Aiming at the problem that the joint point partition strategy expresses the important information of the human body in the human body behavior recognition of bones cannot fully express the behavior,anRCTR-GCNhuman bon...Aiming at the problem that the joint point partition strategy expresses the important information of the human body in the human body behavior recognition of bones cannot fully express the behavior,anRCTR-GCNhuman bone behavior recognition model of the correlation strategy is proposed.First,by adding an association strategy of a refined graph convolutional network model(CTR-GCN)of the smart channel topology,it can dynamically learn different topological structures and efficiently amplify the characteristics of the connection points in different channels while improving the key joint points of associated characteristics.Then,the network model redefines each channel by learning a shared topology and uses a specific channel relationship to unify the model through theoretical analysis;finally,redefining the model structure effectively reflects the associated information of local nodeswithin the channel.Action recognition has stronger aggregation capabilities.The results show that the recognition accuracy in the commonly used NTU RGB+D and NW-UCLA datasets reaches 93.6%(X-View),97.6%(XSub),and 97.2%,respectively.The experimental results show that the accuracy rate is improved.展开更多
文摘Aiming at the problem that the joint point partition strategy expresses the important information of the human body in the human body behavior recognition of bones cannot fully express the behavior,anRCTR-GCNhuman bone behavior recognition model of the correlation strategy is proposed.First,by adding an association strategy of a refined graph convolutional network model(CTR-GCN)of the smart channel topology,it can dynamically learn different topological structures and efficiently amplify the characteristics of the connection points in different channels while improving the key joint points of associated characteristics.Then,the network model redefines each channel by learning a shared topology and uses a specific channel relationship to unify the model through theoretical analysis;finally,redefining the model structure effectively reflects the associated information of local nodeswithin the channel.Action recognition has stronger aggregation capabilities.The results show that the recognition accuracy in the commonly used NTU RGB+D and NW-UCLA datasets reaches 93.6%(X-View),97.6%(XSub),and 97.2%,respectively.The experimental results show that the accuracy rate is improved.
