Collision Risk Assessment Based on Uncertainty of Vehicle Models

Abstract

This article proposes a vehicle collision risk assessment (VCRA) framework based on dual uncertainty, with the aim of improving the precision of collision risk prediction. Firstly, this study utilized models that considered the uncertainty of inputs such as road geometry, environmental factors, and driver operations, as well as the uncertainty of the vehicle model itself. Secondly, the model quantified the uncertainty of vehicle position using an extended Kalman filter and then determined whether to switch the vehicle model based on the weight of the yaw rate caused by the geometric characteristics of the road so that these models are more suitable for the current driving scenario. Thirdly, the collision risk of the predicted trajectory was evaluated in this article, and the collision risk was quantified into vehicle collision probability by Monte Carlo simulation. Meanwhile, this article defined some certain regions that were applied to help determine whether it was necessary to start the assessment algorithm. Finally, the study validated the effectiveness and accuracy of the method through three typical test scenarios specified by the China New Car Assessment Program (C-NCAP). Through comparison, the VCRA framework can accurately assess collision risk at time to collision (TTC) = 2.2 s, which is earlier than the collision risk output time of TTC = 1.7 s in C-NCAP (scenario b). The VCRA framework also provides a new upgrade approach for active safety test scenarios in regulations.