Vehicle Trajectory Prediction in Expressway Merging Areas Based on Self-Supervised Mechanism

Abstract

With the increasing number of vehicles, the growing complexity of traffic environments has led to a rise in traffic pressure. As a critical hub connecting roads between cities, expressway traffic safety cannot be ignored. Merging areas, due to their complex road configuration, have become major accident-prone spots on expressways. Improving driving safety in expressway merging areas is of the utmost importance. Predicting the future trajectory of a vehicle can then be used to avoid traffic conflicts or even crashes by using methods such as future trajectories and active control. Therefore, the emergence of trajectory prediction techniques provides new approaches for intelligent traffic management in these areas. First, this paper takes a global perspective from roadside light detection and ranging (LiDAR) sensors to construct a vehicle trajectory database in the merging area, relying on object detection and trajectory tracking technologies. Then, a vehicle trajectory prediction model based on a self-supervised mechanism is developed, specifically designed for the complex interactive environment of expressway merging areas. Finally, four models, long short-term memory (LSTM), social LSTM (SL), convolutional social LSTM (CSL), and maneuver-aware pooling (MAP), are compared with the proposed model. The evaluation is based on the root mean square error (RMSE) metric for overall, left-turn, right-turn, straight, and merging trajectory accuracies and the Acc metric for horizontal and vertical intention accuracies. Experimental results demonstrate that the proposed model achieves lower errors and higher prediction accuracy in both trajectory prediction and lateral and longitudinal intention prediction.