ABSTRACT Computer contact point constraint technology provides accurate guidance for training by monitoring athletes' contact point data in real time. However, existing systems still have shortcomings in real‐time, dynamic analysis, and personalized feedback. Therefore, an interactive sports auxiliary training architecture based on computer contact point constraints is proposed, which uses an improved object detection algorithm to achieve accurate detection of contact points. Then, it combines graph convolutional networks and improved long short‐term memory networks to dynamically analyze the position and action trajectory of contact points. The proposed method outperformed the comparison algorithms on recall and precision in contact point detection, achieving an average precision of 92.16% and a maximum precision of 95.91%. The mean absolute error in trajectory prediction was only 0.72 px, which was better than comparison methods. In addition, the average feature extraction accuracy reached 93.08%, and the average sensitivity was 91.81%. Overall, the proposed architecture improves the real‐time performance and accuracy of contact point detection and motion trajectory analysis, and provides technical support for adaptive and feedback‐oriented sports training applications, showing promising practical potential.
Minghui Zhu (Mon,) studied this question.