Hierarchical Vehicle Trajectory Prediction with Interactive Multiple Models

In order to enhance the capability to avoid potential collisions, it is essential to accurately predict the future trajectory of autonomous vehicles. In this work, a novel hierarchical vehicle trajectory prediction framework is proposed to promote the safety of autonomous driving. First, to effectively improve the prediction accuracy and environmental adaptability of the model, an interactive maneuver-based trajectory prediction module is established. The vehicle maneuver is estimated by dynamic Bayesian network, and then the most appropriate prediction model will be selected for model interaction in terms of the real-time maneuver of the vehicle. Second, to optimize the compensation of the results, the residual correction module is constructed by utilizing the long and short-term memory network, where the predicted trajectories are used as inputs to the module to estimate the prediction residuals. Finally, this study conducts model validation on the public NGSIM dataset, and the results show that the proposed method has better prediction performance.