Research And Implementation Of Vision System Based On Four-legged Robots

Robot soccer is one of the main concerns in artificial intelligence research. As a key part of robot performance, vision system is the fundamental layers of autonomous robot system. It has been the focus of current international autonomous robot research area. Using advanced SONY 4-legged robot as hardware platform, this article will discuss its theories and implementation, and verify it by experiments.For RoboCup four-legged league, the accuracy of vision process is very crucial when the robots are always moving. This thesis presents an advanced vision system for the robots. To improve the precision of offline calibration, an advanced calibration algorithm is implemented based on YUV gamut, which has abandoned the traditional rectangle color space. Then a segment detection algorithm is applied to the segmented image, which is base on the offline thresholds. At the same time, in order to get the accurate body tilt, body roll and camera pose of the robot, the model of robot body is established through filter the accelerators. Based on this model, an object detection method is applied employing robot pose to recognize the objects. According to the robot model, the distances between objects and robots are calculated by the inverse transformation from two dimensions to three dimensions. The experimental result shows the validity of the method.According to the method presented, the vision module of the soccer system of four-legged league is constructed under the OPEN-R development environment offered by Sony Corporation. And the vision model played well in 2006 Germany Robocup.