Tracking players in highly complex scenes in broadcast soccer video using a constraint satisfaction approach
Jun Miura, Hiroyuki Kubo · 2008
This paper deals with player tracking in broadcast soccer video. In soccer games, players sometimes gather in a small area in the case of, for example, a corner kick. In such a case, due to a heavy occlusion, a simple detection-and-tracking method will certainly fail. We cope with such difficult cases using a constraint satisfaction approach. To integrate pieces of evidence at various places and frames, we construct a graph of player blobs representing possible player transitions. The view of each blob provides a constraint on the number of players in the blob. All such constraints are propagated through the graph to reduce the ambiguities in the numbers. The remaining ambiguities after the propagation is handled by a statistical approach in which a set of the most likely interpretations on the numbers is selected. Finally the players' trajectory are determined based on their smoothness. Experimental results show the effectiveness of the method.