Network Structure for Tracking of Jockeys in Horse Races
Mohammad Hadi Hedayati, Michael J. Cree, Jonathan Brereton Scott · 2014
This article proposes a model to track contenders (jockeys) in horse races from broadcast videos based on the colour property of contenders. To overcome the effects of background noise and occlusion the proposed system merges contextual information of the horse race into a colour-based tracking framework (mean shift). There are two improvements. The first is the design of a model, specifically for horse races, to extract the contenders from the background. The second estimates reliability of a trajectory by building a model called "network structure". The proposed algorithm is tested on eleven broadcast videos of horse races where the mean length in time of video is 51s. To evaluate the proposed model, the races are divided into five intervals and the percentage of correct tracking (PCT) is calculated at the end of each interval. A comparison study of the proposed method, with a mean shift tracking method, shows that network structure improves tracking in the videos.