Professor, Ph.D.


E-mail: vastarit@unina.it

Alessandro Vitale1, Ph.D.


E-mail: alessandro.vitale@unical.it 1. Dipartimento di Pianificazione Territoriale Universit della Calabria Via P. Bucci, cubo 46/B 87036 Rende (CS), Italy Tel: (+39) 0984496780 Fax: (+39) 0984496780 2. Department of Civil and Environmental Engineering University of Waterloo 200, University Street West Waterloo, ON, Canada N2L 6P6 Tel: (519) 888-4567 x. 6596 Fax: (519) 888-5221

Text: Figures: Tables: Total words:

3,659 10 x 250 = 2,500 3 x 250 = 750 6,909

* Corresponding author Resubmitted November 14, 2008

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

ABSTRACT A low cost experimental procedure is presented for obtaining vehicle tracking data from video at a given roundabout location. These data are examined frame by frame to extract individual vehicle position and time as it progresses along its path. Accuracy of the traffic estimates are established by comparing vehicle speed profiles from the vehicle tracking data with speed profiles obtained independently using a laser speed gun at the same roundabout location. The vehicle tracking data was used to estimate safety performance at different segments of the roundabout using different indicators, such as, time to crash, deceleration rate to avoid the crash, crash potential index that incorporates the braking capability of individual vehicles, number and exposure time in conflict. Differences in safety performance are discussed with respect to the type of indicator used, assumed traffic inputs and differences in roundabout geometry. The results of this study provide meaningful experimental indicators of potential safety problems at different roundabout locations subject to behavioural driver responses to different traffic conditions. INTRODUCTION Safety performance measures are becoming increasingly recognized as an effective means of investigating potential safety problems at different road locations, and of guiding costeffective intervention strategies for resolving these problems. These measures are inherently sensitive to both fixed geometric attributes as well as to a myriad of time-dependent traffic conditions affecting safety. As such, they are most useful in identifying real-time high risk driving behaviour for individual vehicles in the traffic stream, and in assessing their potential for crash involvement (1, 2, 3). Safety performance measures can be obtained either through simulation based on calibrated traffic models or experimentally through observational vehicle tracking data. Regardless of the approach adopted (i.e. simulation or observational), the estimation of safety performance needs to be linked to vehicle speed, acceleration and position information observed at a given location over a given period of time (4). The basic aim of this paper is to present a low cost procedure for extracting accurate and reliable vehicle tracking data from a given road location for input into estimation of safety performance. The context for this experiment is the measurement of safety performance at roundabouts with respect to changes in the potential for rear-end vehicle interactions. This exercise has four basic objectives: 1. Develop a low cost experimental procedure for obtaining individual vehicle traffic profiles (speed and position) at different roundabout segments from video. Investigate statistically the precision of vehicle speed/position profiles extracted from video capture when compared to speed profiles obtained from laser speed gun at representative roundabout locations.

2.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

3.

4.

Using the vehicle tracking data, obtain estimates of safety performance at different roundabout locations using several different performance expressions (or indicators). Explore areas of high risk exposure on the roundabout, and investigate how these areas are affected by the measures being used.

Roundabouts have been receiving increased attention especially in North America as an effective means of providing safer intersection control than conventional fixed signal controls. The assertion has been made that roundabouts offer a smoother transition between merging or interacting vehicles, and hence reflect a reduced exposure to abrupt rear end interactions. Proponents of roundabouts suggest that this can produce significant reductions in severe rear-end crashes, as compared to conventional signalized intersections. The case in support of more roundabout control has not been fully made scientifically. A major problem has been the lack of sound vehicle tracking data required to obtain accurate and reliable measures of safety performance. The first challenge encountered in this research was to develop a procedure for extracting such data at a typical roundabout location. VIALE COSMAI ROUNDABOUT DATA GATHERING The Viale Cosmai roundabout offers an excellent opportunity for investigating safety performance at roundabouts using experimental vehicle tracking data. The Viale Cosmai roundabout is situated in the city of Cosenza, Italy, and serves to provide directional control between a major four lane divided arterial (Viale Cosmai) and a minor two lane undivided road (Via Montana). As illustrated in Figure 1, this roundabout comprises of an elliptical two-lane ring with four entry/exit branches at different angles and turning radii. Over 30% of the traffic flow in the roundabout is through traffic on Viale Cosmai. Vehicle trajectories at this case study roundabout were captured by video taping the traffic over a period of one hour using a video camera positioned on the roof of a nearby building at a height of 12 meters above roundabout (shown in Figure 2). The video taping was carried out during a typical weekday between 11:30 am and 12:30 pm, a period coincident with off-peak traffic conditions at this location. The camera position selected for this experiment permits monitoring of almost all the trajectories for vehicles entering/exiting and circulating along the ring. For the one hour taping interval, traffic flow on the ring and on each of the four entry/exit points was found to vary considerably from a low of 120 vph to a high of 2,340 vph.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

