Player Segmentation Evaluation for Trajectory Estimation in Soccer Games

In this paper, we evaluate the player segmentation for trajectory estimation in soccer games. In order to estimate the field trajectories of players in soccer games, we should accurately locate the foot positions of players in each soccer image and transform them into those in the soccer field. However, we cannot always segment the players completely, since players are often motion-blurred due to the fast motion of camera. We use k-means algorithm for accurate segmentation of the player’s legs. Finally, we simulate the trajectory estimation for three different segmentation results: (i) when the legs of the players are accurately segmented; (ii) when the legs under the knees are missing; (iii) when only the torsos are segmented. Experimental results show that foot positions of the players should be located for accurate trajectory estimation.