How Can Real-Time Computer Vision Models Accurately Detect and Track Player and Ball Movement in Tennis Matches, and How Do Their Performance and Reliability Vary Under Different Match Conditions?

Arush Singhania (1)

(1) Student, Saint Stephen's Episcopal School, Bradenton, FL

This study examines real-time computer vision models for detecting and tracking player and ball movements in tennis matches under varying conditions, including changes in lighting, camera angles, occlusion, and motion blur. Accurate detection and tracking are essential for performance analytics, automated broadcasting, and sports officiating. The proposed Real-Time Tennis Analytics System (TAS) integrates YOLOv8 with motion tracking frameworks such as SORT and DeepSORT to achieve high accuracy and low latency match analysis. YOLOv8 detects players, the ball, and court lines, while a motion analysis module computes player trajectories, shot speeds, rally durations, and positional patterns.

Models are trained on diverse professional and synthetic match footage annotated with player and ball positions. Performance is evaluated using precision, recall, Multiple Object Tracking Accuracy (MOTA), and Frames Per Second (FPS). TAS achieves high accuracy under stable lighting and camera conditions, though reliability decreases during high-speed rallies, shadowed courts, or low-resolution footage. The study highlights trade-offs between model complexity and responsiveness and proposes adaptive data augmentation and temporal smoothing to improve robustness. Overall, TAS shows that combining YOLOv8 detection with motion analytics provides real-time insights for coaches, analysts, and broadcasters, supporting data-driven decisions in tennis and other racket sports.

Keywords: YOLOv8; Tennis Analytics; Object Detection; Motion Tracking; Computer Vision; Sports AI

Article Type: Original Research

Final Paper.docx.pdf

Download PDF • 2.68MB