Skeleton-Based Mutual Action Recognition Using Interactive Skeleton Graph and Joint Attention

Skeleton-based action recognition relies on skeleton sequences to detect certain predetermined types of human actions. The existing related works are inadequate in mutual action recognition. We thus propose an innovative interactive skeleton graph to represent the skeleton data. In addition, because the GCN pays attention to the information about the edges in the skeleton graph which represent the interaction between joints, we propose a joint attention module that assists the model in paying attention to the pattern of vertices which represent the joints in the skeleton graph. We validate our model on the NTU RGB-D datasets, and the experimental results demonstrate the superiority of our model against other baseline methods in terms of recognition effectiveness in understanding mutual actions.