|

Predicting Crashes with Safe Systems Surrogates Obtained from Video Analytics’ Implications for Evaluation of Vision Zero Safety Treatments

Author(s): Jafari Anarkooli, Milligan, Penner, Persaud, Saleem
Student Paper Competition: 2nd Place | Dr. Charles H. Miller Award

Slidedeck Presentation:

7C_Jafari_Anarkooli

Abstract:

Background/Context: Fundamental to credible vision zero programs is the evaluation of implemented safety treatments, especially for innovative treatments for which there is no prior, reliable information on safety effects. This can be challenging to accomplish with crash-based evaluation for treatments targeted at rare crash types as well as for cases where the knowledge is required before sufficient crashes materialize to detect statically significant effects.

Aims/Objectives: The paper aims to address this challenge for treatments at urban signalized intersections by providing a methodology that uses surrogate measures of safety obtained from video analytics. While the paper focuses on the challenge described, a secondary aim is to cast light on all related methodological challenges in evaluating vision zero safety treatments.

Methods/Targets: The approach applies the concept of a safe systems surrogate, which is a safety performance indicator that quantitatively incorporates the human body's limited ability to withstand force in its assessment of risk severity. To develop this approach, vehicle-vehicle traffic conflicts based on post encroachment times, along with corresponding conflicting vehicle speeds, are first measured from video observations at signalized intersections in York Region, Ontario and Winnipeg, Manitoba.

These measures were derived from road user trajectories that were automatically extracted using video analytics software, which detects and classifies objects according to user class. In this, a spatial holography and scaling process is used to obtain the spatial position and speeds of road users. Statistical models are then estimated using generalized linear modeling to relate left turn opposed crashes at the same intersections to the corresponding conflicts that are classified by severity using the speeds of conflicting vehicles.

Results/Activities: The modeling results indicate that those conflicts classified as severe consistently increase the risk of crashes after controlling for left turning traffic volume. While the results for York Region and Winnipeg were in a general agreement in terms of the significant variables and the direction of the effect, they do imply that the classification of severe conflicts may need to reflect operational characteristics of the intersections, such as speed limit, to improve the precision of crash predictions.

Discussion/Deliverables: The models can be applied to estimate the change in crashes following a safety treatment by observing, through video analytics, the change in conflicts and speeds and using the crash-conflict-speed model. The premise of the approach is that a safety treatment may alter the frequency of conflicts and the speed of conflicting vehicles but will not change the relationship between those variables and corresponding crashes.

Conclusions: The methodological approach is viable for quickly evaluating all vision zero treatments and, in particular, innovative ones for which knowledge on safety effects is sparse or non-existing. Currently the approach has only been developed for treatments targeting left turn opposed crashes at signalized intersections. It should and could be expanded for other crash and site types, including pedestrian crashes.