Research On Target Recognition And Positioning And Robot Grasping System Based On Binocular Stereo Vision

Industrial robots are widely used in product grabbing on conveyor belts.At present,the common application scenarios are:the product on conveyor belts is single and the product cycle is determined,and then the robot grabbing is taught accurately.But there are more than one product on the conveyor belt of some production lines(such as the lamp housing and mask production line of the automobile lamp factory),and the product cycle is uncertain,so the traditional robot will fail to grasp.The combination of vision technology and robot technology will effectively solve this problem,especially that binocular vision can obtain three-dimensional pose information of different objects on the conveyor belt,which will effectively solve the problem of robot grasping when there are two or more products on the conveyor belt.Based on the ABB IRB1200 robot platform,this paper constructs a binocular vision target location and grasping system,and studies the technology of target location,recognition and robot grasping based on binocular vision.The main contents of this paper include:(1)The domestic and foreign research status of binocular vision robot grasping is summarized.The selection basis of the hardware(robot,industrial camera,lens,light source and robot terminal grabber)is studied,and the core hardware selection of the whole visual system is determined.Secondly,according to the camera’s common field of view and the measurement error of depth direction of binocular vision,the installation mode of the camera and the baseline distance as well as the shooting distance are reasonably determined.(2)In the visual system calibration part,the camera imaging principle and the calibration principle are first studied,and the camera calibration experiments are carried out with OpenCV and Matlab to obtain the internal and external parameters of the camera and the location parameters of binocular vision system,and the accuracy of the results obtained by the two calibration methods is analyzed by the method of distance measurement.Secondly,the robot hand-eye model is studied,and hand-eye calibration parameters are obtained through a simple hand-eye calibration method.(3)In the part of target recognition,the method of target recognition based on SIFT algorithm is studied in depth.Aiming at the problem of low accuracy of target recognition found in the study,the process of target recognition is optimized by combining RANSAC algorithm to improve the recognition accuracy.(4)In the part of target location,the edge contour of the template image is obtained by Canny edge detection method,and the centroid of the target in the template image is obtained according to the contour,and the spatial coordinates of the centroid of the target are identified by SIFT algorithm.On the basis of obtaining the object’s centroid,the position and attitude information of the object is determined by determining the slope of the bus bar of the object in the scene graph.Finally,the target location is accomplished by combining the two.(5)Finally,based on OpenCV and VS2015,a binocular visual positioning and grabbing software is developed by using MFC programming.After analyzing the grabbing reliability,grabbing experiments are carried out,and the centroid Z coordinates are optimized to improve the grabbing success rate.