Player extraction in sports video sequences

This paper presents the results of the work on developing methods for object segmentation and preparation of algorithms for tracking objects in 3D space. This paper focuses on the case where video sequences recorded by many, fast-moving cameras with changing focal length lenses are processed. It describes in detail two main processes of the segmentation algorithm, namely the global camera motion estimation and object model fitting, using calculation of camera parameters from homography matrix. In the paper, the results for player detection in sport video sequences are presented.