A fast feature extraction algorithm for omnidirectional vision system

Inspired by the fact that human beings will concentrate on the Region of Interest(ROI) when facing a complex scene, a novel feature extraction algorithm named biSCAN is proposed for omni-directional vision system in RoboCup domain. The algorithm can mainly be divided into two parts: the rough scanning stage and the fine scanning stage. In rough scanning stage, the Field Region of Interest(FROI) is fast extracted through a few radial scanning based on the prior knowledge that the robot lies on the field. In fine scanning stage, feature points for white lines, robots and the ball are accurately extracted and classified by more precise scanning with both color and edge information. With the feature points extracted, self localization, robots recognition and ball tracking are performed. Experiments show that the proposed method is not only effective but also more robust to variable light.