Design and implementation of a cheap middle size soccer robot with wide vision scope for RoboCup

Omnidirectional catadioptrice vision systems that can capture the scene around themselves is a very hot research for robot localization. Costly hardware for catoptric image and complicated algorithms for unwarped image however hinder them to be developed as popular tools. In this paper, an omnidirectional catadioptrice vision system with cheap components and a fast algorithm to get unwarped image are designed and implemented in a middle-size soccer robot for RoboCup. In particular, the vision system engages a Web camera connected to USB port in a host computer and two convex mirrors for catoptric image captured in system memory. Compared with traditional vision system usually using expensive cameras and expensive grabber cards, the cheap vision system spends almost the same time for grabbing the image data. The relationship between real space and image space is formulated as a transformation matrix. In addition, experiments of a middle-size soccer robot to localize its position and orientation, to find the target ball and to shoot it for getting the score are also included.