Peril Set-Aided Lane-Change Intention Inference on Highway

Understanding lane-changing (LC) intention on highways is a challenging task for satisfying the safety requirements and supporting the decision making of autonomous vehicles in dense traffic environments. We propose a peril set-aided LC intention inference model (PS-LCIIM) to accurately and timely recognize a driver’s LC intention before the vehicle crosses the line. To save computational resources of the onboard processor, a peril set-based vehicle selection strategy is designed to select the target vehicles based on collision risk and perception salience. Then, a dynamic time-series Bayesian network is introduced to model time-varying vehicle interactions. The real-time intention classification is conducted based on the outputs of a newly developed population evolutionary particle filter algorithm. A naturalistic vehicle trajectories data set is applied to train and validate the proposed model. The results demonstrate that the proposed model predicts the LC intention earlier with higher accuracy while maintaining a low computation cost.