Design And Research Of Robot Disordered Grasping System Based On Binocular Structured Light

The disordered grasping of workpieces is one of the important research issues in the field of industrial manufacturing.In the face of disordered materials and complex environments in industrial production,the two-dimensional information obtained by traditional machine vision is difficult to meet this general demand.Binocular vision can effectively obtain the surrounding three-dimensional information,which greatly improves the flexibility of robot operation.Based on machine vision technology and the characteristics of high degree of freedom of industrial robots,this paper uses Gray code structured light combined with binocular camera to reconstruct three-dimensional point cloud,and develops a set of workpiece disordered grasping system with FANUC six-axis robot.The robot grasping based on binocular structured light is studied through experimental platform construction,theoretical calculation and experimental verification.Firstly,according to the project requirements,the system hardware is selected and the system software is designed.The visual imaging part of binocular structured light adopts the synchronous hard trigger design of light machine and camera,which improves the acquisition speed of structured light coded image and the stability of image acquisition,and realizes the automatic grasping function of robot.In the aspect of three-dimensional point cloud reconstruction,the binocular camera placement method of converging optical axis is selected,and the monocular and binocular image models of the camera are established.The binocular camera is calibrated by Zhang Zhengyou calibration method,and the robot coordinates are calibrated by binocular camera stereo matching.Aiming at the problem of long timeconsuming corresponding point search in epipolar constraint,the search step size improvement combined with Gray code condition is proposed to improve the speed of point cloud reconstruction.By removing the imaging shadow area to eliminate the wrong structured light coding,the quality of point cloud image is finally improved.In the aspect of three-dimensional point cloud processing,straight-through filtering,voxel filtering and statistical filtering are used to simplify and denoise the point cloud.RANSAC algorithm is used to extract the point cloud of the loading plane.The Euclidean clustering segmentation algorithm is compared with the region growth method based on normal and curvature,and the latter is selected to obtain better target segmentation effect.The distance between the centroid of each target point cloud and the plane is compared to select the top target in the stacking area.The key points of the template point cloud and the target point cloud are extracted by grid down sampling,the key points are described by FPFH feature descriptor,the coarse registration is carried out by SACIA,and then the target pose is obtained by ICP algorithm.Finally,the experimental platform is built to test the workpiece grasping.The workpiece grasping experiment is carried out under three conditions of no light,normal light and strong light,and the expected grasping effect is obtained.