Research On Target Recognition And Pose Determination In Grasping Stacked Box Based On Binocular Vision

With the rapid development of computer science,as an important branch of machine vision,binocular stereo vision has been applied in all fields of production and daily life.The vision guidance technology based on binocular stereo vision can be upgraded from the recognition position of the two-dimensional space to the three-dimensional space,and the three-dimensional information of the target can be obtained,thereby realizing that more complicated grasping and sorting can be achieved in the environment with higher degree of freedom.In industrial production,the level of automation represents the level of production.At present,many robots that capture and sort can complete the identification and positioning of the target by visual guidance technology,but most of them are two-dimensional vision guidance technology.Aiming at this problem,this paper use the research of three-dimensional vision guidance technology based on binocular stereo vision to identify and locate the box in the stacked state,after obtained the position and posture information of the box,grabbed the three-dimensional space.At present,the visual guidance technology of binocular stereo vision is an important subject,however,at the same time it faces many challenges.In the research of visual guidance technology,binocular stereo vision technology has been unable to balance real-time efficiency and high accuracy.It can complete visual guidance work under a single working condition,but it's accuracy was greatly reduced while under a complicated stacking condition,and it's system robustness was not strong.In allusion to the above issues,in this paper,the visual guidance technology applied in stacking box grabbing is as follows:(1)For Zhang Zhengyou's calibration method,only the influence of radial distortion is considered,which makes the calibration result inaccurate.This paper proposes the calibration method of adding the tangential distortion coefficient based on the Zhang's calibration method.In order to calibrate the algorithm,the Zhang's calibration method only considers the most influential radial distortion.However,since the lens cannot be strictly parallel to the imaging plane during camera assembly,tangential distortion can be introduced.In order to improve the calibration accuracy,according to the distortion principle,the camera is calibrated considering the effects of radial and tangential distortion.(2)For the efficiency and accuracy of stereo matching,this paper proposes an improvement based on the SGBM algorithm,which use the RANSAC iterative algorithm to reject the mismatch.The parallax map was processed by using the Delaunay triangulation technique to make the parallax result more accurate.Compared with the current matching algorithm of SGBM and ELAS with high matching efficiency and high matching precision,the SSG algorithm was improved by combining ELAS algorithm processing flow.(3)For the problem of position estimation of stacking box,this paper make an in-depth study on 3D reconstruction technology,and propose a method based on parallax information to estimate the pose of stacking box.Through the Blob image processing technology,split the foreground plane of the stacking box you need to grab,extract the parallax information and the fitted plane normal vector in the plane.The box can be grabbed according to the obtained position information.