Research And Software Development Of 3D Object Reconstruction Based On Monocular Camera

Three-dimensional reconstruction based on monocular camera has broad application prospects due to its convenience.Among them,various methods based on camera parameter estimation are popular because many fields of application rely on partial key point depth estimation.However,it is difficult to perform fine-grained restoration of the three-dimensional point set of objects in the scene using only a small number of images.To this end,this paper has made further research on this,and the main work results are summarized as follows:(1)A parameter estimation method for monocular camera based on smaller reprojection error is implemented.The accuracy of results calibrated by general Zhang’s calibration method is often affected by the number of images of the input calibration object.Therefore,this paper uses different number of calibration plate images to carry out many experiments,and obtains a number of convergence range.In this range,to select several calibration images automatically and randomly.After eliminating the large error corners in all calibration images,then participate in camera calibration,so as to ensure the accuracy of internal parameter calculation.Finally,the intrinsic matrix of the image is calculated by combining the acquired camera internal parameters,and the matrix is decomposed by SVD method to obtain camera external parameters with higher accuracy.(2)An adaptive maximum effective feature point matching scheme is designed.In the process of feature matching,it is found that the camera shooting angle and image quality will affect the effective matching number of the final feature points.Therefore,firstly,feature matching observation is carried out on the shooting results of six degrees of freedom shot by a monocular camera,thus finding out an optimal double-view shooting method.Then,denoising and detail information enhancement are carried out on the double views,and the FAST operator is used to extract feature points twice by adapting its internal threshold value,so as to ensure that a set of larger number of feature point sets(for three-dimensional information recovery)and a set of moderate number of feature point sets(for external parameter calculation)are simultaneously obtained;Finally,the feature points are matched under the constraint of the number of the two groups of feature points,and the matching set is optimized by using the random sampling consistency method to obtain the two groups of the most suitable effective feature matching sets.(3)A method for obtaining fine point clouds of three-dimensional objects is given.Firstly,the depth information of the object in the graph is estimated by using triangulation method combined with monocular camera parameters and feature point matching set of double views to obtain the initial point set.Then,the spherical neighborhood control method is used to denoise the initial point set.Finally,the denoising point set is projected in two dimensions,the point sets in different regions are divided by the connected region labeling method,and the regions with large dispersion and few points are eliminated.Then the two-dimensional point sets are restored in three dimensions to obtain refined point cloud data belonging to each object only.(4)A 3D object reconstruction software based on monocular camera is developed.Software component technology is used to construct the software,and functions including image acquisition,image preprocessing,monocular camera calibration,image feature point detection and matching,three-dimensional information recovery,point cloud data denoising,point cloud independent segmentation,independent object fitting and other functions.The effectiveness of the method is verified by software.