现代电子技术2025，Vol.48Issue(16)：45-49,5.DOI:10.16652/j.issn.1004-373x.2025.16.008
投影子空间下人体动作序列预测研究
Prediction of human motion sequence in projection subspace
摘要
Abstract
In order to comprehensively extract motion features from different viewing angles and improve the robustness of perspective changes,the prediction method of human motion sequence in projection subspace is proposed.In the projection subspace,the 3D human movements captured by Kinect camera are converted into 2D human movement sequences with different angles of main view,left view and top view.In the multi-neighborhood global adaptive graph neural network,the adaptive graph convolution is used as the encoder to extract the 2D human action sequence features from the three viewing angles,so as to improve the robustness of the network to the changes of viewing angles.The gated recurrent unit based on multi-neighborhood global adaptive graph convolution is used as the decoder,and combined with the human motion sequence characteristics to obtain the human motion sequence prediction results under different viewing angles.The fusion module is used to fuse the predicted results of action sequences from various perspectives by means of the voting fusion strategy,so as to obtain the final prediction results.The experimental results show that the proposed method can effectively convert 3D human motion and obtain 2D human motion sequence.This method can effectively predict human action sequence,and the determination coefficients of action sequence prediction are higher under different viewing angles,that is,the prediction accuracy is higher.关键词
人体动作序列预测/投影子空间/2D动作/特征提取/自适应图卷积/门控循环单元Key words
human action sequence prediction/projection subspace/2D action/feature extraction/adaptive graph convolution/gated recurrent unit分类
信息技术与安全科学引用本文复制引用
岳晓玉,章璐璐..投影子空间下人体动作序列预测研究[J].现代电子技术,2025,48(16):45-49,5.基金项目
江苏省社会科学基金项目研究成果(22TYB012) （22TYB012）
