FULBO

Overview

FULBO is a computer-vision pipeline for analyzing football matches over the SoccerNet dataset. It detects players, referees, and the ball with YOLO-based models, estimates player pose keypoints, and tracks everyone across frames. By detecting pitch keypoints and solving the homography, it projects the action onto a top-down tactical minimap and clusters players into teams, the raw material for automated match analysis.

Highlights

• YOLO detection plus pose models for players, referees, and ball, with dedicated ball fine-tuning.
• Pitch keypoint models (29- and 57-point) drive a homography that projects play onto a tactical minimap.
• Team clustering from appearance, plus tracking and segmentation experiments with Segment Anything 2.

Report