A Motion Capture Framework for Table Tennis Using Optimized SVM and AdaBoost Algorithms

Abstract

Table tennis requires high technical and tactical skills. The application of motion capture technology can improve athletes' training effectiveness and competition strategies. To further improve the collection and capture efficiency of table tennis sports data, a high-precision optical motion capture system, and inertial measurement unit sensors are first used to collect table tennis sports data. Data preprocessing is carried out using action windows and sliding windows. Secondly, a support vector machine classifier optimized with a Gaussian radial basis kernel function is used for training. The sample weights are updated based on the feature classification results in each iteration. Finally, the adaptive boosting algorithm is combined with it to propose a new type of table tennis motion capture model. The experimental results showed that the optimized model achieved the highest classification accuracy of 96% with the best kernel parameters and a normalization factor of 1.0. The model's motion capture errors ranged from 1.5% to 9.7% and had the shortest runtime of 7.66 seconds. In addition, the model achieved the highest capture accuracy of 93%, 92%, 91%, and 90% for the four motions of forehand kill, forehand putt, backhand kill, and backhand putt, respectively. This demonstrates significant advantages in terms of accuracy and computational efficiency. Therefore, the proposed model not only improves the accuracy and efficiency of motion capture but also performs well in terms of resource consumption. This model has high practical application value and can provide a new reference for technological development in this field.