Research on the Recognition and Evaluation Method of Sparring Movements Based on Skeletal Keypoints

Motion recognition and assessment of Sanda experiments show potential for motion capture technology through the detection and analysis of bone key points. Based on a comparison of multiple bone key detection models, the AlphaPose algorithm attains excellent accuracy and speed in the performance of accurate extraction of athlete movements. Key frame extraction algorithm is deployed to effectively reduce the amount of data, which in turn optimizes the subsequent classification process. The strict experimental validation has brought significant improvement in the accuracy and performance of three classification models, which are random forest, support vector machine, and neural network. Among them, neural networks show overwhelming superiority in classifying complex actions. The similarity evaluation method tests Euclidean distance, cosine similarity, and Mahalanobis distance one by one, while results show that Mahalanobis distance is the best in evaluating action similarity. In general, the movement recognition based on bone key points not only improves the accuracy of motion analysis, referral for the athletes' technical training and competition preparation but proves the wide application potential of this technology in sports science. The identification accuracy and evaluation consistency have shown excellent results in terms of analyzing experimental data, thus proving the innovativeness and practicality of this method.