Lane Change Trajectory Prediction based on Spatiotemporal Attention Mechanism
- Resource Type
- Conference
- Authors
- Yang, Shichun; Chen, Yuyi; Cao, Yaoguang; Wang, Rui; Shi, Runwu; Lu, Jiayi
- Source
- 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC) Intelligent Transportation Systems (ITSC), 2022 IEEE 25th International Conference on. :2366-2371 Oct, 2022
- Subject
- Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Computing and Processing
Engineering Profession
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Transportation
Decision making
Predictive models
Feature extraction
Mathematical models
Trajectory
Spatiotemporal phenomena
Safety
- Language
The motion intentions and future trajectories of traffic participants have great influences on the decision-making and path planning processes of autonomous vehicles. Lane change behaviors needs to be studied with accurate mathematical representation to realize long-term and reliable intention and trajectory prediction. Traditional studies have applied the specific probability model to perform prediction, but this model is limited by strict assumptions and constraints. With the development of deep learning methods, better prediction results have been realized through the introduction of data-driven concepts. In this study, we focused on the spatiotemporal interaction between the ego and surrounding vehicles by mining hidden trajectory features to effectively predict future lane change intentions and the trajectories of the vehicles surrounding an autonomous vehicle. We constructed spatiotemporal attention mechanism-based long short-term memory (LSTM) networks to perform lane change prediction within the future 5 s using the next generation simulation (NGSIM) dataset. The prediction results were represented in a certain trajectory form and were obtained using the regression fitting method. It was shown that the proposed model can accurately predict lane change behaviors within the future 5 s and provide new ideas for future lane change behavior prediction.