Skeleton-dominated two stream graph convolutional networks for skeleton-based action recognition

In daily life and industrial development, action recognition have been widely used. However, most of the current RGBbased action recognition models focus too much on background information and not on the posture information and action information of the behavior, which may cause many problems in the industry and even be detrimental to the development of the industry. For this purpose, we propose a Two-stream network architecture that fuses the skeletonbased posture and action information and the RGB-based fine-grained image information. And the effect of our model was verified on the Mimetics data set which eliminated background interference.