A Maneuver-Based Threat Assessment Strategy for Collision Avoidance

作者单位：Jilin University;General Motors LLC

发表期刊：SAE International Journal of Passenger Cars - Electronic and Electrical Systems, 2018

摘要： Advanced driver assistance systems (ADAS) are being developed for more and more complicated application scenarios, which often require more predictive strategies with better understanding of driving environment. Taking traffic vehicles’ maneuvers into account can greatly expand the beforehand time span for danger awareness. This paper presents a maneuver-based strategy to vehicle collision threat assessment. First, a maneuver-based trajectory prediction model (MTPM) is built, in which near-future trajectories of ego vehicle and traffic vehicles are estimated with the combination of vehicle’s maneuvers and kinematic models that correspond to every maneuver. The most probable maneuvers of ego vehicle and each traffic vehicles are modeled and inferred via Hidden Markov Models with mixture of Gaussians outputs (GMHMM). Based on the inferred maneuvers, trajectory sets consisting of vehicles’ position and motion states are predicted by kinematic models. Subsequently, time to collision (TTC) is calculated in a strategy of employing collision detection at every predicted trajectory instance. For this purpose, safe areas via bounding boxes are applied on every vehicle, and Separating Axis Theorem (SAT) is applied for collision prediction, so that TTC can be calculated efficiently and accurately. Finally, a threat level index based on reverse TTC is used to quantize the threat degree of every traffic vehicle potential collision to the ego vehicle. Experimental data collected in field test are used in the model training, and the overall strategy is validated under PanoSim. Simulation results show that MTPM can accurately identify maneuvers such that the effective prediction on trajectories can be generated. TTC and threat index can be calculated timely. The proposed threat assessment strategy can not only assist collision avoidance systems to foresee dangerous situations, but also eliminate false alarm to certain extent.