Position Detection for Badminton Tactical Analysis based on Multi-person Pose Estimation

In the badminton game, the identification of the players position plays an important role in tactical analysis. This paper proposes to automaticly extract player positions for tactical analysis in badminton videos using computer vision techniques. Particularly, the pose estimation problem was formulated to obtain the accurate body parts of players. This problem can be solved by using deep learning and graph-based techniques. Then, the positions of a player are calculated through the positions of both feet and the nine-cell badminton court. Experimental evaluation on the self-constructed database illustrates that, proposed method obtains satisfied performance in most situations, such as near and far court, International competition and training, and male and female.