ICAIET 2025

Computer vision is transforming sports analytics by providing valuable insights into player performance, team dynamics, and tactical strategies through data-driven approaches. This manuscript introduces an integrated approach that utilizes complex computer vision techniques to examine footage from football matches. The proposed approach employs YOLOv5 for detecting objects, K-means clustering to classify teams based on their jersey colors, optical flow for stabilizing camera motion, and perspective transformation to map player positions accurately. A custom-trained YOLOv5 model achieves a remarkable mean average precision (mAP@0.50) of 84.5%, showing the capability to track players, goalkeepers, referees, and the ball under various match conditions. K-means clustering enables accurate estimation of ball possession by analyzing spatial proximity, and optical flow correction improves the accuracy of speed and trajectory measurements. Transforming perspective allows precise estimation of distance and position within real-world coordinates. The proposed system provides a strong, automated approach to match analytics, giving coaches, analysts, and broadcasters valuable, objective insights that can improve strategic decision-making.

Aug 28, 2025 2:30 PM — 5:00 PM