Automatic soccer players tracking in goal scenes by camera motion elimination

In this paper, we propose a novel and effective algorithm for tracking soccer players in goal scenes, by eliminating fast camera motions effect through the correspondence between line marks in soccer field model and image sequences. The proposed algorithm comprises four steps. At the first step, we introduce an automatic grass field extraction algorithm that is tested for various soccer video types and conditions. The field line marks, which are used to locate the players, are detected at the second step using Hough Transform and tracked in all frames by predicting their positions using Kalman Filter. At the third step, we introduce a novel approach for estimating the players’ positions during the course of tracking players. We estimate a player’s position in the current frame by observing his position in the old frame within the soccer field model, using Perspective Transformation of the old frame to the real world coordinate system, and then by projecting back the obtained player’s position into the current frame. At the final step, the players are tracked by applying the region-based detection algorithm, the Histogram Back-Projection algorithm or a combination of the Merge-Split approach and the Template-Matching algorithm around the estimated positions, depending on whether or not the occlusion has occurred and if yes, how it has occurred. Then their memberships are identified by an algorithm that employs both appearance and spatial information of the players. Image sequences of different soccer games were captured from different sources, and all experiments were performed off-line. The results of our experimentations show that our algorithm is highly robust to occlusion, different soccer field colours, different lights such as sunlight and spotlight, shadows of players and fading the whole screen due to fast camera movements.