| The large numbers of sport games brought the explosive expanding of sport video directly. Therefore, efficient sport video analysis is necessary. In this paper, the segmentation and tracking of basketball are studied for video analysis of basketball game. Firstly, according to the characteristic of basketball in close shots and establishing shots, we design the corresponding algorithms for segmentation of basketball. Then, the algorithms of basketball tracking are designed and implemented. The details are as following: 1. Firstly, the basketball segmentation in close shots is studied. According to the consistency of basketball color, color segmentation is implemented. Then with the help of steerable filters, the method detects image edges and the orientations of them. Lastly, based on edges and their orientations, Hough transform is improved to get the circle center and the radius of basketball. 2. In the basketball segmentation in establishing shots, Gauss filter is adopted to smooth noise in the image firstly. Secondly, the background is separated by the difference of inter-frame and the connected regions are labeled. Thirdly, a strategy is designed to identify basketball by the characteristic of basketball. Finally, the edge deviation is revised and the optimal result is obtained using improved Snakes. 3. Tracking of basketball is studied. Firstly, the basketball object template and adaptive object model are established by using the basic characteristic of segmented basketball in the former frame. Then an improved block-matching algorithm is proposed for tracking the basketball. Finally, the edge of basketball is revised, and the mechanism for tracking validity is established. 4. Finally, the extraction and description of basketball trajectory is studied preparatorily. And experiment platform is built.
|
PDF Full Text Request