Ball Carrier Detection and Behavior Recognition in Basketball Match Using Covariance Descriptor

This paper presents a novel approach for ball carrier detection and behavior recognition in a basketball match. Considering the cluttered background, fast motion of the players and the low resolution of the player images in the basketball match video, a covariance descriptor is adopted to fuse multiple visual features of the player, which can be characterized as a point on the Riemannian manifold and can be projected to the tangent space through the homeomorphic mapping. Therefore, detection and behavior recognition of the ball carrier is simultaneously completed on the tangent space through the trained multiclass LogitBoost. Experimental results demonstrate that the proposed method performs well on video recorded in the NBA basketball matches.