A New Omni-vision Based Self-localization Method for Soccer Robot

Abstract

Self-localization is one of the key technologies in soccer robot system. However in the soccer robot system based on the omni-vision, it is hard to match the features extracted from the image to the real features existing in the world due to the large distortion in omni-vision images. This has become the main obstacle to precise self-localization based on the omni-vision. To solve this problem, a new method with the fusion of multiple kinds of vision information including the color, edge and the space relationship between them is proposed for feature matching. And at the same time, to solve the problem of "kidnapped robot" which often happens in Soccer Robot system due to the collision between the robots, a new localization method called Mixture Sampling-based Evolutionary Monte Carlo Localization (MS-EMCL) is proposed, which applies mixture sampling technology to draw samples from the most newly observed information and applies evolution operators, cross and mutation introduced from the genetic algorithm, to make samples move towards regions with high post density so that the samples can represent the pose of the robot much better after robot being "kidnapped".