Golf video tracking based on recognition with HOG and spatial–temporal vector

Abstract

The hand and club movements contain golfer’s swing information, which can be obtained to provide good visualization to be shared on Internet and be summarized in golf studying. In this article, a hand and club tracking framework based on recognition with a complex descriptor combining histograms of oriented gradients and spatial–temporal vector is proposed to obtain their movement trajectories in golf video. After the hand and club are recognized in initial windows defined by the body region, a boosted classifier trained by the proposed descriptor is utilized for recognition and tracking in a searching window predicted by trajectory fitting with previous four object positions. Experiments show that the boosted classifier can have a precision and recall rate both better than 97%, and the hand and club tracking are basically correct in our testing videos.