Study On Some Problems Of Vision For RoboCup Four-Legged League

The intelligent robot soccer competition ——the RoboCup starts from 1997. As the top legged robot soccer game, the RoboCup Four-Legged League is appealing more and more universities and scientific institute because it concerns several researching fields such as machine vision, artificial intelligence and multi-robot cooperation. This paper focuses on some problems on vision in the RoboCup Four-Legged League. Because of the importance of the program structure, the paper firstly introduces the structure of the Tsinghua University program for Four-legged League, followed by detailed explanation of each module and their relationships. The vision module and its affiliated modules is stressed and some problems in the module are proposed as the focus of following chapters. As for color processing, the currently popular Color Lookup Table algorithm is accurate and efficient in fixed illuminating environment and cannot adapt varying illumination. However, the trend of Four-Legged League is to illuminate the field with natural sunlight, thus a new algorithm that makes the robot adopt varying illumination is worth researching. This paper proposes a segmenting method using histogram thresholding technology, and the advantage is proved by the comparison between the result of same images processed by different algorithms. The most important object during the game is the ball, whose position when near the robot is not evaluated accurately by any programs. Applying circle fitting, the method in this paper can achieve plausible result even though the ball is not complete in the image. Although the method is somewhat complicated, the enhanced processing ability of the new type robot makes the new method applicable. As the markers around the field are to be canceled gradually, the robot needs to find something else to help positioning. We have chosen the field lines as reference. Points on the lines are extracted from the image by color variation, and the types and parameters of the lines are calculated. Experiments have proved that the algorithm is applicable and reliable.