文献类型：会议论文

英文题名：Human Action Recognition Method Based on Spatio-temporal Relationship

作者：Yu, Haigang[1]; He, Ning[2]; Liu, Shengjie[1]; Han, Wenjing[2]

第一作者：Yu, Haigang

机构：[1] Laboratory of Information Service Engineering, Beijing Union University, Beijing, 100101, China; [2] Smart City College, Beijing Union University, Beijing, 100101, China

第一机构：北京联合大学

通讯机构：[2]Smart City College, Beijing Union University, Beijing, 100101, China|[1141734]北京联合大学智慧城市学院;[11417]北京联合大学;

会议论文集：Image and Graphics - 12th International Conference, ICIG 2023, Proceedings

会议日期：September 22, 2023 - September 24, 2023

会议地点：Nanjing, China

语种：英文

外文关键词：Extraction - Motion estimation

摘要：Traditional human motion recognition algorithms cannot fully utilize the spatio-temporal information in the video. In this paper, we propose a human motion recognition method with spatio-temporal modeling from local to global perspective. For local spatio-temporal information, the motion feature extraction module and the multi-scale spatio-temporal feature extraction module are designed. The local motion features are extracted by establishing cross-frame level correspondence within the network and combined into spatial features to obtain local spatio-temporal information. For global spatio-temporal information, the segmented network fusion module is designed. The representative recognition results are filtered from the video segments, and the fused filtered recognition results predict the final classification results. The results show that the model can effectively obtain the spatio-temporal dynamic information in the video for action recognition, achieve high recognition accuracy, and greatly reduce the model’s complexity. ? The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.