Hockey Player Localization

Our first competition team competed in the Player Localization competition on the platform AIcrowd. This competition was about creating a top-down view of a hockey field with hockey players and the goal was to be able to locate the hockey players during the match.

The Ice Hockey Player Localization competition was about automating the process of analyzing Ice Hockey games. The purpose of the competition was to keep track of Ice Hockey players throughout the game. The given dataset consists of 17 videos of Ice Hockey matches, in which the camera was often rotating and zooming. The footage also differs hugely in quality, camera position and light exposure.

Each frame was annotated with the coordinates of players on the frame itself. Our job was to predict the coordinates of players on a top-down view of the field.

Our approach detects static points on the field, such as corners, goals and certain marks on the border. Once we have mapped these points to their corresponding coordinates on the top-down view of the field, we can apply holography to transform image coordinates to coordinates on the top-down fields.

Using this approach, we ultimately finished in second place on the public leaderboard!