Detection and Player Tracking on Videos from SoccerTrack Dataset

Analysing player movement on the field during a match can enhance individual player performance and overall team play. Existing solutions often rely on bulky Global Navigation Satellite Systems devices attached to players or a complex array of multiple cameras, each covering a portion of the football field and tracking players' movement. This paper presents a novel approach using a single high-resolution fisheye camera and state-of-the-art object detection and tracking algorithms. The proposed algorithm provides accurate player coordinates and player IDs at every moment. We explore various YOLOv8 algorithm variants with different image pre-processing techniques and evaluate three object tracking algorithms. This approach significantly reduces the overall system cost, enabling even smaller teams to leverage accurate match statistics for performance improvement.