Event identification, which refers to defining important moments within a movement, is a fundamental process in biomechanical analysis. This study aimed to develop and evaluate deep learning models for automated detection of the moment of racket-ball contact during tennis strokes. For this purpose, Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks were implemented using kinematic data collected from a markerless motion capture system. Kinematic data were obtained from 16 collegiate tennis players performing standard stroke techniques. Among various joint combinations, a set of ankle, elbow, knee, and wrist data provided the highest performance with 95% accuracy. The LSTM model achieved a precision of 95.77%, an F1-score of 97.09%, and an AUC of 98.06%, while the GRU model achieved a precision of 96.23%, an F1-score of 97.71%, and an AUC of 99.20%. These findings demonstrate that automated detection of sport-specific biomechanical events can effectively replace labour-intensive manual annotation. Moreover, markerless motion capture systems enable large-scale and ecologically valid data collection, offering a viable alternative to laboratory-based methods. The proposed approach provides a methodological basis for near-real-time monitoring and practical feedback in tennis performance analysis and training.
Lee et al. (Mon,) studied this question.