Samenvatting

This thesis explores person re-identification in the football domain. BallJames generates tracking data by tracking football players during a match. When an occlusion happens between multiple players, BallJames loses track of the player. Person re-identification could solve this problem by comparing the appearances of the players. A standard person re-identification model is not able to differentiate between players of the same team due to the similar apparel. By training a new person re-identification model on domain specific training data this problem is solved. The domain specific training data is collected using a newly created data extractor which can create semi automatic labeled training data in a time feasible manner.