FIGURE 1 Plan view of Viale Cosmai roundabout with geometric features. Video images were processed using Adobe Premiere software to permit the analysis of different trajectories on a frame by frame basis. The time interval between two sequential frames was established at 1/30 of a second. This was found to provide a suitable precision for measuring individual vehicle progression along its path 0.1 sec time increments. This is consistent with minimum time increments normally used in many current microscopic simulation models, such VISSIM or PARAMICS. In order to obtain real time microscopic parameters of vehicle progression every 0.1 seconds, it was necessary to position a series of virtual detectors at various points along the roundabout. Given an average speed of 25 km/h along the ring, an optimal spacing of 1 m was selected between successive virtual detectors. A total of 176 virtual detectors were positioned as illustrated in Figure 2, corresponding to eight entry/exit paths: 0-84, 0-326, 0414, 101-84, 101-326, 101-414, 201-84 and 201-414.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

FIGURE 2 Superimposition of detectors on an example frame. An accurate juxtaposition of virtual detectors was maintained with reference to three known roundabout feature points and the known location of the camera (obtained from a gps survey). This permitted a precise positioning of all 176 detectors along the roundabout with an accurate matching of vehicle progression as tracked along each path over time and space. For the purpose of demonstration, the extraction of vehicle tracking data has been restricted to a selected number of eight representative paths or trajectories. These have are illustrated in Figures 3 for two of the eight representative paths (0-84 and 0-326) with their corresponding detector ID numbers. To accurately track each vehicle position and time, information is needed on the time the vehicle crosses three sequential detectors along its path. Two detectors are used to estimate vehicle speed and three adjacent detectors are used to yield vehicle deceleration/acceleration rates. Given the short distance between adjacent detectors, the assumption is made that the change in speed between these detectors is constant. For each trajectory or path, the time each vehicle crosses a detector is obtained through a frame by frame analysis of the video images. Each element of this matrix represents the transit time on the detector observed for each vehicle. A total of 78 vehicle trajectories were examined and their time profiles were obtained at each virtual detector along their specific paths. The results are shown in Figure 4 for two of eight paths being considered: 0-84 and 0-326. The information shown in this Figure provides traffic flow inputs for the estimate of individual vehicle safety performance as it progresses along a given path.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

FIGURE 3 Representative vehicle trajectories and associated detectors for paths 0-84 and 0-326.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

FIGURE 4 Vehicle time-space diagram for paths 0-84 and 0326. In order to use these estimates of vehicle progression for the estimation of safety performance, it is necessary to first link the time and speed profile for a given following vehicle (FV) to its corresponding lead vehicle (LV). Depending on the position of the LV on roundabout (entry, exit or ring), its status with respect to the FV may change along the

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

roundabout. The pairing of FV to LV requires an in-depth analysis of vehicle positions over time from both a longitudinal and lateral perspective. For the eight paths being considered in this study, a total of 58 FV-LV pairs were examined. Relative speed and spacing between LV and FV along each path are being obtained from the transit time spreadsheets. Acceleration rates for individual FV were obtained in 0.1 sec increments by comparing the speed values between each FV-LV pair at their respective virtual detector positions. ASSESSING ACCURACY OF VEHICLE TRACKING SPEED PROFILES In order to determine as to how well the speeds extracted from video capture at virtual detectors compares with observed speeds in the field, an independent speed survey was applied to a selected sample of vehicles crossings a given station using a laser speed gun (Laser 500, Traffic Safety Systems). The speeds determined from the laser gun were then compared to speeds estimated at the same stations from the video capture data. A total of six representative stations were selected for this comparison as shown in Figure 5. Four stations were set at roundabout exits, one at a merging section of the ring, and one station at a entry section on the ring. For each section, 30 measures of speed were observed using the laser gun and compared to speeds extracted from video capture at the same location.

FIGURE 5 Representative stations for checking speeds estimates. A statistical analysis of vehicle speeds was carried out to compare video capture and radar gun speed profiles. Speed measures derived from both methods were determined to

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

follow the normally distribution, as illustrated in Figure 6 are for observations obtained at detector station n. 30. A Chi-Square test confirmed the results that the speed profiles extracted from the laser gun application compared closely to those extracted from the vehicle tracking data at station no. 30. The measured speed profiles for Laser gun and vehicle tracking samples at this station were both found to follow a normal distribution with a mean of 10.84 km/h and 10.90 km/h and variance of 5.33 km/h and 5.25 km/h respectively. The differences were not found to be statistically significant at the 5% level.

