Real-Time Football Match Analysis: Leveraging YOLO for Enhanced Object Detection and Possession Tracking

This paper presents a novel, real-time computer vision system for comprehensive football match analysis, focusing on the accurate detection and tracking of players, referees, and the ball. Leveraging advanced YOLO-based models, this research conducted a rigorous comparative evaluation of YOLOv8x and YOLOv12x models, marking the first documented application and performance comparison of YOLOv12x in football analytics. Our findings show YOLOv8x outperforms YOLOv12x in detection accuracy (F1-score), despite YOLOv12x's higher efficiency in terms of fewer parameters and lower inference time. To enhance ball detection, we integrated a Kalman filter, significantly boosting its F1-score. The system also precisely calculates ball possession based on player-to-ball proximity. Tested on both local Jordanian and diverse global football matches, the proposed system offers substantial potential for improving sports broadcasting and advanced analytics, providing deeper insights into game dynamics and player performance.