Automatic player detection, tracking and mapping to field model for broadcast soccer videos
Quang Tran, Bac Thi Vo, Tien Ba Dinh, Duc Anh Duong · 2011
We present an automatic soccer analysis framework through detecting, tracking players, and reflecting the information into the field model. First, a player detector is built on four-seed edge feature approach which extracts player regions applied in long view shots including playfield extraction, shot view classification, and player segmentation. Second, a multi-player tracker uses a high general reversible jump Markov chain Monte Carlo (RJMCMC)-based approach to associate player regions detected in each frame. Third, a fast calibration algorithm matches field model by fitting two regions: the center circle and the penalty areas so as to provide the geometry transformation enabled to map player positions in a video frame to the real-world coordinate. The experimental results on broadcast videos of the 2010 FIFA World Cup South Africa have evaluated to prove the effective design and the promise of the proposed framework.