3D Reconstruction and Video Annotation of Soccer Game

Recently, sports analysis and annotation become a hot issue in the fields of multimedia. In this paper, we present a practical system to convert a video sequence of soccer game into an animated virtual 3D scene. The system tracks players with a color based segmentation algorithm and template matching method. The 3D positions of players can be computed from the rectified mosaics of the whole scene. A Kalman filter is used to remove noise. The height of ball in air can be recovered by parabola approximation. According to the trajectory of players and ball, the system can recognize the behaviors of players and comment on the game automatically like a human being. At last, an OpenGL based virtual soccer scene will be constructed.