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

Failure-cause sed Traffic Conflicts s – Theory, Applications s and Validation

Vision zero for traffic fatalities and serious injuries – research questions and challenges 32ndICTCT Conference in Warsaw, Poland 24– 25October 2019 Andrew Tarko, PhD Professor of Civil Engineering Director of Center for Road Safety Purdue University, Lyles School of Civil Engineering West Lafayette, Indiana, USA

Primer to Traffic Conflicts

(Near-crash Events)

• https://www.youtube.com/watch?v=wo_u5Ncv-Ko (right angle)
• https://www.youtube.com/watch?v=HIRKH6MtY2Y (side swipe)
• https://www.youtube.com/watch?v=XLbfJ3ocjnA (rear end)
• https://www.youtube.com/watch?v=8SeVc3itItI (pedestrian)

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

• Understand the mechanism of crash occurrence
• Propose a conflict model that includes the causal

mechanism of unobserved crashes

• Estimate the model using conflicts data
• Validate the model

Connecting Crashes with Conflicts

Counterfactual A Approach

The Road to Counterfactual Traffic Conflict cts Method

1964 Error as a necessary condition of a traffic conflict. Klebelsberg, D., Derzeitiger Sand der Verhaltensanalyse des Kraftfahrens. Zrbeit und Leitsung.

• Ablt. Arbeitswissenscaft soziale

betriebspraxis vol. 18, 33–37. 1980 Idea of probabilistic continuity of safety-related events - Glauz, W. D., D.J.

• 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

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 benefits of active safety technologies. UMTRI-96-18 Final Report. University of Michigan Transportation Research Institute, Ann Arbor, MI. 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. 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. Accident Analysis and Prevention, 43, 1907-1919. 2018 Theory of failure-based conflicts and practical method of estimating crash frequency

• Tarko, A. P. Estimating the expected number of crashes with traffic conflicts and the Lomax

Distribution – A theoretical and numerical exploration. Accident Analysis and Prevention,

• Vol. 113, pp. 63–73.

Frequentist or Counterfactual Approach?

Traffic Encounters Road Users Nearness Crashes EV Distr. of Road User Nearness Crash Probability

Black-box Data EV Regression Model

Statistical extrapolation of the observable events. The causal character of the estimated regression relationship is an open question.

Observable

Counterfactual Persp spective

Factual Counterfactual Failure No recovery Crash Recovery No crash No recovery Crash

Data Generating Mechanism Crash Generating Model

Remarks: (1)Failures make encounters (conflicts) and crashes etiologically consistent. (2)Failure generates severe conflicts. (3)Past research confirmed correlation between severe conflicts and crashes.

Evasive maneuver begins Time t Time of potential crash End of successful response Time of response

Traffic c Conflict Conce cept

Instantaneous Time to Collision (ITTC) vs. Time to

• Col
• llision (TTC)

Time to collision (TTC) at response time

Conflicts s and Crashes in Heterogeneous s Conditions

Consistent with the Lomax distribution

P(C|N)

Theory Implications

QN

Estimating k from Observed Conflicts

(Single Parameter Estimation, Tarko, 2018)

Probability of Crash

(Tarko, 2018)

Road Departures in Driving Simulator

Experiments in a Driving Simulator

Road length = 27 miles Outside Shoulders = 2 ft Driving Lanes = 12 ft Inside Shoulders = 4 ft Median = 36 ft Small distance ⇒ Near departure Track of the front right tire

Lane or Road Departure

Risk of departure can be analyzed through near-departure events within the framework of traffic conflicts. Traffic Conflict Near-departure TTC Time to Collision DTD Distance to Departure TTCo Threshold Time to Collision DTDo Threshold Distance to Departure dr = TTCo-TTC Delay Time dr = DTDo-DTD Delay Distance c = 2-k expected number of crashes c = 2-k Expected number of departures k estimated based on observed TTCs k estimated based on observed DTDs

Near-departure Ev Event

Path of the front right tire Right edge

• f

pavement Direction

• f travel

Shoulder

Traffic lane

Lateral Dist (ft) Time t (s)

Near-departure Ev Events s Analysi sis

Lateral Distance (ft) Time (s) Potential departure

10 runs

DTD* (ft)

Example Esti timati tion of the Departure Risk

Subject 219, 10 runs, 270 miles

0.029 Distance travelled between departures = 1/0.029 x 270 = 9,310 miles

near-departures

Take-away

• The results closely followed the anticipated trends prompted by the

theory - flattening trend of the expected number of crashes for sufficiently small threshold separations.

