Non-stationary crash risk modelling of powered two-wheelers using extreme value analysis of surrogate crash events

Evaluating the impact of evasive actions such as braking and steering on the crash risk assessment of vehicles is a scarce endeavor due to the lack of relevant data. This study uses Extreme Value Theory to investigate and model the effect of evasive actions on the sideswipe crash risk of powered two-wheelers (PTWs) moving on multilane rural highways. The crash risk was projected from the observed sideswipe conflicts that were quantified using a surrogate safety indicator called anticipated collision time (ACT). The vehicle trajectory data extracted from traffic videos, collected using an unmanned aerial vehicle, was used as the input for the analysis. The data was denoised using a state-of-the-art trajectory reconstruction method called recursively ensembled low pass filtering. Once the conflicts were identified from the trajectory data, the crash risk models were developed considering five covariates: maximum deceleration rate, maximum yaw rate, and the times spent in decelerating, accelerating, and steering during a sideswipe conflict. These covariates were used to capture the non-stationarity in the traffic conflict extremes. The best performing non-stationary model was selected by comparing the negative log-likelihood values with the stationary-one. The findings suggest that the PTWs experience significant sideswipe crash risk on four-lane (crash risk 0.09%) and six-lane (crash risk 0.17%) highways. The sideswipe crash risk of PTWs increases with the increase in the intensity of braking and steering actions measured in terms of maximum deceleration and yaw rates. Further, this study emphasizes that incorporating the effects of evasive actions in the crash risk estimation and developing non-stationary models could significantly improve the pre-cision of crash frequency estimates. Based on the findings it can be concluded that for the safety improvement of PTWs on multilane highways, lane-restriction should be imposed which can increase the safety margin during sideswipe conflicts.

Automated summary

This study investigates the impact of evasive actions on the sideswipe crash risk of powered two-wheelers (PTWs) on multilane rural highways using Extreme Value Theory. The findings suggest that PTWs experience significant sideswipe crash risk on four-lane and six-lane highways, and the risk increases with the intensity of braking and steering actions. The study emphasizes the importance of incorporating evasive actions in crash risk estimation and developing non-stationary models for more accurate crash frequency estimates.