摘要
Abstract
In recent years,the application of emotion recognition in mental computing,human-computer interaction,and mental state moni-toring has attracted considerable attention.Compared with other methods such as facial and electroencephalogram(EEG)emotion recognition,gait data does not require high-precision shooting equipment and can be collected at a long distance.Although some research has been con-ducted in this field and corresponding progress has been made,there are still two problems that remain unsolved.First,most of the existing work on gait-based emotion recognition focuses on exploring the local correlation of human joints from skeleton images through graph convolu-tional networks(GCNs),while ignoring the global correlation of human joints.Second,the human body's naturally connected joint skeleton graph is used,and the original fixed connection will limit the network's ability to capture the interaction between distant joints.To solve these problems,this paper proposes a graph convolutional network based on pseudo-node cross-attention,which realizes the timely transmission of information of all joints through the pseudo-node method and uses the cross-attention method to capture efficient gait representation for emo-tional state recognition.The proposed method is evaluated on the dataset Emotion-Gait,and the accuracy rate reaches 88.63％,it has better performance compared with the existing classic advanced models.关键词
步态情绪识别/交叉注意力/图卷积神经网络/关节邻接矩阵Key words
gait emotion recognition/cross attention/graph convolutional networks/joint adjacency matrix分类
信息技术与安全科学
