Research On Mid-sized Soccer Robot's Vision Recognition And Localization

In this thesis, the vision system of the middle-size league of RoboCup is studied. Through the omni and forward lens, the robot's target recognition and localization based on RoboCup Middle-Size Soccer Robot are given.In this thesis color is mainly used to recognize target. The process of recognition includes choice of color model, image preprocessing, image segmentation and image recognition. Through the analysis of various color space models, we choose the HSI model which is not intensive for the illumination disturbance et al.. We used the median filtering to smooth the color image, and reduced the image noise. The style for the segmentation of 2-D gray-level histogram is improved, mainly used improved 3-D gray-level histogram segmentation to make the robot recognize different targets with the H IS color model, and proved this way of image segmentation is efficient. The field's white lines of the Soccer Robot are used to robot self-localization.Firstly, the white lines of the field are detected through an improved Hough transform voting scheme. After Canny edge detection, each collinear set of pixels is clustered together in the edge image. Next, we find a best fitting line for each cluster and estimate the uncertainties from pixels of each pixel set corresponding to the best line. An elliptical Gaussian kernel of the fitting line is then calculated according to these uncertainties to represent the importance of it. The voting process will be performed on major lines only. With transformations between 3D spaces and the imaging principal, we can estimate the position of each line target in the image, and then perform fast Hough transform on the region around the position and search the line. This can significantly reduce the computing consumption; and also improve the real-time line detection of Hough transform. Based on this we can detect the white lines of the field in the real time, so the method of robot self-localization through the detected white lines based on the omni vision system is given. By the experiment on Voyager III platform showed, the method of robot's self-localization has well ability, rapidity and robustness.The precision can be accepted.