To address the problem of the lack of interpretability of vehicle trajectory prediction models based on deep learning, this paper proposes a Fusion Neural network with the Spatio-Temporal Attention (STA-FNet) model. The model outputs a predictive distribution of future vehicle trajectories based on different vehicle trajectories and traffic environment factors, with an in-depth analysis of the Spatio-temporal attention weights learned from various urban road traffic scenarios. In this paper, the proposed model is evaluated using the publicly available NGSIM dataset, and the experimental results show that the model not only explains the influence of historical trajectories and road traffic environment on the target vehicle trajectories but also obtains better prediction results in complex traffic environments.
Vehicle Trajectory Prediction Model Based on Fusion Neural Network
Lect.Notes Social.Inform.
International Conference on Communications and Networking in China ; 2022 November 19, 2022 - November 20, 2022
10.06.2023
12 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Vehicle trajectory prediction based on LSTM network
IEEE | 2022
|Graph and Recurrent Neural Network-based Vehicle Trajectory Prediction For Highway Driving
ArXiv | 2021
|Vehicle trajectory prediction method based on time-space fusion attention
Europäisches Patentamt | 2025
|