Real-Time Tennis Stroke Classification Using Deep Learning: A Comparative Analysis of YOLO-Based Models

This study presents the development of a real-time pose estimation system for classifying four key tennis strokes: ready position, backhand, serve, and forehand. Accurate pose estimation is crucial for performance analysis, training feedback, player coaching, and injury prevention. Despite significant advancements in object detection algorithms, the challenge of selecting the most effective model for specific, real-world applications persists. In this research, we implemented and evaluated three advanced versions of the You Only Look Once (YOLO) algorithm YOLOv8, YOLOv11, and YOLOv12 specifically optimized for tennis stroke estimation. Our approach utilized a large, annotated dataset of tennis stroke videos, ensuring model robustness and generalization. The system demonstrated classification accuracies of 99.08% with YOLOv8, 99.12% with YOLOv11, and 98.54% with YOLOv12, all measured at a confidence threshold of 50%. Among the models tested, YOLOv11 achieved the highest accuracy, making it the most suitable for real-time implementation in a controlled tennis environment. These results highlight the importance of selecting the appropriate deep learning model for sports analysis and emphasize YOLOv11's superior performance for precise and efficient stroke classification in tennis. The proposed system offers a reliable tool for coaches, analysts, and athletes aiming to enhance performance and minimize errors.