Abstract
This paper addresses a tracking problem of the golf swing from a video sequence. A particle filter is widely used for the problem. However, it is not easy to achieve good performance result for golf swing tracking problem by using conventional particle filter. To overcome the problem, we combine a Pictorial Structure Model (PSM) as a global pose estimation, with a particle filter approach. In particular, the proposed method enhances the recovery performance of lost track by the global pose estimation. The proposed matching method consists of four steps. First, the golf player's position is found by using the person detector. Second, a pose estimation algorithm using silhouette features is applied for the global pose estimation. Third, left arm of the golfer is tracked by using PSM. Finally, the grip of the golfer is tracked by particle filter. Some experimental results demonstrate that the proposed method works out more accurately than the standard particle filter in tracking of uncontrolled golf swings.
