Research On Image-based 3d Reconstruction

In recent years,more and more attention has been paid to 3D technology-related products.The 3D reconstruction technology has been continuously developed and has become increasingly important in computer vision.As an important branch of computer vision research,which can directly imitate the human vision to process the objective scene and has the advantages of simple equipment,low cost,easy to use,etc.It has been widely used in the market.The study of image-based 3D reconstruction has been a hot research issue in computer vision due to its diversity of applications.The main research contents of this paper based on the three-dimensional image reconstruction are camera calibration and stereo matching.In the process of camera calibration,because this paper is based on the parallel vision model,the image data acquisition code is first written to achieve the simultaneous acquisition of the image data in the two cameras.The internal and external parameters of the camera are obtained by using the calibration method between the self-calibration method and the traditional calibration method in OpenCV(Open Source Computer Vision Library),and the stereo correction was performed using the Bouget polar correction principle.Finally,the correctness of the calibration result is verified by the projection error analysis of the calibration result of the Matlab camera calibration toolbox.In stereo matching part,this article studied feature-based and semi-global matching based stereo matching algorithm.The algorithm proposed in this paper is based on the extraction of SURF features,feature matching through KD-tree,and random sampling consistency algorithm and distance constraint-based method to eliminate false matching points.Using self-built binocular vision system to captureimage pairs to verify correctness of image correction and matching results.Experimental results show that this algorithm is good for extracting evenly distributed feature points and reduce the matching point aggregation.This paper also presents an improved algorithm for semi-global stereo matching based on matching costs,by using the combined method of gray level and gradient information of similarity measure,and combines the binomial interpolation method to optimize the disparity map,which can fully preserve the image edge information.At the end of this paper,based on the established binocular stereo vision system,according to the triangulation principle and the Delaunay triangulation method,the actual scene is reconstructed by the indoor shooting left and right images,and the final 3D model is obtained.