文献类型：会议论文

英文题名：Spatial-Temporal Hypergraph Neural Network based on Attention Mechanism for Multi-view Data Action Recognition

作者：Wu, Zhixuan[1]; Ma, Nan[2,3]; Zhi, Tao[4]; Xu, Genbao[2,3]

第一作者：Wu, Zhixuan

机构：[1] Beijing Union University, Beijing Key Laboratory of Information Service Engineering, College of Robotics, Beijing, 100101, China; [2] Beijing University of Technology, Faculty of Information Technology, Beijing, 100124, China; [3] Beijing University of Technology, Engineering Research Center of Intelligence Perception and Autonomous Control, Ministry of Education, Beijing, 100124, China; [4] Yunji Technology, Beijing, 100081, China

第一机构：北京联合大学北京市信息服务工程重点实验室|北京联合大学机器人学院

通讯机构：[2]Beijing University of Technology, Faculty of Information Technology, Beijing, 100124, China

会议论文集：Proceedings - 2023 10th International Conference on Dependable Systems and Their Applications, DSA 2023

会议日期：August 10, 2023 - August 11, 2023

会议地点：Tokyo, Japan

语种：英文

外文关键词：action recognition; Attention mechaine; component; multi-view; spatial-temporal hypergraph

摘要：To address the issue of large network computing parameters for spatial-temporal features of actions in multi-view video sequences, this paper proposes a Spatial-Temporal Hypergraph Neural Network based on Attention Mechanism (STHGNN-AM). This method consists of a Temporal Attention Mechanism based on Trainable Threshold (TAM-TT) and a Multi-scale Spatial-Temporal Residual Module (MS-STRM), achieving multi-view data action recognition. Specifically, TAM-TT is constructed by using a learnable threshold to extract key frames of actions from different view video frames input to the module. MS-STRM is employed to further improve the model performance, and high-order semantic features of actions are learned in a hypergraph neural network. The MS-STRM extracts features using a multi-scale approach, modeling long-term and short-term semantic information to capture the temporal information changes between different frames. Comparative experiments on the NTU RGB+D and imitating traffic police gestures datasets evince the superior performance and heightened recognition accuracy exhibited by the proposed methodology, effectively enhancing the cognitive ability of machine-human body language interaction. ? 2023 IEEE.