FIGURE 6 Statistic analysis of speed distribution on detector 30 obtained by Laser measurements and video capture.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

10

MEASURING SAFETY PERFORMANCE Several measures of individual vehicle safety performance have been introduced in the literature including: time-to-collision (TTC), extended time-to-collision (TET), post encroachment time (PET), initial deceleration rate (DR), maximum deceleration to avoid collision, etc. (5, 6, 7, 8). In this paper, safety performance for individual vehicles is expressed using three distinctive indicators: 1. Time to Crash (TTC)

Time to crash was defined by Hayward (5) to reflect the time separating a given FV from its corresponding LV, where their differential speeds are such that both vehicles are closing in on each other. When TTC reaches a threshold value of 1.5 sec (minimum perception/reaction time) suggested by Van der Horst (9), the two vehicles are assumed to be in conflict or in an unavoidable collision path. 2. Maximum Deceleration Rate to Avoid a Crash (DRAC)

A recent FHWA report and a subsequent article by Archer (3) have explicitly recognized the relevance of DRAC as a measure of safety performance. DRAC explicitly considers the role of speed differentials and decelerations in crash occurrence. DRAC was defined by Almquist et al. (10) in terms of the speed differential between FV and LV divided by their closing time. The LV is responsible for the initial action (braking for a traffic light/stop sign, changing lanes and/or accepting a gap), while the FV responds to this action by braking. For rear-end interactions this can be expressed as:
DRAC RVt +1 =
REAR

(V FV ,t VLV ,t ) 2 ( X LV ,t X FV ,t ) LLV ,t

(1)

where, t = time interval X = position of the vehicles L = vehicle length V = velocity DRAC is updated every time interval (0.1 sec) based on driver reaction in the previous interval. Archer (3) suggests that a given vehicle is in traffic conflict if its DRAC exceeds a threshold braking value of 3.35 m/s2.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

11

3.

Crash Potential Index (CPI)

Several researchers have argued that the conventional DRAC measure fails to accurately reflect traffic conflicts, since it does not consider the vehicles braking capability (or Maximum Available Deceleration Rate) for prevailing road and traffic conditions. To address this concern, Cunto and Saccomanno (4, 11) introduced a Crash Potential Index (CPI) expressed in terms of the probability that a vehicle DRAC exceeds its assigned braking capability of MADR. Both DRAC and MADR are estimated for each FV as it progresses along its path in 0.1 sec increments. On the basis of the CPI measure, a vehicle is assumed to be in traffic conflict if its DRAC at a specific point in time exceeds its assigned MADR.

MEASURING SAFETY PERFORMANCE AT THE VIALE COSMAI ROUNDABOUT


In this paper safety performance is measured based on the three indicators (TTC, DRAC and CPI). For CPI, two inputs for MADR are considered (MADR1 and MADR2): 1. MADR1 is obtained assuming that the maximum deceleration rate depends only on pavement surface condition (wet or dry), such that:
(2)

MADR = g ( f a i ) where, g = gravity acceleration (9.81 m/sec2) fa= coefficient of friction (0.40) i = cross grade (0 %)

These values, that reflect the real conditions observed on the roundabout during the survey, derive from the application of Italian rules, according with the Decreto Ministeriale 05/11/2001 (12); 2. MADR2 is a random variable following normal distribution truncated to minimum and maximum values of 4.2m/s2 and 12.7m/s2, according to a previous research (4).

Tables 1, 2 and 3 indicate three measures of safety performance (TTC, DRAC and CPI) for all vehicles trajectories along the roundabout. For each measure, the number and percentage of vehicles in conflict and the average exposure to conflict are obtained for each trajectory of interest. For CPI, two MADR values were used as defined above.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

12

TABLE 1 Vehicles in conflict and average exposure to conflict based on TTC and DRAC values for all paths analyzed
Path Total 0-84 0-326 0-414 101-84 101-326 201-84 201-414 # Veh. 58 16 8 9 2 6 1 16 # veh. in conf. 23 6 4 4 1 1 1 6 TTC % veh. in confl. 39.66 37.50 50.00 44.44 50.00 16.67 100.00 37.50 Aver. Exp. (veh.sec) 1.81 1.67 2.65 1.00 0.50 1.10 3.10 2.05 # veh. in conf. 21 5 4 3 0 1 1 7 DRAC % veh. in confl. 36.21 16.00 50.00 33.33 0.00 16.67 100.00 43.75 Aver. Exp. (veh.sec) 0.92 0.84 0.88 0.80 0.00 0.80 1.10 1.06

