Study On Robot Capture-Oriented Binocular Stereo Vision Detection

As robots have been finding more and more applications in many fields of modern industry such as welding, spraying, assembly, transporting and so on, vision, as its most important kind of input information, is mainly used to detect object and manipulator. In a robot-oriented vision servo system, for the robot to grab an object, the vision system requires not only the 3D information about the object, but also the 3D posture of the manipulator, so that the robot controller can well perform path planning. What with cluttered background and illumination disturbance under real circumstances, it could usually be difficult for a vision system to work properly. In this paper, therefore, we make some studies on this, in an attempt to build a binocular vision detection system for the 6 DOF serial robot, which is then analyzed, some experimental results acquired eventually.The whole architecture of the vision servo system is presented, involving the hardware platform and relevant submodules, which can help improve the Real-time performance of the system. Video frames are sampled by using the DirectShow technology, and are then changed into IplImage data format for OpenCV's uses.This paper circles around the detection of object and manipulator. With a small red ball as our object, three algorithms, including Hough transformation, color recognition, and motion-based method, are here analyzed and compared with each other, the result showing that to detect the ball by using its motion information together with color feature is the best choice. On account of the complexity of the manipulator itself, we apply the so-called Haar-like features based AdaBoost algorithm to the detection of the robot, which yields a good outcome.With the established mathematical model of the robot's 3D posture, the 3D posture parameters, are then secured, which will serve as feedback for the robot controller.The summary and vistas of the whole work are listed in the end.