Designed Soccer Robot Vision System Based On Color Images

Holding soccer robot game has a great benefit, i.e. it can combine theories of the Artificial Intelligence(AI) and Computer Vision with practice. We can prove those theories and explore how multiple intelligent robots cooperate in uncertain dynamic invironment in game. We also can perfect those theories and speed up the process that those theories are commercialized and industrialized by the game. The real-time of the vision system of soccer robot is a state of the art research task in image processing. Soccer robot provides an instrument for the experimentation of the Artificial Intelligence(AI) and Computer Vision. The vision system is the most essential sub-system of soccer robot. The significance of studying vision system is apparent.Color has proven to be a powerful visual cue and image feature for computer vision. It is computationally feasible, and it overcomes many problems associated with geometric cues and gray scale intensity. The capability to endure rotation in depth, and scale and resolution changes makes it an attractive approach to image segmentation. Despite the many advantages, certain problems are also associated with color. The fact that the perceived color is greatly dependent on illumination is one of the major challenges in color-based segmentation.In this paper, a real-time color-based vision system is designed. Mobile robot has the real-time perceptual capability by employing the vision system. In some literature, we know some color-based image segmentation and tracking algorithms and analyze some characteristics of them, we discuss some issues that may affect the quality of color image, and so on. It is difficult to achieve a real-time and self-adaptive vision system. Furthermore, it is necessary to enhance robust and precision for vision sysem.In the practical section, three pixel labling algorithms are introduced, i.e. invariable threshold method, adaptive threshold method, and statistical model method. At the same time, three region segmentation algorithms also are introduced, i.e. global-based segmentation algorithm, tree-based region find in RLE map image, and statistic-based region feature computation algorithm. Experiments and results indicate that those algorithm are computationally feasible and effective. These is sufficient to achieve a real-time and self-adaptive vision system and enhance robust and precision. The implemented system, which based on the proposed architecture, is robust too.