Detection and Classification of Correctly and Incorrectly Played Strokes in Tennis

Abstract

This study presents an automated approach for detecting and classifying tennis strokes using the YouTube videos of professional Tennis players as a benchmark dataset, enhanced with MediaPipe for pose estimation and CNNs for classification. MediaPipe, an open-source library by Google, identifies key body landmarks for precise motion analysis, providing essential data for the classification of forehand, backhand, and serve strokes. Our model achieves a 95.31% accuracy in stroke identification and 96.51% in distinguishing between correct and incorrect strokes, supporting real-time feedback for players and coaches. This system aims to optimize training, improve player performance, and mitigate injury risks through consistent, objective stroke analysis.