An evaluation of possession information in playfield zones from soccer video using mid-level descriptors

In this paper, we propose a method to acquire the possession information in different zones of the playfield from soccer video by using view type and playfield zone mid-level descriptors. First, each video frame is classified into three kinds of view type according to a domain-specific feature, grass area ratio and series of classification rules. Then, the classified frames are used to determine the currently active playfield zone in the match. The history of active playfield zones is post processed to acquire the possession information in playfield zones during the game. The efficiency and effectiveness of the proposed method are demonstrated over a large collection of soccer video data with different stadiums and conditions.