Research On 3D Reconstruction System Based On Zoom Binocular Stereo Vision

With the continuous development of computer vision,the demand for acquiring the three-dimensional point cloud of the target to be measured quickly and accurately is increasing.As an important way to achieve 3D reconstruction,binocular stereo vision has the characteristics of low cost,high efficiency,and strong flexibility in addition to its human-like processing goals,so it is widely used in business,education,culture and other fields.Compared with traditional fixed focus cameras,zoom cameras have the advantages of high flexibility and strong control.It can change the focal length according to the scene changes,so as to obtain a more accurate and complete three-dimensional point cloud.Therefore,the 3D reconstruction system based on zoom binocular stereo vision has extremely important research significance.This paper studies and designs a three-dimensional reconstruction system based on zoom binocular stereo vision,which focuses on camera calibration,feature point extraction and matching,and three-dimensional reconstruction.The main research contents of this article are summarized as follows:(1)Aiming at the problems of large workload and low accuracy of the zoom camera calibration,a two-step calibration method based on fixed focus-zoom binocular cameras is proposed.Firstly,we can use the Zhang's calibration method to perform stereo calibration to obtain the initial internal and external parameters of the binocular camera.Secondly,according to the corresponding point relationships on the four imaging planes before and after zooming,we can solve the internal and external parameters after zooming of the binocular camera.Finally,according to the internal and external parameters solved and considering the distortion parameters,nonlinear optimization is used to estimate all the parameters of the binocular system.Experimental results show that the method in this paper is simple and easy to operate and has high calibration accuracy.(2)Aiming at the problems of low feature extraction and low matching accuracy of existing methods,a feature matching method combining oFAST-SURF and GMS is designed.In the feature extraction part,oFAST(Oriented FAST)detector is used for feature detection,and SURF descriptor is used for feature description.The feature matching part uses bidirectional constraints instead of unidirectional constraints based on the GMS algorithm,and uses the RANSAC algorithm based on epipolar constraint to filter feature matching pairs.Experiments show that the method in this paper has improved the number of feature points and the accuracy of matching.(3)Design and implement a 3D reconstruction system based on zoom binocular stereo vision.The system is mainly composed of six modules.They are image acquisition,camera calibration,stereo matching,sparse point cloud reconstruction,point cloud stitching and dense visualization.Firstly,the system firstly uses the calibration method and feature matching method of this paper to obtain the three-dimensional sparse point cloud at the initial focal length and the variable focal length respectively.Secondly,the system uses ICP algorithm to stitch the two point clouds to obtain a more complete point cloud.Finally,the Delaunay triangulation method and OpenGL technology are used to achieve dense visualization.At the same time,the system uses the images collected in the real scene for experiments.The experiments prove that the three-dimensional reconstruction system in this paper is feasible and effective.