A new efficient real-time arbitrary colored ball recognition method for a humanoid soccer robot

Abstract

The ball recognition algorithm in Robocup is challenging with regard to the computing complexity and false positive rate. This paper proposes a new method to deal with the real-time arbitrary colored ball recognition for a humanoid soccer robot. At the beginning, the image is matched with a look-up table under the HSV color space. After we get the edge pixels information, the ball is recognized by the contour feature method (CF) or the modified Gradient Circle Hough Transform method (GCHT) based on different contour conditions. The CF approach proved to be fast and effective in non-intersection condition. While the modified GCHT as a complement is more reliable under intersection condition. The novelty of this ball recognition method is the combination of the CF and modified GCHT method and taking advantages from each of them. In addition, the new ball recognition method focuses on the arbitrary colored ball recognition rather than the single uniform colored ball detection. Besides, the variance of illumination, the shadows and occlusion have also been considered. A great number of experiments have been carried out showing that the proposed new method achieves a balanced point with the consideration of timeliness and efficiency.