Research on the Application Model of Computer Vision in Posture Analysis of College Sports

This paper proposes a sports posture analysis model based on GCN - SelfAttention-PoseNet. It combines GCN and self-attention mechanism to improve the accuracy and real-time performance of sports posture analysis in college sports teaching. The key innovation of this model is to deal with the spatial relations of athletes' joints by means of graph convolutional networks. GCN can model the spatial structure between joints and capture the spatial dynamic characteristics of athletes in different sports movements, while the self-attention mechanism helps the model focus on key time steps, thereby reducing the interference of redundant information and improving the accuracy of temporal feature extraction. In order to verify the effectiveness of the model, this paper uses sports data sets collected in college sports teaching, including running, jumping, shooting and other common sports movements. GCN-SelfAttention-PoseNet shows significant advantages over traditional posture estimation methods (such as OpenPose). The model improves the accuracy of posture estimation by 6.2 %, the real-time performance by about 25 %, and can still maintain a high recognition accuracy of 93.7 % in the face of complex backgrounds and lighting changes. In addition, the model can run stably in dynamic scenes, significantly improving the robustness of athlete posture recognition.