LAUSR

In view of the fact that the study of human action recognition systems is a very important research field in the field of artificial intelligence, the use of a continuous image of the human skeleton’s key points for deep learning action training and identification is the current development focus. However, it is limited by the image shooting angle and the motion attitude transformation of the visual masking problem, resulting in the misjudgment of human skeleton key points affecting the accuracy of motion training and identification. This research paper puts forward a human body motion recognition system that uses a human skeleton key point correction method to improve the accuracy of human body motion recognition. The basic correction algorithm of a human skeleton key candidate point is based on the principle of the symmetrical characteristics of the human skeleton key points, and the human skeleton key candidate point advanced correction algorithm is based on using the human body shield map as the limit of the human skeleton key point range. This research paper proposes a performance comparison of this system and the relevant human action identification systems ST-GCN, 2 S-AGCN, GCN-NAS, with Gaussian filter for the ST-GCN, GCN-NAS and 2 S-AGCN. Human motion recognition accuracy is increased by at least 68%, 40%, 56%, 68%, 50% and 46%, respectively, for the experimental data.