Research Of Ilumination Invariant For Color Image Of Mirosot Based On Particle Swarm Optimization

The robot soccer is not only an interdisciplinary research area, referring to signalprocessing, multi-agent control, path planning, wireless communication and patternrecognition, but also a multidisciplinary technical which involves computer vision,robotics, intelligent control and image processing. It is a research platform which isfull of high technologies as challenges.Vision subsystem is very important in MiroSot soccer robot competition and thetarget recognition of color image is completed based on this system. The robot gamevenues is a standard color structuaral environment and color feature is considered tosolve problems. The color of the object under illumination change is different, whichseriously affects the color image segmentation algorithm, so illumination problem inrobot soccer becomes one of the bottlenecks.This paper presents the structure and key technologies of MiroSot soccer robotsystem and analyzes the impact of illumination change on the soccer visionsystem.This paper identifies the focus on dichromatic reflection model based on theresearch of light source, illumination model, color vision theory and color imageprocessing method. On the basis of dichromatic reflection model, combining thetraditional classical color model, all kinds of illumination change insensitive colormodels and components are studied. Finally, this paper comes up11illuminationinvariant color components, namely r, g, b, H, S, y, u, v, C1, C2and C3.In this paper, particle swarm optimization algorithm is used to select the bestillumination invariant color model. MiroSot soccer robot color samples are regardedas input datas, and within-class scatter matrix and between-class scatter matrix areused to establish fitness function.The larger the fitness value is, the better theclassification ability of color model is. The serial number of the color model as acoding parameter, inertia weights diminishing strategy used to renew particle, thecorrelation between components as constraint conditions, through continuous optimization process, this paper proposes rSC2color model suitable for MiroSotsoccer robot competition.Finally, the paper introduces MiroSot soccer robot platform, from this platformwe can acquire different illumination intensity pictures. The uSb color model isproposed according to the previous method, and this paper does the experiment ofobject recognition effectiveness of the newly proposed color models compared withthe traditional color models. The result shows that the new color model is better thanthe traditional color models and the rSC2color model based on particle swarmalgorithm has the best illumination invariant performance.