In the field of autonomous driving, vehicle trajectory prediction using information about the surrounding environment is crucial for improving system safety. Existing trajectory prediction models often fail to effectively capture temporal features under different speed conditions when dealing with historical trajectories with speed differences. In addition, these models rely excessively on the self-learning of the network when modeling inter-vehicle interactions and lack effective integration of expert knowledge. To address these problems, this article proposes a group vehicle trajectory prediction model based on dynamic view-distance enhancement. The model adopts a spatio-temporal decoupled modeling framework, in which the spatial branch combines the driver’s field of view angle and inter-vehicle distance information under different speed conditions to dynamically optimize inter-vehicle interactions, while the temporal branch extracts temporal information in successive and inter-timesteps through speed-adaptive temporal convolution and graph convolution. The experimental results show that the model in this paper significantly improves the accuracy of trajectory prediction on NGSIM and highD datasets with a small number of parameters.
Li et al. (Wed,) studied this question.