Abstract:
Tennis ranks among the most popular sports worldwide, with over 89 million participants in 2021. Success in tennis requires technical skill, strategy, mental resilience, and peak physical conditioning. Traditional performance analysis relies on manual observation, which is time consuming and requires specialized equipment. Calculating shot speeds and player movements manually is inefficient. This research introduces an automated approach using object detection to quickly and accurately extract performance metrics, reducing manual effort and analysis time. The study collected tennis match footage from YouTube, using a dataset of 578 images from various court angles to detect the tennis ball. Videos were standardized through pre-processing, and annotation was performed using Roboflow to label players and the ball. A separate dataset was used to mark court key points for performance calculations. The YOLOv8 model was trained to detect players, the ball, and key points on the court. YOLOv8 balances speed and accuracy, with an improved anchor-free detection mechanism for better generalization across datasets. It also supports multiple vision tasks, making it a versatile choice. The model successfully drew bounding boxes, tracked trajectories, and identified key points for precise measurements. Frame by frame analysis determined player and shot speeds, providing insights into match intensity and efficiency. Visual inspection confirmed accurate player identification and key point mapping. This research minimizes manual effort in performance analysis and delivers actionable data. The system enables coaches to track player performance, identify weaknesses, and tailor training programs. Players can analyse matches, assess shot efficiency, and refine strategies based on objective data
