Development of an Annotation Tool for On-the-ball Soccer Player Localization

抄録

Recognition of which player is currently having the ball is necessary to perform soccer player analysis. Our previous work showed that conventional object-tracking methods are limited in detecting the ball consistently. Focus on the player's pose could be an alternative approach to localize on-the-ball players. We devised an Annotation Tool with Pose Estimation to collect the players' poses and Object Tracking to apply consistent ID for each player's pose. The annotation task selects the on-the-ball player and stores its ID. The result is an annotation file with a localized pose representing the on-the-ball player per frame.