Trajectory-Based Detection And Tracking Approach For Semantic Video Object

A trajectory-based ball detection and tracking approach for soccer video isproposed in this paper. In view of difficulties of direct detection, an indirect strategybased on non-ball elimination is applied. We distinguish the ball with a coarse-to-fineprocess. Game field is firstly extracted and the rest regions are considered as candidateobjects. Then, at the coarse step, some distinct non-ball objects are removed viaevaluation of several simple color and shape features. And at the fine step, a confidencevalue representing the ball region's reliability is assigned to every candidate objects.Finally, based on the ball trajectory restriction, we detect and track the optimal one thatis mostly like the ball region. The experimental results on six soccer video clipsconfirmed the effectiveness and robustness of our framework.