Automatic parsing of TV soccer programs

Abstract

While automatic parsing of general video images are difficult or nearly impossible, parsing a specific type of video image can be achieved if the video structure can be explicitly identified and its model can be constructed. The paper presents ongoing work on using domain knowledge to parse content of soccer video programs. The proposed system can classify a sequence of soccer frames into various play categories, such as shot at left goal, top left corner kick play in right penalty area, in midfield, etc, based on a priori model comprising four major components: a soccer court, a ball, the players and the motion vectors. Approaches in applying the model for parsing soccer programs are presented in detail. Experimental results are included to evaluate the performance of the system.