Vehicle Trajectory Prediction with Driving Style Identification and Intention Fusion

Precise trajectory prediction is crucial to the decision-making of autonomous vehicles, especially the surrounding vehicle performs a lane-changing maneuver. However, the vehicle trajectory is affected by driving intentions and diverse driving styles, which makes it difficult to be predicted accurately. In this study, a vehicle trajectory prediction method considering the driving intentions and driving styles is proposed. Firstly, the Dynamic Bayesian Network (DBN) is used to model the generation of vehicle lane-changing intention. It predicts the probability of lane-changing maneuver. Furthermore, considering the diversity of driving styles, an online identification method for driving styles is proposed by combining the K-means and Bayesian filtering. Moreover, the Long short-term memory (LSTM) encoder-decoder is used as the trajectory prediction network. Vehicle's historical trajectories, as well as the identification results of lane-changing intentions and driving styles, are integrated as the input of the trajectory prediction network. The proposed method is verified through making prediction 14s in advance. The results show that the proposed method has significant advantages compared to the baseline algorithms.