Research On 3D Reconstruction Method Of Industrial Binocular Endoscope

In today's intelligent era,precision and complex industrial products put forward higher requirements for detection technology,and the demand for detection technology automation is more and more urgent.An industrial endoscope is critical testing equipment of industrial testing automation.But at present,the threedimensional measurement technology of industrial endoscope based on machine vision is limited due to technical and cost reasons.The two-dimensional industrial endoscope is mainly used in industrial inspection,which seriously affects the accuracy of damage identification and maintenance plan,and greatly reduces the maintenance efficiency.In order to solve this problem,this paper focuses on the research of an image matching algorithm which is easy to be applied to threedimensional industrial endoscope.Firstly,on the basis of studying the principle of binocular vision,we design a binocular vision system,select the appropriate model for the camera,lens,optical endoscope,master computer and other hardware facilities,build a hardware platform,select software development tools,and introduce its functions.Secondly,the principle and process of Zhang Zhengyou's calibration method and the transformation relationship between coordinate systems are described.The camera calibration is carried out,and the parameters of camera calibration are solved by experiments.Thirdly,Gaussian filter is selected to preprocess the acquired binocular image.Considering the advantages and disadvantages of various stereo matching methods,an improved Oriented FAST and Rotated BRIEF(ORB)algorithm is proposed.In the feature point detection stage,ORB algorithm and Speeded Up Robust Feature(SURF)algorithm detects feature points at the same time,o FAST and SURF algorithm detects left and right image feature points,and r BRIEF descriptor describes feature points;in the stereo matching stage,Hamming distance is used for rough matching of feature points,and epipolar constraint is introduced to filter feature points and carry out fine matching,so as to reduce the matching search range,speed up the matching speed and improve the matching accuracy.This paper compares the improved ORB algorithm with Scale-invariant Feature Transform(SIFT)algorithm,SURF algorithm and ORB algorithm,and verifies the scale invariance of the improved ORB algorithm.Finally,based on the analysis of the depth information estimation method based on binocular vision,the least square method is used to calculate the three-dimensional coordinates of the matching points,and the Bowyer-Watson algorithm,which is the most commonly used point by point insertion method,is used to generate the Delaunay triangulation,and finally the three-dimensional model of the measured object is obtained.The experimental results show that the improved ORB algorithm is 3.9 times faster than SIFT algorithm and 2.7 times faster than SURF algorithm.In terms of accuracy,the improved ORB algorithm is 2.9 times of SIFT algorithm,3.37 times of SURF algorithm and 1.2 times of traditional ORB algorithm.In addition,the improved ORB algorithm has scale invariance.The improved ORB algorithm studied in this paper can be applied to image matching recognition,target tracking,3D reconstruction and industrial 3D detection.