A Novel Algorithm for Crash Detection Under General Road Scenes Using Crash Probabilities and an Interactive Multiple Model Particle Filter

Abstract

Driver inattention causes the majority of vehicular crashes, and these accidents produce extensive economic and social costs, as well as injuries and fatalities. Thus, the development of imminent crash detection systems is one of the most important issues in automotive safety. Various crash detection algorithms have been proposed, but the coverage of these algorithms has been limited to one or two crash scenarios. To widen the coverage of crash detection systems to include various crash modes, driver behaviors that are dependent on road scenes and vehicle dynamics should be considered. This paper proposed an algorithm for detecting an imminent collision in general road scenes. The proposed algorithm consists of crash probability data generated from Monte Carlo simulations that consider driver behavior and vehicle dynamics, a tracking algorithm that uses an interactive multiple-model particle filter, and a threat assessment algorithm that estimates crash probabilities. To reduce nuisance and false-positive alarms, the algorithm discriminated between normal and dangerous road scenes, and a point of no return was detected using three driver models that addressed different levels of driver input. The performance of the proposed algorithm was evaluated under three scenarios, and it successfully discriminated between collision and near-miss cases, and it adjusted warning times depending on the road scenes. It is expected that the proposed algorithm would have good driver acceptability based on the results of the near-miss cases. The proposed algorithm can be used as an integrated crash detection algorithm for crash warning, avoidance, and mitigation purposes while incorporating tracking information from multiple sources.