A novel algorithm for shuttlecock tracking

Real-time accurate ball tracking in sport games is important for automated game analysis and augment reality display. It is very difficult because the ball is usually very small, with few features and large variance appearance, sudden motion change and occlusion occurs quite often as well as the background is noisy. Badminton is one of the fastest ball games in the world and so tracking a shuttlecock accurately is a challenging work. In this paper, we propose a novel real-time non-recursive tracking algorithm to addressing this issue by formulating the motion model and the correlation within scene context, which is different from standard tracking approaches. We build a multi-layer filters focusing on the rationality to eliminate false candidates step by step and choose the optimal candidate without previous results. Experiment results show that the proposed tracker can track the shuttlecocks in real-time with satisfying performance.