Research And Development Of Intelligent Sorting System For Parallel Robot Based On Binocular Vision

In modern industrial production,automatic grasping of workpieces is an important sign of intelligence.At present,most robots are faced with a fixed environment,repeating a single task.Binocular vision technology can entitle robots the function similar to human eyes,and carry out 3D reconstruction of the target object according to the parallax principle,so as to improve the industrial production efficiency and automation level.This thesis will study the object 3D reconstruction and robot grasping technology based on binocular vision.Firstly,the transformation of camera pinhole imaging model and monocular coordinate system are studied,the causes of camera distortion and the basic principle of binocular vision are analyzed.Through the selection and design of main components of the visual platform,a parallel binocular vision system is designed and constructed.The software environment of the system is analyzed,and the calibration and accuracy of the vision system are verified by Zhang Zhengyou calibration method.Secondly,three preprocessing methods of image smoothing,graying and information enhancement are studied.Different template matching methods are analyzed,The feature point template matching method based on SURF descriptor is used to identify the target object.RANSAC algorithm is used to remove the false matching point pairs.Through the target recognition experiments in different scenes,the accuracy of the proposed algorithm is proved to be preferable.Then,SGBM semi-global stereo matching algorithm is used to match the target object.The parallax map obtained by left and right image matching is reconstructed in Visual Studio 2017 software environment,and the reconstruction results are evaluated.A robot grasping strategy is developed based on 3D centroid reconstruction of the target workpieces in this thesis.Finally,according to the sorting needs,the sorting system experiment platform composed of parallel robot and binocular vision system is designed.The communication between PC and parallel robot is established by TCP/IP protocol.Eye-to-hand model is used to complete the hand-eye calibration of the sorting system,and the positioning results of the vision system are sent to the robot.The grasping experiment of milk slice bottle and cylinder block was carried out by parallel robot,It is proved that the method used in this thesis can accurately and stably realize the positioning and grasping of the target object.