A visual compass for robot soccer

This paper proposes a cheap image processing method for visual compass in the robot soccer context. An offline map is constructed by using color transition histogram features. When a feature is extracted, we calculate maximum likelihood orientation of that feature over the pre-calculated map. This instantaneous orientation information updates orientation belief with given mixing constant. We presented a fast method to disambiguate goals in symetric field in RoboCup standart platform league. The method can correctly identify the goals with 94% accuracy for Webots simulations experiments and 86% accuracy for real robot experiments.