Recognizing football game events: Handball based on Computer Vision

Football or Soccer is one of the most popular games in the world. “Handball” event in the game is one of the most controversial and important decisions by a single referee. This paper proposes a method to define a “handball event” in a football game by recognizing the hand, ball, and their interaction from a single camera. We trained a model to identify hands and balls using detectron2, then used the vertices of two objects to find out the overlapping situation between these two objects to determine the handball event in a football game. The train and test result of detection was satisfactory with 96% for hand and 100% for the ball respectively.