Abstract: Trajectory prediction is a hot topic in the field of computer vision and has a wide range of applications. Trajectory prediction refers to predicting the future trajectory of a target based on its past trajectory. This paper proposes a method based on graph neural network and attention mechanism, in order to update trajectory characteristics by implement global pedestrian interaction. And, a direct relationship between history and future is introduced with the attention module for reducing error propagation. The method was evaluated on several real-world crowd datasets, the results demonstrate the effectiveness of our method.
Authors: Zhe Liu (University Of Electronic Science And Technology Of China, China); Lizong Zhang (University of Electronic Science and Technology of China, China); ZhiHong Rao (China Electronic Technology Cyber Security Co., Ltd & Cyberspace Security Key Laboratory of Sichuan Province, China); Guisong Liu (UESTC, China)
Email: 201852080833@std.uestc.edu.cn, l.zhang@uestc.edu.cn, 13608184980@139.com, lgs@uestc.edu.cn