Primer to Traffic Conflicts Failure-cause sed Traffic Conflicts s - PDF document

Primer to Traffic Conflicts Failure-cause sed Traffic Conflicts s – (Near-crash Events) Theory, Applications s and Validation • https://www.youtube.com/watch?v=wo_u5Ncv-Ko (right angle) • https://www.youtube.com/watch?v=HIRKH6MtY2Y (side swipe) Andrew Tarko, PhD Professor of Civil Engineering • https://www.youtube.com/watch?v=XLbfJ3ocjnA (rear end) Director of Center for Road Safety Purdue University, Lyles School of Civil Engineering • https://www.youtube.com/watch?v=8SeVc3itItI (pedestrian) West Lafayette, Indiana, USA Vision zero for traffic fatalities and serious injuries – research questions and challenges 32 nd ICTCT Conference in Warsaw, Poland 24– 25October 2019 Connecting Crashes with Conflicts Frequentist st A Approach • Observe traffic conflicts in relatively short periods on multiple roads • Use corresponding crashes reported on the same roads in similar conditions in long periods • Calculate crash-conflict ratio, or • Estimate a crash count regression model that includes traffic conflicts Causality not considered Conditions in the two periods different Underreporting crashes by police Transferability of the ratio questionable Connecting Crashes with Conflicts The Road to Counterfactual Traffic Conflict cts Method Counterfactual A Approach 1964 Error as a necessary condition of a • Understand the mechanism of crash occurrence traffic conflict. Klebelsberg, D., Derzeitiger Sand der Verhaltensanalyse • Propose a conflict model that includes the causal des Kraftfahrens . Zrbeit und Leitsung. Ablt. Arbeitswissenscaft soziale mechanism of unobserved crashes betriebspraxis vol. 18, 33–37. • Estimate the model using conflicts data 1980 Idea of probabilistic continuity of safety-related events - Glauz, W. D., D.J. • Validate the model Migletz. Application of Traffic Conflict Analysis at Intersections . NCHRP Report 219, Transportation Research Board, Washington D.C.

The Road to Counterfactual Traffic Conflict cts Method Frequentist or Counterfactual Approach? 1996 Application of the EV theory to evaluate the effect of in-vehicle technologies on safety - Campbell, K., Joksch, H. C., Green, P.E. A bridging analysis for estimating the Black-box Data benefits of active safety technologies . UMTRI-96-18 Final Report. University of Michigan Transportation Research Institute, Ann Arbor, MI. Traffic Encounters 2006 Application of EV theory to estimating the expected number of crashes - Songchitruksa, P., Tarko, A. The extreme value theory approach to safety estimation . Accident Analysis and Prevention, 28, 811–822. Road Users Crashes EV Distr. of Road User Nearness Crash Probability Nearness 2011 Discussion of counterfactual analysis of conflicts and crash causality - Davis, G. A, Hourdos J, Xiong H, Chatterjee I. Outline for a causal model of traffic conflicts and crashes . EV Regression Model Accident Analysis and Prevention, 43, 1907-1919. 2018 Theory of failure-based conflicts and practical method of estimating crash frequency Statistical extrapolation of the observable events. -Tarko, A. P. Estimating the expected number of crashes with traffic conflicts and the Lomax The causal character of the estimated regression relationship is an open question. Distribution – A theoretical and numerical exploration. Accident Analysis and Prevention, Vol. 113, pp. 63–73. Traffic c Conflict Conce cept Counterfactual Persp spective Instantaneous Time to Collision (ITTC) vs. Time to o Col ollision (TTC) Factual Counterfactual End of Data Generating Mechanism Evasive successful maneuver response begins Observable Crash Failure No recovery Crash Generating Model Time of No recovery Recovery potential crash Remarks: (1)Failures make encounters (conflicts) and Crash No crash crashes etiologically consistent. 0 Time t (2)Failure generates severe conflicts. Time of (3)Past research confirmed correlation between Time to collision (TTC) at response time response severe conflicts and crashes. Conflicts s and Crashes in Heterogeneous s Conditions Consistent with the Lomax distribution

