Binocular Parallax Reconstruction Methods Of 3D Surface And Its Experiment Research

Computer vision technology has been widely applied in industrial production,such as the UAV real-time navigation and positioning,fingerprint identification,face recognition.Compared to the human eye,computer vision technology posses a higher measurement accuracy in the distance and size measurement.Measurement techniques based on computer vision mainly include monocular vision and binocular vision measurement.Because the binocular vision is more like the human eye,it has become a research focus.Presently,binocular vision measurement has been widely used to the intelligent distance-measuring,the intelligent positioning system,VR,and 3D reconstruction.And it has the advantages of simple system structure,non-contact and automation.However,it also have some shortages such as low efficiency for the processing of big data,poor robustness,limited precision and applicable scope.Around the vision measurement of 3D surface,this thesis puts forward the reconstruction and measurement method based on binocular parallax graphs.Main aims include the robustness research of camera calibration,image preprocessing,the reconstruction of disparity map and 3D reconstruction.The main works are listed as follows:1.The robust binocular camera calibration.To solve the robustness and real-time problem of camera calibration,the initial parameters are obtained by the Zhang Zhengyou's plane calibration method.Then,the initial parameters are optimized by the LM algorithm.Finally,this thesis analyzes the reasons influencing the stability of the calibration.2.Image preprocessing algorithm of the 3D surface.The image denoising method and the image segmentation method are studied respectively.The target image is processed by the Gaussian filtering algorithm and Grabcut algorithm.In order to enhance images' anti-noise and provide more information for the reconstruction of disparity map,the intensity is merged with the gradient.3.The reconstruction of disparity map and 3D reconstruction.This thesis research the construction method of disparity map based on all pixels.On basis of the principle of the parallel binocular vision,we rectify pair of images.A graph-cut-based matching algorithm is used to disparity map reconstruction on constrains of parallax continuity and consistency constraints.Finally,we reconstruct the 3D surface based on the dense disparity map.4.Experimental test and analysis.The 3D surface measurement system consists of two CCD cameras,image acquisition card,a platform of three dimensional motion measurement and so on.Firstly,we conduct experiments on the robustness influences of binocular camera calibration.we finish the binocular camera calibration.Secondly,experiments on image preprocessing algorithm are carried out by using standard test images and the regular specimen images.Finally,based on the gray image and the IG image,we obtain the reconstruction of disparity map using this thesis' s algorithm.The disparity map using this thesis' s algorithm compare with the algorithm based on block-matching algorithm.Consequently,we realize reconstruction of the 3D surface.Then,this thesis analysis the proposed method's feasibility.In this thesis,recommendations include the image sample acquisition environment,the number of image samples and calibration plate selection are proposed to improve the camera calibration robustness.A graph-cut-based matching algorithm for the reconstruction of disparity map posses better result than algorithm based on block-matching.The The IG image has better noise immunity for the disparity map reconstruction.Finally,the average error of 3D reconstruction is less than 0.2mm.The satisfactory results are obtained.