Research on Table Tennis Movement Evaluation Method Based on Human Posture Estimation

Aiming at the problems of poor visualization performance and low accuracy of movement evaluation in traditional sports motion capture methods, this paper proposes a table tennis swing movement evaluation method based on human posture estimation. By combining Support Vector Machine (SVM) and Random Forest algorithm, an improved SSVM classification model is designed, and SHAP values are introduced to enhance the model interpretability and quantify the contribution of human skeletal key point features to action classification. The Mediapipe framework is used to extract 14 skeletal keypoints of the upper body, and the key frame extraction, normalization and histogram equalization pre-processing steps are combined to construct a table tennis swing action dataset (covering the standard actions of forehand and backhand attacking as well as the non-standard actions of going down to the net and going out of bounds). The experiments compare the performance of Random Forest, traditional SVM and SSVM algorithms, and the results show that: the accuracy of SSVM in the classification of forehand and backhand motions reaches 0.94 and 0.96±0.1, respectively, and the F1 scores are 0.95±0.2 and 0.96±0.1, which are significantly better than the other algorithms; the analysis of the SHAP values shows that the key points of elbow and shoulder have the greatest contribution to the identification of non-standard motions (e.g., ball off the net). with the greatest contribution. The method not only improves the accuracy and real-time performance of movement classification, but also provides targeted guidance for sports training by visualizing the importance of features, which verifies its practical value in table tennis teaching and sports injury prevention. The method can be further extended to the evaluation and training of other sports movements, which has wide application significance, especially in the field of intelligent fitness and sports rehabilitation.