Research On Spatial Pose Detection Of Box Workpiece Based On Binocular Vision

As a current research hotspot,machine vision technology has been more and more widely used in the process of industrial automation production.In the traditional mode of assembly line,the robot often has to complete the grasping operation of the workpiece,and,in this process,the acquisition of the spatial position of the workpiece as well as the accuracy of the posture calculation are the prerequisite and key to the implementation of the grasping operation.When the grasping task requires the spatial pose information of the target,unlike monocular vision which can only obtain the two-dimensional information of the target,Binocular vision has the ability to obtain three-dimensional information of the target.For problem in target recognition and spatial pose detection in the process of box workpiece spatial pose detection,a spatial pose detection scheme of box workpiece based on binocular stereo vision system was proposed.The plan aims to identify the box workpiece from the complex background picture and quickly and accurately determine the spatial pose of the box workpiece.Combining the geometric constraints of the box workpiece itself,the characteristic corner points are extracted,and then the appropriate stereo matching algorithm is used to determine the correspondence between the characteristic corner points of the left and right images.After that,combined with the binocular vision theory the spatial pose of the box workpiece is quickly and accurately determine.In the calibration part of the binocular camera,in order to realize the above scheme,not only the imaging model of the camera and the conversion relationship between its four reference coordinate systems were studied,but also theory of binocular vision 3D reconstruction was introduced.The Zhang’s calibration method was introduced in detail and the internal and external parameter calibration of the binocular camera in the eye-tohand installation mode was completed through the calibration experiment.In the part of target recognition,starting from image feature extraction,how to use SURF operator to extract and build image feature descriptors was studied.An offline template library of the target was established by collecting pictures of the target workpiece at different angles,and then the area where the target workpiece is located in multiple scenarios was determined through the homography matrix conversion relationship between the template picture and the scene picture.In the workpiece pose detection part,Corner features were used to characterize the pose information of the box workpiece.In order to solve the problem of misdetection and missed detection of traditional corner detection algorithms,the coordinates of the intersection of the edge fitting straight lines were used as the characteristic corner points,which improves the detection accuracy of the corner points.A local stereo matching algorithm based on epipolar constraints was used to determine the correspondence between the feature corners of the left and right scene images.Finally,the spatial pose of the box workpiece was calculated by the mapping relationship between the corner pixel coordinates and the user coordinates.The workpiece pose detection experiment is designed to verify the effectiveness of the proposed scheme.At different depths and distances,the estimated size and pose angle of the box workpiece obtained form the system were compared with the actual value.The experimental data show that the estimated error of the workpiece size does not exceed2 mm,and the average error of the pose angle does not exceed 2°,which verifies the feasibility of the proposed scheme.