Considering the spatial interaction and vehicle history temporal features of approaching driving vehicles is a key step in decision making and trajectory planning for autonomous vehicles. In order to solve the problems of insufficient spatial-temporal feature extraction and multimodal vehicle trajectory prediction, a model for multimodal vehicle trajectory prediction integrating spatial and temporal features is proposed in this paper. First, a dynamic attention-based graphical neural network and a fused attention-based temporal convolutional network (TCN) are used to capture spatial interaction features as well as the remote temporal dependence of vehicles in the target region. Secondly, a gating unit for feature fusion is designed to achieve effective fusion of spatial-temporal features, and then a decoder is used to generate probability distributions of future vehicle trajectories in the target region. Finally, the proposed model is evaluated on the publicly available NGSIM dataset and compared with previous models. Simulation results show that the proposed model provides advanced improvements in both the prediction error and the probability distribution of the predicted future trajectories.
A multimodal vehicle trajectory prediction method with spatio-temporal feature fusion
2023-08-04
768335 byte
Conference paper
Electronic Resource
English
Deep spatio-temporal feature fusion with compact bilinear pooling for multimodal emotion recognition
British Library Online Contents | 2018
|SHIP TRAJECTORY FEATURE POINT EXTRACTION-BASED SPATIO-TEMPORAL DP METHOD
European Patent Office | 2022
|Air route network flow prediction method based on spatio-temporal feature fusion
European Patent Office | 2023
|