Research On Robot Classification-and-grab Technology Based On Machine Vision

The degree of automation is higher and higher in modern society, and the robot is widely used in all fields. The robotics, based on machine vision has become a popular direction in robot development. Based on MOTOMAN-UP6 robot, this paper structures a classification-and-grab system which includes CCD camera, image acquisition card, computer, robot and so on. The system gets image through camera, and converts analog images into digital images, then it deals with the digital images by using developed software to get the target classification and areal coordinate. At last, the robot can achieve motion and finish classification-and-grab operation.At first, this paper studies the camera calibration. It implements the Direct Liner Transform algorithm, perspective projection matrix algorithm and Zhengyou Zhang method on Matlab and finally takes Zhengyou Zhang method as the method of the camera calibration. In the identification of static objects, this paper does threshold to images, extracts the target area and perimeter and classifies the target through the method of clustering. In the tracking of dynamic target, it extracts the moving target by the difference of background algorithm, mainly studying data association based on nearest neighbor algorithm, it suggests a way that combines the Kalman prediction with the template match algorithm in dynamic target recognition, which can shorten the matching rang and enhance the real-time.Through several experiments, debugs and modifications, the author finishes a primary classification-and-grab system which is based on machine vision. The experiment verifies the feasibility of the classification-and-grab system, which provides a useful reference for the robot classification-and-grab technology. At the same time, the experiment also reflects the disadvantage of the system and the direction need to be strengthened in future.