Automatic soccer field line recognition by minimum information

Abstract

Automatic analysis in soccer scenes is still a difficult task in the absence of soccer field information. The first and most important step in almost all analysis, is soccer field line recognition and homography extraction. The aim of this paper is introducing a novel approach for automatic detection and recognition of soccer field lines and arcs by minimal information. A simple camera model and perspective map is assumed to reduce unknown parameters. An accurate method is utilized for detecting line pixels. The side of playfield area is determined based on the orientation of lines and arcs. Based on the detected playfield area side, an initial perspective map is obtained. An optimization algorithm then adjusts the parameters of perspective transform and camera. The proposed method needs only some minimal information in theory and practice. It is applied to some typical soccer videos. The achieved results demonstrate its robustness and accuracy.