Human action recognition is an important branch in the field of computer vision. The graph convolution method based on the human skeleton diagram is one of the most mainstream methods used by researchers. The human skeleton diagram contains all the information about the movement and is less influenced by the surrounding environment. However, less work has been done so far on the feature extraction for human skeleton diagrams. In this paper, a new skeleton feature extraction algorithm is proposed, which fully considers the characteristics of the motion object itself in each action video, and it will be adjusted according to the scope of the motion concentration that can reduce the impact of the motion object's characteristics on the accuracy of the action recognition. The spatial temporal graph convolutional networks (ST-GCN) is used as the basic model and significantly improves the recognition accuracy on the NTU-RGB-D dataset.