Key Frame Detection in Badminton Swings and Its Application to Physical Education

The use of video analysis in sports training has revolutionized the way coaches and players evaluate performance and develop strategies. This paper presents a machine learning based approach for key frame detection in badminton swings aimed at improving the learning experience for beginners through visualization and real-time feedback.  Our proposed method uses the MediaPipe framework to extract 3D coordinates of skeleton joints, which serve as input features for a machine learning based model that accurately predicts key frame positions in badminton swing videos.  This model is integrated into a mobile app developed for Android tablets, allowing learners to record their swings and compare them with those of professional players, thereby enhancing badminton learning.  Comparative studies show that models with graph convolutional networks, a prominent approach commonly used in skeleton-based human action recognition, outperform other existing methods in terms of accuracy and reliability.  Experimental studies demonstrate the app's positive impact on performance, motivation, and self-perception.  One main contribution of this research is the development of a robust key frame detection model. Another main contribution is that we implement this key frame detection model in a tablet and verify it through extensive experiments.