Research And Design Of Vision Systems For Soccer Robots In RoboCup Middle-Size Matches

As a new research branch of autonomous robot, the research of soccer robots has attracted more and more attention. The main idea of the robot soccer competition is to develop relative technologies of robots by supplying a standard competition platform. The design of soccer robots involves many subjects such as mechanics, electrics, automation, pattern recognition, image processing, and artificial intelligence. Generally speaking, there are four parts in a soccer robot. They are decision-making subsystem, vision subsystem, wireless communication subsystem and motion control subsystem. The vision subsystem, as an important approach of acquiring environment information, is critical to the performance of the whole system. In the RoboCup competition, although the object is known , because of its rivalry, the vision subsystem must be robust enough real-time to insure the rapid reaction to changes of the environment and accurate enough to measure the distance between robots and objects and robust enough to adapt itself to different environment, all of these are great challenges for the design of vision subsystems.In this thesis, we introduce the origin and development of soccer robot, relative knowledge of middle size soccer robot competition, the important and difficult issues on designing vision systems. After a general description of robot vision systems, some technologies about robot vision systems are discussed including image preprocessing, color image segmentation, image representation and object recognition and so on.We design our vision system based on Omni-vision and Front vision. The main work I have finished include: choosing the HSL color space which is more suitable for image segmentation, developing a method of pixel classification with lookup tables that remarkably reduces calculation used in the process of image segmentation, adopting the RLE (Run-Length Encoding) to reduce the calculation furthermore, proposing a new algorithm for recognition of gates and corner posts, designing a linear interpolation algorithm for Omni-vision and a grid interpolation algorithm for Front vision to calculate world distances. The results show that our vision system is a real-time and effective system. We also discuss the color threshold calibration problem which is a key issue of vision systems and propose a cluster segmentation algorithm based on HSL color space solve the problem.There are still some drawbacks of the system. The adaptability to changes of ambient light should be improved, and the calculus of distances should be more accurate.