TABLE 2 CPI calculated with MADR1 for all paths analyzed


Path Total 0-84 0-326 0-414 101-84 101-326 201-84 201-414 # Veh. 58 16 8 9 2 6 1 16 # veh. interacting 19 5 4 3 0 1 1 5 % veh. interacting 32.76 31.25 50.00 33.33 0.00 16.67 100.00 31.25 CPI Aver. Exp. (veh.sec) 0.82 0.68 0.65 0.67 0.00 0.70 0.70 1.24 CPI/veh. 0.045 0.031 0.028 0.026 0.000 0.058 0.055 0.081 CPI 85th 0.072 0.051 0.038 0.037 0.000 0.058 0.055 0.151

TABLE 3 CPI calculated with MADR2 for all paths analyzed


Path Total 0-84 0-326 0-414 101-84 101-326 201-84 201-414 # Veh. 58 16 8 9 2 6 1 16 # veh. interacting 16 4 4 3 0 1 1 3 % veh. interacting 27.59 25.00 50.00 33.33 0.00 16.67 100.00 18.75 CPI Aver. Exp. (veh.sec) 0.38 0.40 0.28 0.40 0.00 0.40 0.10 0.57 CPI/veh. 0.020 0.018 0.011 0.016 0.000 0.033 0.008 0.038 CPI 85th 0.033 0.029 0.016 0.022 0.000 0.033 0.008 0.052

A more in-depth analysis was carried out to highlight specific areas on the roundabout where measures of safety performance (summed over all vehicles and paths) reflect higher potential risks. This analysis is based on the hypothesis that high risk CPI values takes place when FV LV spacing, relative speeds and deceleration yield FV DRAC that exceeds its MADR for a given time increment. When summed over all FV vehicles and paths, a mapping of high risk behaviour is obtained for different areas of the roundabout.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

13

Figures 7 to 10 illustrate the areas or zones of higher risk associated with the Viale Cosmai roundabout, as expressed by the three safety performance measures (TTC, DRAC and CPI), respectively. For CPI, two estimates of braking capability were considered: MADR1 and MADR2. The shading is an indication of individual vehicle contributions to safety performance for all paths considered, i.e. the darker the shading the higher the risk. The assumed risk thresholds used in this analysis are given in the legends. It is evident from this analysis (tables 1 to 3 and figures 7 to 10) that vehicles proceedings along path 201-414 exhibit higher safety performance values (risk) than for the other paths. This could be due to the presence of vehicles on this trajectory that will have to yield precedence to vehicles on the ring (at detector No. 217) and then progress through the merging/diverging maneuver. Here these vehicles are exposed to abrupt changes in speeds, deceleration/acceleration rates and associated turbulence. The result is higher risk of crashes.

FIGURE 7 Potential conflict areas according with the TTC.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

14

FIGURE 8 Potential conflict areas according with the DRAC.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

15

FIGURE 9 Potential conflict areas according with the CPI (MADR1).

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

16

FIGURE 10 Potential conflict areas according with the CPI (MADR2).


It should be noted that along path 0-84, where the same number of vehicles are involved as per path 201-414, the value of CPI/veh is moderately lower, despite reflecting similar areas of high risk on the roundabout. It is reasonable to suggest that the vehicles travelling along path 0-84 are exposed to risk over a longer period of time than vehicles travelling along path 201-414 for the same volume. This translates to a lower CPI per unit time for the time interval a vehicle progresses through the roundabout. A comparison of CPI measures for the two MADR input values suggests higher risk profiles result from the application of the Italian standard MADR1 and the areas covered by elevated risk values is also more extensive for these Italian standards. In general the two MADR inputs into CPI consistently point to safety problems in similar locations of the roundabout (i.e. corresponding to major entry/exit and merging points). Higher risks are also associated along the roundabout ring in the area preceding the exiting the westbound direction.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

17

The other measures of safety performance (TTC and DRAC) produce results that tend to mirror those obtained from CPI. These measures tend to over-estimate the percentage of vehicles in conflict and the time exposed to conflict relative to those obtained from CPI, and they yield high risk areas that are more extensive than CPI-based measures. This was not entirely unexpected since CPI takes into account braking capability, whereas TTC and DRAC are estimated solely on the basis of differential speeds and spacing associated with each vehicle pair.

