Research On PMVS Monocular 3D Reconstruction Method

In the process of monocular reconstruction,a motion recovery structure system is often used to obtain the relative position between each camera and the spatial point cloud corresponding to some image feature points.Based on the output model of the motion recovery structure system,the multi-view stereo matching algorithm(PMVS)can further estimate more matching point pairs on different images to recover denser surface point clouds of objects.However,the PMVS algorithm has the problems of severe loss of details of reconstructed objects and inaccurate reconstruction of the position of the point cloud.The problem is particularly serious when the sequence of captured pictures is small and the texture of the reconstructed objects is not obvious.In this paper,the dense reconstruction work in the monocular 3D reconstruction system is studied,and the corresponding improvement methods are proposed for the shortcomings of the application of the PMVS algorithm in the reconstruction process.The main work of this article is as follows:1.Sparse 3D reconstruction using incremental motion recovery structure system.The SIFT feature points in the multi-view image sequence are extracted,and the false matching is removed by the reverse screening method and the RANSAC algorithm.Among them,use the camera CCD size to estimate the camera’s inherent internal parameter coefficient,and use the matching point and internal parameter coefficient to complete the external parameter coefficient estimation.Finally,nonlinear optimization is used to adjust the position of the camera’s internal and external parameter matrices and spatial points to reduce the reprojection error of the spatial points.2.In view of the shortcomings of the application of PMVS algorithm in the reconstruction process,a corresponding improvement method is proposed.First,the PMVS algorithm based on patch diffusion is used to obtain a good quasi-dense point cloud.The projected matching point of the point cloud is obtained through the projection matrix,and then the method of matching points based on the distance constraint of neighboring points,ZNCC stereo matching constraint and epipolar constraint is used Regional diffusion;then use the template matching algorithm to obtain the matching block of the point cloud hole on the image,use the adaptive window size ZNCC stereo matching algorithm to obtain the matching point in the matching block,and finally obtain the matching through subpixel interpolation and triangulation The spatial point corresponding to the point finally reconstructs the dense point cloud on the surface of the object.3.Complete the verification of the improved algorithm through experiments.Shoot objects with partial weak texture areas,and use the obtained image sequence in a system composed of motion recovery structure and improved algorithm in this paper to realize three-dimensional reconstruction of the object surface.Through the understanding and design of the incremental motion recovery structure system,it is applied to the data set for sparse three-dimensional reconstruction,including image feature point detection and matching,image registration,triangulation of matching points,and finally through the beam Method adjustment optimization parameters.The spatial position of the camera and the sparse point cloud of the object surface are obtained,and the sparse three-dimensional reconstruction is completed.Then,the PMVS algorithm based on the point cloud patch diffusion and the improved algorithm in this paper are used to reconstruct the dense point cloud.The experimental results show that the improved method in this paper not only improves the density of the point cloud,but also fills in some hollow areas on the surface of the object,improves the quality of the point cloud,and completes the reconstruction of the dense point cloud of the object.