Research On Technology Of Target Location Based On Vision Robot

With the continuous improvement of science and technology of our country, technology of vision robot as the forefront of the development of contemporary high technology has became the hot topic of the current research. As one of the important means of robot perception information, vision technology not only enhances the ability of adapting to the environment and perceiving environment of the robot, but also improves its work efficiency. But, as an important branch of robot, industry robot is still facing with a problem, which is how to develop an appropriate vision technology, and then solve intelligence and informationization of the robot. Recognition and localization of target in industrial environment are the most common operation of industry robot, and they are also important parts of flexible manufacturing system(FMS) and computer integrated manufacturing system(CIMS). Therefore, it is of great theoretical and practical significance to study recognition and localization of target of vision industry robot.This paper will focus on researching how to achieve 3D position of industry target and spatial attitude acquisition using vision technology, and achieve sorting and grabbing of target by combining vision technology and industry robot. In order to accomplish this goal, this paper detailed overall process and technical methods of vision technology-based industry target recognition and localization including obtaining target 2D position information from images, acquiring depth information based on binocular vision and preliminary research and simulation experiment based on target attitude acquisition of spatial point cloud. The main research work is as follows:1) The basic theory and technical process of vision localization were detailed, including: the imaging model of the camera, the conversion relationship between the different vision coordinate systems, the theory and methods of the camera calibration and the theory of binocular stereo vision-based measuring distance. Meanwhile, by comparing with two kinds of calibration methods based on Open CV and MATLAB, a conclusion that the calibration method based on Open CV had good program portability on the basis of guaranteeing calibration accuracy was obtained.2) In recognition and localization research of target, target firstly was recognized by the edge gradient-based shape matching algorithm, and recognition speed was improved by the sub-pixel-based edge thinning algorithm, target accurate localization information in the two-dimensional surface was obtained by combining the camera calibration and coordinate transformation results, then, on the basis of this, depth information was acquired by the triangle measuring theory of binocular vision.3) After acquiring the 3D coordinates of the target, a method of obtaining target space attitude was proposed using the rigid transformation relationship between the point cloud of model and the point cloud of target, and this rigid transformation was calculated using the iterative closest point algorithm(ICP)-based 3D point cloud registration method. Finally, according to the effect of the proposed method, a MATLAB simulation was carried out.4) A grasping experiment was implemented by combining vision technology and industrial robot, on the basis of conversing vision coordinate to robot coordinate, the target positioning result is transmitted to the robot control system, guiding the robot to complete the target positioning and grasping. The experimental result showed that the robot grasping success rate basically could meet the actual demand.