Lightweight Combat Motion Recognition Based on Human Keypoint Extraction by MediaPipe Models

With the development of computer vision technology, motion recognition based on key points in the human body has also found many promising applications in sports analysis, intelligent monitoring and human-computer interaction. This paper proposes a lightweight motion recognition method for combat actions. The first is the MediaPipe framework for extracting human keypoints. Build a Fusion-Score Algorithm Next. Systematically calculate multi-dimensional features of the distance, speed and joint angle to detect punch motion. Two groups of comparative experiments have been conducted to assess the performance of the feature fusion method. Based on the above experiments, the multi-feature fusion algorithm is better than those employing only one feature in terms of both accuracy and robustness, with a success rate of 50%. Experimental results show that the proposed multi-feature fusion algorithm outperforms single-feature methods in both accuracy and robustness, verifying the effectiveness and application potential of the proposed method in real-world action recognition.