Research On Technology Of 3D Reconstruction Of Workpiece Based On Binocular Vision

At present,most factories at home and abroad use the teaching device to pull the welding torch to weld the workpiece.When the position of the workpiece is moved or the workpiece is changed,the teaching device must be retrained to perform welding.This method has low welding efficiency and high cost.Based on binocular vision to realize the processing of the three-dimensional shape of the workpiece,obtain the three-dimensional structure of the workpiece,and to pull the welding torch to realize the automatic welding of the workpiece,which is the current research hotspot and the demand of intelligent development.This thesis is based on the principle of binocular stereo vision parallax,according to the work characteristics of the welding workpiece,systematically study the key technical issues of system calibration,image preprocessing,workpiece target extraction,feature extraction and matching,3D reconstruction,etc.,and design a set of effective realization of the workpiece 3D Rebuild the hardware and software system.First,set up a binocular camera hardware measurement system to perform camera calibration.Camera calibration is the first step of the binocular stereo vision system.The accuracy of the calibration result is directly related to the accuracy of the 3D restoration.This article explains the principle of camera calibration,discusses the advantages and disadvantages of camera calibration technology,and chooses the Zhang Zhengyou calibration method based on Open CV as the calibration method of this article.Secondly,because the collected image of the workpiece contains a lot of irrelevant background and noise,the binocular images are not corrected,misaligned,etc.,this paper performs extreme calibration,cropping,filtering and other preprocessing on the collected binocular images,After the processed image,the workpiece target is extracted,which can completely preserve the workpiece area and eliminate the influence of irrelevant factors.Finally,in the case of a large number of mismatches in the matching method of scaleinvariant feature conversion,this paper proposes a RANSAC method based on homography matrix to correct the mismatch;for the matching result map is a sparse disparity map,the workpiece is three-dimensional The structure is difficult to determine.The theory of this thesis explains how to convert from sparse disparity map to dense disparity map,and finally 3D reconstruction of the workpiece based on the disparity map,the reconstruction results are in good agreement with the actual workpiece.