Multi-layer Graph Convolution Neural Network for Human Behavior Recognition Incorporating Spatial Attention

Aiming at the problem that the low accuracy in recognizing human behavior in complex environments using existing graph convolution related algorithms , this paper proposed a human body behavior recognition algorithm based on improved graph convolution network. The algorithm alternately used GCN and TCN to extract human features by stacking multi-layer spatiotemporal graph convolutions, and utilized spatiotemporal attention mechanism to enhance spatiotemporal feature extraction. This algorithm used OpenPose to extract human joint information, passed it into an improved spatiotemporal convolutional network model, and finally classified and recognized actions through Softmax. The algorithm was validated on the NTU-RGB+D dataset.Experimental results show that the algorithm can be effective in human behavior recognition ,with recognition accuracy of 88.9%.