Multiple Object Tracking

Tracking players in occlusion by softly clustering conceptual soccer field cells. Accurately tracking soccer players is a non-trivial task due to existence of various challenges. Soccer players try to confuse each other with unexpected changes in their velocity. Moreover players almost look identical and they are frequently involved in possession challenges and tackles in which they could be completely occluded by another resulting in tracking ambiguities. Player tracking algorithm is specifically designed to address the problems above and precisely track players in rain or shine. “In our approach, the soccer field is conceptually modeled as a two dimensional world and partitioned into a grid consisting of dense spatial cells. Each cell corresponds to unique ground point on the soccer field and is represented by a fixed image patch in the video.” We employ a well-known machine learning technique and a classifier, trained with over 100k player samples, to detect the set of cells containing players in the video. This allows our tracking algorithm to function in a variety of challenging environmental conditions.