Research Of Illumination Invariant Color Image In Soccer Robots

Robot soccer match provides a remarkable and challenging topic for the development of intelligent robot, artificial intelligence and other areas including communication, electronics, sensors, machinery, image processing, etc. As a complex multi-agent system, one soccer robot system consists of control subsystem, communication subsystem, robot subsystem and vision subsystem. Recognizing target accurately and timely will impact the soccer robot's capability of locating target, planning path and synchronizing between each other.Robot soccer match field is a standard color structured environment in which every object has a specific color. So the specification of color looks like a better way, other than the specification of shape, frequency or statistical features, for a soccer robot to recognize the objectives, which is also a popular way currently. However the color of object varies when illumination changes, which will affect the effectiveness of target recognition. Therefore illumination impact becomes one of the critical problems in the soccer robot vision system objective recognition.The purpose of this paper is to reduce the impact of illumination variance on the soccer robot color recognition by the research of improving the color models. The paper identifies the focus on dichromatic reflection model based on the research of illumination color image processing methods. It identifies the focus of sequential forward feature selection method and genetic algorithm based on the research of suboptimal feature selection and optimal feature selection theories. The paper comes up 11 illumination invariant color components based on the research of the color components that are compliant to illumination invariance. The research highlights in this paper are as below:1) The paper comes up soccer robot vision image sequential forward feature selection method, and sets up color models for soccer robot that are illumination invariant. It comes up color model {uSb} that applies to semi-autonomous soccer robot, as well as color model {Syc2} that applies to autonomous soccer robot. The experiment analysis shows the new color models are better than the traditional color models in terms of the object recognition effectiveness.2) The paper comes up soccer robot vision image genetic algorithm method, and sets up color models for soccer robot that are illumination invariant. It comes up color model {rSc2} that applies to semi-autonomous soccer robot, as well as color model {yc2c3} that applies to autonomous soccer robot. The experiment analysis shows the new color models are better than the traditional color models in terms of the object recognition effectiveness.3) The paper comes up a set of generic algorithm parameters that are fit for the research of soccer robot color models. It does research of the 4 generic algorithm parameters and their impact on the result.4) The paper comes up the scope of the 4 generic algorithm parameters that work for the research on soccer robot color models. It summarizes the procedure of identifying the generic algorithm parameters, which also helps the research of generic algorithm in other areas.The paper does the experiment of object recognition effectiveness of the newly proposed color models comparing with the traditional color models. The result shows that the new color models are better than the traditional color models.