Trajectory Prediction for Turning Vehicles at Intersections by Fusing Vehicle Dynamics and Driver’s Future Input Estimation

Abstract

The accuracy of vehicle trajectory prediction is a key factor that has a considerable effect on the reliability of collision warning systems. Most previous trajectory prediction studies relied mainly on vehicle dynamics and led to some prediction errors because of the lack of information about the driver’s future maneuvers, especially when a vehicle made turns. This paper reports a novel trajectory prediction approach for predicting vehicle movements at intersections; the approach relies on fusing vehicle dynamics and estimation of the driver’s future maneuvers. The current vehicle state was estimated by a double Kalman filter consisting of a yaw angle Kalman filter and a position Kalman filter. The estimation results were then used as the initial inputs to the trajectory prediction loop that was iterated over the prediction horizon. The kinematic parameters (e.g., velocity and yaw angle) used for the prediction at the next time step were corrected by the driver’s future operation inferred from the driver’s desired trajectory. The desired trajectory was in turn determined from prior information about road geometry and the particular driver’s driving behavior. Results show that the prediction accuracy of this proposed approach is noticeably improved compared with common trajectory prediction methods, especially for turning vehicles at intersections. The superiority of the proposed method can help improve the reliability of collision warning systems.