Vehicle motion trajectory prediction fusion algorithm with driver adventurousness correction factor based on CS-LSTM

Abstract

Predicting the trajectories of adjacent vehicles plays an important role in the driving safety of adaptive cruise control system. It affects the safety and stability of the vehicle following the target vehicle during the vehicle cruising driving vehicle. However, due to the uncertainty of vehicle dynamics, driver character, and the complexity of the surrounding environment, vehicle trajectory prediction faces great challenges. Hence, a dynamic vehicle trajectory prediction system is proposed based on identifying driver intentions. First, based on a convolution LSTM, the driver adventurousness factor is introduced to describe the driver’s lane-change behavior heterogeneity and improve the accuracy of long-term lane-change trajectory prediction of adjacent lane vehicles. Second, the trajectory prototype predicted trajectory is updated by adjusting the minimum value function until the vehicle model corresponds to the planned sampling trajectory to improve the accuracy of the adjacent lane vehicle’s short-term lane-change trajectory prediction. Finally, the trajectories are fused using the trigonometric fusion algorithm, and the optimal trajectory is the output. The suggested strategy can predict lane-change intentions 2–5 s in advance. The prediction accuracy of the lane-change trajectory was approximately 21% higher than the normal prediction outcomes. The proposed method can be used to improve passenger comfort and the stability of a vehicle following a target vehicle that is separated from the adjacent lane vehicle.