Modeling collision risk for unsafe lane-changing behavior: A lane-changing risk index approach

Abstract

The lane-changing maneuvers are challenging and contributes to traffic accidents and crashes. They are complicated task that often lead to an increased risk of vehicle collisions and deteriorates traffic flow. This paper proposes a vehicle collision risk model based on vehicle lane-changing movement for autonomous vehicles using the probability threat assessment approach and the lane-changing risk index (LCRI). Given two vehicles, A and B, we first estimate the movements of vehicle B using a GPS device, which accurately identifies the vehicle while changing lanes. We then evaluate the collision risks between vehicles using the probabilistic assessment approach to determine whether a collision event occurs at a later stage. Then, we propose the LCRI, which aims to identify the risk level associated with vehicle lane-changing movement and determine the severity of the crash level. In this study, the HighD vehicle trajectory dataset is used to obtain accurate traffic information and help us investigating collision risks among vehicles. The results show that the proposed model can identify the risks of collision ahead of time. Furthermore, the random parameter ordered logit (RPOL) with heterogeneity model achieves better performance and provides a good fit model and measures for improving traffic safety.