Semi-Automatic Estimation of Body Orientation in Football

Body orientation is a crucial component for tactical analysis in professional football. This allows coaching staff to obtain insights into players' attention, decision-making, and strategic positioning. Likewise, body orientation might be an indicator of tactical readiness in football. This paper presents a computer vision pipeline for automatic body orientation estimation during pass events using player pose estimation. To evaluate the aforementioned pipeline, we used a series of video clips from the Women's Super League match between Brighton and Aston Villa in the 24/25 season. To validate our results, we compare them against manual annotations by a professional football analyst. We achieve 75% accuracy on real-world match footage, demonstrating the potential to reduce manual annotation in performance analytics. However, the wrong body orientation estimations were due to issues in the player and ball detection and the pass detection events, and not to the body orientation model. Therefore, we believe that the proposed pipeline can assist coaching staff to semi-automatically determine body orientation in professional football. However, more research is required to improve the ball and players' detection.