Research Of Vision System For Soccer Robot

Nowadays, research of robotics is increasingly thorough. The dream of giving a similar human visual perception to robots is realized step by step. In this paper, RoboCup soccer robot is used as the research object, and related research for omnidirectional machine vision has been finished.By comparing the methods of existing omnidirectional vision and the advantages and disadvantages of five kinds of reflection mirrors, the hyperbolic mirror is selected as a reflector to form omnidirectional vision system. Then an analysis of its imaging principle is made for designing the parameters of the hyperbolic mirror. And a kind of calibration method based on experimental measurement is designed. The calibration method is suitable for most models of omnidirectional vision.In the aspect of image segmentation, a kind of Otsu method is applied to getting multi-threshold value for color images automatically. And then an improved segmentation method is proposed, according to the characteristics of color images of the RoboCup soccer robot, which is, in the image areas with middle-high saturation and low saturation, using different components of color space for threshold selection. Experiments proved that this method is rapid, efficient and can be adaptive to updating threshold and better robustness.Then, The process of tracking for moving target in video image is studied. Kalman filtering algorithm to tracking moving target is analyzed, which can predict the moving path. And so is Mean Shift algorithm, which has nothing to do with movement pattern, has a little amount of calculation and can be adaptive to target’s part of being covered and deformation. And a method is proposed, which is to judge whether the target is covered. Then Kalman filtering and Mean Shift algorithm are combined to achieve target tracking accurately for the situation of target being completely occluded or changing motion forms suddenly.