Real-Time Motion Recognition and Evaluation Based on Computer Vision

This study proposes an intelligent motion analysis system combining ResNet and GCN, which aims to recognize athletes' movements in real time and make accurate evaluations. The system first uses ResNet to extract spatial features from sports videos, processes athletes' posture information and key movements, and then uses the GCN model to model the temporal relationship between the joints of the athletes' bodies to capture the dynamic changes of movements. Through this combination of spatiotemporal features, the system can efficiently recognize and analyze complex sports movements. This study used a video dataset containing 5,000 annotated movements, covering a variety of sports types (such as running, weightlifting, yoga, etc.) for algorithm training and testing. Experimental results show that the system has an accuracy of 94.7% in motion recognition, which is 9% higher than the traditional SVM and KNN methods. In addition, the system also performs very well in real-time motion evaluation, and can provide instant feedback based on the quality of the movement, helping athletes adjust their postures and optimize training effects.