Theory Implications Estimating k from Observed Conflicts (Single Parameter Estimation, Tarko, 2018) P(C|N ) Q N Probability of Crash (Tarko, 2018) Road Departures in Driving Simulator Experiments in Lane or Road Departure Road length = 27 miles Outside Shoulders = 2 ft a Driving Driving Lanes = 12 ft Inside Shoulders = 4 ft Risk of departure can be analyzed through near-departure events within the Simulator Median = 36 ft framework of traffic conflicts. Traffic Conflict Near-departure TTC Time to Collision DTD Distance to Departure TTC o Threshold Time to Collision DTD o Threshold Distance to Departure Small distance ⇒ d r = TTC o -TTC Delay Time d r = DTD o -DTD Delay Distance Near departure c = 2 -k expected number of crashes c = 2 -k Expected number of departures Track of the front right tire k estimated based on observed TTCs k estimated based on observed DTDs

Near-departure Ev Event Direction Lateral Dist (ft) Distance (ft) of travel Lateral Traffic Near-departure Path of the lane Ev Events s Analysi sis front right tire Potential departure Shoulder Time t (s) 0 Right edge Time (s) of pavement Example Esti timati tion of the Departure Risk Take-away Subject 219, 10 runs, 270 miles near-departures • The results closely followed the anticipated trends prompted by the theory - flattening trend of the expected number of crashes for 10 runs sufficiently small threshold separations. • The estimates were stable even for a relatively small number of claimed conflicts (confirms efficiency). 0.029 • Although the number of traffic conflicts carries safety information, the conditional probability of crash must be considered. • A conservatively small threshold separation and a longer observation DTD* (ft) period are needed. Distance travelled between departures = 1/0.029 x 270 = 9,310 miles Right-angle Collisi sions s – The Lesso son Learned Studied signalized intersections in 2003 West Lafayette and Lafayette, Indiana No. Site Date 8 hours of PET Observations 1 87905 Friday, June 13, 2003 0745-0845, 1000-1600, 1630-1730 Right-angle Collisi sions s – The Lesso son Learned 2 87906 Monday, June 16, 2003 0730-0830, 1000-1600, 1630-1730 3 87907 Thursday, May 22, 2003 0900-1000, 1000-1600, 1630-1730 4 87909 Wednesday, June 25, 2003 0800-0900, 1000-1600, 1630-1730 5 87915 Friday, April 11, 2003 0800-0900, 1000-1600, 1630-1730 6 87930 Wednesday, July 02, 2003 0730-0830, 1000-1600, 1630-1730 Wednesday, April 02, 7 87933 2003 0900-1000, 1000-1600, 1630-1730 8 97901 Tuesday, April 08, 2003 0900-1000, 1000-1600, 1630-1730 9 97903 Tuesday, April 29, 2003 0815-0915, 1000-1600, 1630-1730 10 97905 Monday, April 21, 2003 0745-0845, 1000-1600, 1630-1730 11 97911 Wednesday, May 21, 2003 0830-0930, 1000-1600, 1630-1730 12 97920 Tuesday, April 01, 2003 0745-0845, 1000-1600, 1630-1730

Right-angle PETs Observations s (2003) TTC vs. PET TTC is a pre-crash nearness consistent with the counterfactual concept while PET is a Remaining Time Time to post-event nearness when the probability of collision crash is zero. • Intersections videotaped from elevated Time to position Nevertheless, PET is an acceptable conflict Envelope time to • Manual extraction of PET values frame by point alternative to TTC if: conflict point Count frame – good quality Counterfactual (1)evasive maneuvers are similar (braking at time to collision • Field of view did not include the approach similar rates), and Time of areas potential crash (2)only considerably different are response • All PETs were measured without possibility delays. of observing evasion maneuvers Response • In fact, evasive maneuvers were not Then, PET-based and TTC-based response delay considered important (Extreme Value Time t delays are approximately linearly dependent approach) and the Lomax distribution is applicable to PET data. 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 PET Threshold (s) PET ET-base sed PET-based Resu sults s for Results for Site site 87930 si 87909 Correlation between Predicted and Reported Crash shes PET-based Results for Predicted and reported Site 87933 crashes correlated (traffic Overestimation ratio = 20 congestion) co Removing three very short PETs has changed the results considerably.