The Recognition Of Field Features Based On Vision System Of Humanoid Soccer Robot

RoboCup (World Cup Soccer Robot Game) has become the world’s largest robottechnology exchange platform since it was held for the first time, which played a hugerole in the development of the robots and its related technology. The need ofautonomous real-time confrontation during the game puts forward higher requirementon the real-time performance, robustness and accuracy of the system. As the soccerrobots obtain the outside information mainly through the visual system in the game,so it is very important for the soccer robots to identify the stadium quickly. Based onthe humanoid medium-sized group of soccer robot game platform, this paper did theresearch about the visual system of the humanoid soccer robots to realize the colorfulimages segment, edge extracting, line extracting, intersection recognition and goalpostidentification.With the analysis of the soccer robot competition environment and rules, thepaper divides the process flow of the visual system into three parts including the edgeextracting, the straight line extraction and the intersection recognition to complete therecognition of the characteristic stadium.For the part of the edge extraction of the visual system processing, the HSVcolor model is adopted to image segmentation. The HSV color model has its uniqueadvantages compared with RGB color model and YUV color model. The advantagesmainly reflected in that HSV color model is not sensitive to light, and it can be betteridentified by the HSV color model when the color is similar but the saturation isdifferent. For example, it can distinguish the color of dark red and magenta. Theresearch improves the segmentation process of the color images threshold, through Hchannel histogram combined with auxiliary S channel of HSV color model tocomplete the image segmentation of the field. By adopting the method of this paper,two or more colors intersection area in the stadium can be better segmentated, and bychannel H histogram, it’s easy to set the tone threshold segmentated by colorthreshold. In addition, a kind of edge extracting algorithm using four linear templatesis proposed in this paper, whose detection speed is relatively faster than Krischoperator and detection is more accurate. In addition, it has better continuity compared with Soble operator and Prewitt operator.In order to improve the real-time performance of Hough transform linearextracting and reduce its amount of calculation, the classification method of edgepixel clustering is proposed in this paper to realize the edge pixel clustering of thefield. It can substantially reduce the image pixels number of scanning of Houghtransform.According to analysis the type of intersection line, it is found that theintersection’s essence is formed by intersect linear, and the point of straight lineintersecting is called the intersection, which be seen as the initial point, to determinethe type of intersection line by judging the number of rays emitted from theintersection point. In this way it can reduce the Hopfield neural network learning andmemory templates and make learning memory templates to be more appeal.The color threshold segmentation of the stadium can be extracted effectively inthe goalpost to judge the left or right goalpost by the relative position of goalpost andthe intersection which detected by Hopfield neural network.