Passing-yielding intention estimation during lane change conflict: A semantic-based Bayesian inference method

Intention estimation has been widely studied in lane change scenarios, which explains a vehicle's behaviour and implies its future motion. However, in dense traffic, lane-changing is more tactical and interactive. Due to the conflict between merging vehicles and adjacent vehicles, driving intentions become interdependent which fuses passing and yielding. In addition, lane change occurs without a fixed location. Drivers should be aware of each other's intentions along conflict process, and take instant responses. To address these challenges, this paper proposes semantic-based interactive intention estimation (SIIE), to understand driving intentions during lane change conflict. The problem is addressed by combining driving semantics with probability inference model based on dynamic Bayesian network (DBN). Firstly, the DBN is modelled for the interaction process with Condition-Intention-Behaviour relationships. Secondly, the semantics are extracted from the lane change conflict and are inferred with observation methods. Thirdly, SIIE is trained and verified with real-world driving data. The intention estimation results are demonstrated, and then utilized for multi-modal motion identification and trajectory prediction. Lane change in dense traffic requires interactive cognition of driving intentions, the findings of this research shall inspire future studies into related scenarios, and promote interactive driving technologies.