Intention-Enhanced Vehicle Trajectory Prediction for Autonomous Driving in Highway Scenarios

Trajectory prediction has always received widespread attention due to its important role in decision-making and path planning for autonomous driving. Most existing studies focus on improving the accuracy of trajectory prediction by changing the backbone neural network or embedding static scene information such as high-definition maps, while ignoring the use of signals such as turn signals that represent the vehicle′s movement intention. In this paper, we propose an intention-enhanced vehicle trajectory prediction method that improves the prediction performance by introducing lane-changing intention into the graph-based trajectory prediction model. In detail, we first embed the observed historical trajectories of the ego vehicle and surrounding vehicles into a scene graph and feed it into a graph convolutional network to extract the interaction features between vehicles. Then, the developed prediction module fully considers the lane-changing intention of the target vehicle to output a more reasonable predicted trajectory. Finally, we evaluate the performance of the proposed method on the public dataset HighD. Experimental results demonstrate that introducing lane-changing intention can effectively improve the performance of trajectory prediction.