• The estimates were stable even for a relatively small number of

claimed conflicts (confirms efficiency).

• 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

period are needed.

Right-angle Collisi sions s – The Lesso son Learned 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 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 7 87933 Wednesday, April 02, 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

Time t Response delay Time of potential crash Remaining Time Time to conflict point Time to collision Counterfactual time to collision Envelope time to conflict point TTC is a pre-crash nearness consistent with the counterfactual concept while PET is a post-event nearness when the probability of crash is zero. Nevertheless, PET is an acceptable alternative to TTC if: (1)evasive maneuvers are similar (braking at similar rates), and (2)only considerably different are response delays. Then, PET-based and TTC-based response delays are approximately linearly dependent and the Lomax distribution is applicable to PET data.

TTC vs. PET Right-angle PETs Observations s (2003)

• Intersections videotaped from elevated

position

• Manual extraction of PET values frame by

frame – good quality

• Field of view did not include the approach

areas

• All PETs were measured without possibility
• f observing evasion maneuvers
• In fact, evasive maneuvers were not

considered important (Extreme Value approach)

4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 PET Threshold (s) Count

PET-based Results for Site 87909

PET ET-base sed Resu sults s for si site 87930

PET-based Results for Site 87933 (traffic co congestion)

Removing three very short PETs has changed the results considerably.

Correlation between Predicted and Reported Crash shes

Predicted and reported crashes correlated Overestimation ratio = 20

Lesso sons s Learned

• Evasion actions should be detected to claim traffic conflicts.
• An alternative Lomax data generation process – random arrivals with

possibly violated red signals.

• TTC and ITTC should be used instead of PET.
• PET threshold of six seconds was too large.
• Traffic encounters at low speeds do not include hazard sufficient to be

claimed conflicts.

• The observation period should be several time longer to provide sufficient

number of observations.

• Modern object tracking allows data collection in extended periods.
• Standardized selection of appropriate separation threshold is needed.

Validation with SHRP2 rear-end Conflicts Data SHRP2 and Rear-end Conflicts s Data

• Strategic Highway Research Program 2 (SHRP2) between 2006 and 2015
• Naturalistic driving study – routine use of instrumented vehicles by regular

drivers

• Nearly 5.5 million trips made by 3,400 participant drivers
• Over 4,300 total years of naturalistic driving time between 2010 and 2013
• 1,549 validated crashes
• 234 were rear-end collisions including
• 119 rear-end crashes SHRP2 drivers

following other vehicle

• Traffic conflict data come from

randomly selected 1.7% trips analyzed

Validation with SHRP2 rear-end Conflicts Data

ITTC Strong Jerk

• Beg. of Strong

Braking End of Evasion Conflict

ITTC values represented the crash nearness instead if TTC when estimating the k parameter (approximation).

Lomax analysis results

Female drivers 45-64 Male drivers 45-64 Male drivers 16-25

Driver Type Based on Crashes in All Trips Based on Conflicts in 1.7% Trips Mean 90% Conf. Interval Bootstrap Mean 90% Conf. Interval Lower Limit Upper Limit Lower Limit Upper Limit Males 16-25 1086 749 1475 1008 251 2187 Males 45-64 456 198 799 402 91 936 Females 45-64 137 25 325 95 13 259

Validation with SHRP2 rear-end Conflicts Data

The differences in the mean estimates are statistically insignificant.

Estimated rear-end crash rates per 100 million miles of following another vehicles

Technology Traffic Scanner TScan – a LiDAR sy syst stem for intersections

Research Unit Prototype

Demonstration

• f the Results

39

Guidelines for Estimating Safety with Traffic Conflicts

by Purdue Center for Road Safety

Funded by Federal Highway Administration and Indiana Department of Transportation through Joint Transportation Research Program Expected in 2022 Expected on Nov 8, 2019

Closure

• Failure present during an encounter is the necessary condition of

the results validity

• Multiple measures of crash nearness that meet the condition of

reasonableness are applicable (for example, distance, ITTC, TTC)

• Needed further research on detecting evasion maneuvers reliably

and on predicting crash outcome severity from observed conflicts

• Pedestrian traffic conflicts pose the biggest challenge
• Estimation of individual safety effects needed
• Recent progress observation technology, and estimation methods

makes conflicts practical

Thank you