CONCLUSIONS
The experimental procedure presented in this paper has provided a low cost means for extracting vehicle tracking data from videotaped images. The vehicle tracking data provides observational information for input into measures of safety performance at a given location. These measures are sensitive to both fixed geometric attributes as well as real-time changes in driving behaviour and traffic conditions. A number of measures of safety performance have been explored in this study. While in general, all measures have yielded consistent insights into potential safety problems on the case study roundabout, some differences were observed. While all measures point to higher potential risks on the roundabout at or near major entry/exit points along the ring, the extent of the risk zone varies with the safety indicator. The CPI-based measures yield more confined risk zones. Risk areas reflect locations with higher than normal vehicle interactions, more frequent braking and speed changes, and hence more turbulent traffic conditions. The result is an increased potential for crashes. The Viale Cosmai case study has highlighted a number of potential safety problems for different geometric features along the ring. The highest potential for crashes is associated with areas of the roundabout where the incidents of entry, exit and merging maneuvers are high. Given the higher number of vehicle interactions at these locations we would expect more abrupt speed changes and increased traffic turbulence, and hence, increased potential for crashes. In estimating safety performance some indicators require the specification of more inputs than others. In this study, CPI requires specification of both DRAC and MADR (i.e. braking capability per vehicle). This measure yields a more definitive region of higher risk along the roundabout. For the same CPI-based measure the use of the different braking norms (MADR1 and MADR2) resulted in different areas of high risk on the roundabout. Despite these differences, however, the case study application has yielded remarkably consistent results with respect to potential safety problems inherent on the roundabout for different geometric and traffic conditions.

ACKNOWLEDGMENTS
We would like to thank Prof. Demetrio Carmine Festa for his encouragement and valuable comments on this study and Prof. Gian Franco Capiluppi, Ing . Rosolino Vaiana and Ing. Vincenzo Gallelli for their support in the data gathering stage. We are also grateful to Ing. Flavio Cunto for his comments and suggestions.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

Guido, Saccomanno, Astarita and Vitale

18

REFERENCES
1. Saunier, N. and T. Sayed. Automated Road Safety Analysis Using Video Data. In Transportation Research Record, No. 2019, Transportation Research Board, Washington, D.C., 2007, pp 57-64. 2. Abdel-Aty, M.A., J. Keller and P.A. Brady. Analysis of the types of crashes at signalized intersections using complete crash data and tree-based regression. In Transportation Research Record: Journal of the Transportation Research Board, No. 1908, Transportation Research Board of the National Academies, Washington, D.C., 2005, pp. 37-45. 3. Archer, J. Methods for the assessment and prediction of traffic safety at urban intersection and their application in micro-simulation modelling, PhD Thesis, Department of Infrastructure, Royal Institute of Technology, Sweden, 2005. 4. Cunto, F., F.F. Saccomanno and D. Duong. Comparison of Simulated Safety Performance to Observed Crash Occurrence. Submitted to Transportation Research Board Annual Meeting, Washington, D.C., 2008. 5. Hayward, J. Near misses as a measure of safety at urban intersections. PhD Thesis, Department of Civil Engineering. The Pennsylvania State University, Pennsylvania, 1971. 6. Minderhoud, M. and P. Bovy. Extended time to collision measures for road traffic safety assessment. Accident Analysis and Prevention, Vol. 33, 2001, pp 89-97. 7. Gettman, D. and L. Head. Surrogate Safety Measures from Traffic Simulation Models. Federal Highway Administration report RD-03-050. McLean, VA, 2003. 8. Huguenin, F., A. Torday and A. Dumont. Evaluation of traffic safety using microsimulation. In Proceedings of the 5th Swiss Transport Research Conference STRC. Ascona, 2005.
9. van der Horst, A. R. A. Time-to-collision as a Cue for Decision-making in Braking. In I. D. Brown, C.M. Haselgrave, I. Moorhead, & S. Taylor (Eds.), Vision in vehicles: Vol. 3, 1991, pp. 19-26.

10. Almquist, S., C. Hyden and R. Risser. Use of speed limiters in cars for increased safety and a better environment. In Transportation Research Record: Journal of the Transportation Research Board, No. 1318, Transportation Research Board of the National Academies, Washington, D.C., 1991, pp. 3439. 11. Cunto, F., and F.F. Saccomanno. Microlevel traffic simulation method for assessing crash potential at intersections. In Proceedings of the 86th Annual Meeting of the Transportation Research Board, Washington D.C., 2007. 12. Norme funzionali e geometriche per la progettazione delle strade, Decreto Ministeriale 05/11/2001.

TRB 2009 Annual Meeting CD-ROM

Paper revised from original submittal.

You might also like