The Effectiveness of Using YOLO11 Real-Time Object Detection to Determine Holding Penalties in Football

This study used the YOLO11 computer vision model to detect holding penalties in images of people bocking in football games. This study aimed to determine if a computer vision model can accurately detect holding penalties in football games. It was hypothesized that the YOLO11 model would determine the difference between a holding and blocking penalty with an 85% accuracy. For the study, the model was trained on 200 images, 100 holding penalties, and 100 regular blocking during football games. Then, 50 images, 25 holding, and 25 blocking, were tested with the model to determine whether the algorithm could detect when a holding penalty occurred and when it didn’t. The test found that the model achieved an accuracy of 74% with 37 images being correctly detected as either holding or blocking and 13 being incorrectly detected. The hypothesis was not supported as the model did not accurately detect holding or blocking 85% of the time. In comparison to previous studies, the model was less accurate. The model is not viable for NFL usage as NFL referees are accurate on 98.9% of calls. For future research, the model could be improved by increasing accuracy. Different penalties could also be tested in future studies.