Visual Recognition Of Robot Soccer

Robot soccer is a complex competitive game. Vision system is an important part in the robot system. Quickly and correctly identifying the position and the movement of the golf ball and robot vehicles is the foundation of decision making and control in winning robot competitions. In a robot system, the ball, robot vehicles and background are painted with specified colors and these different specified colors serve as indexes for identifying the moving objects.This paper researches the vision system based on RoboCup series, and puts forward a new algorithm to locate the moving objects rapidly and accurately.Based on existing technology in image segmentation and identification, and according to the characteristics of RoboCup. This thesis compares some common used algorithms, and proposes some improved algorithms. The main contributions include:First, the development of image segmentation and feature extraction in image process particularly in robot soccer system are reviewed.Second, the color information is commonly used in the robot soccer system. Considering the characteristics of robot soccer system, some different color spaces are analyzed and finally YUV color space is selected as the work space of the vision system, because it is closer to characteristics of human cognition, and can reduce the disturbance of the change of illumination. To meet the real-time demand in the robot soccer competition, this thesis proposes a color image segmentation method using BP ANN in YUV color space. The new algorithm can classify all used colors in one time, so the speed of the algorithm can be improved a lot.As the important part of the vision system, objects identification is the important part of our work. Most existing identification algorithms used in robot soccer vision system need to scan the image another time after all the pixels' color have been classified, which is time consuming. In order to save time, a binary search algorithm based on grid search before game and pre-decision in game is introduced in this paper. After locating the robot area, bring forward a recognition scheme which can recognize the robot's team, serial number, andorientation by the location and the color information of the color-pack on the robots. The new algorithm proves to be much faster and more exactly.Last, with respect to objects location which provides the actual position to the decision system and to meet the real-time demand, we provide a scheme in which geometrical transform is adopted after objects location to save time, while pre-processing before the recognition is omitted.