Tennis requires precise and rapid movements, making biomechanical assessments essential for optimizing player performance and preventing injuries. Traditional motion capture methods, while effective, are often expensive, highlighting the need for markerless alternatives. This study presents a dual-camera setup for 3D motion capture, expanding on previous research in this field. The system was tested using videos of tennis players recorded at a provincial Tennis Club, with MocapMe, a framework built on DeepLabCut and OpenPose.

The dataset includes synchronized recordings from two cameras capturing tennis strokes such as serves, forehands, and backhands, providing a robust foundation for analysis. A ResNet152 architecture was fine-tuned on this dataset to enable accurate 3D pose estimation. Preliminary tests indicate high accuracy in detecting key points, especially during complex movements like serves, and demonstrate the system’s stability across multiple trials. Although exact performance metrics are still being finalized, early results suggest that the system's accuracy is expected to significantly outperform previous 2D models, showing lower error rates and greater consistency.

This system offers rapid performance analysis and valuable insights for technique improvement and injury prevention. It holds significant potential for applications in sports biomechanics research, offering a practical solution for 3D biomechanical analysis in tennis. Future work will expand the dataset and integrate additional machine learning models to further enhance the system's capabilities.