Multi-camera tracking of soccer players through severe occlusions

Accurate visual tracking of soccer players is in high demand in the industry. Existing solutions are however unable to track players during severe occlusions, making manual supervision necessary. In order to get to a fully autonomous system we propose a setup based on the fusion of 14 cameras. An algorithm with at its core visual hull calculations combined with an HOG and HSV based appearance model is employed to track the players. It is shown that the proposed system is able to reduce the number of identity switches while maintaining tracking accuracy during severe occlusions.