Research And Realization Of The Repair Algorithm Of The 3D Reconstruction Point Cloud Model Based On PMVS

Currently,three-dimensional reconstruction based on images has been a hot topic of research,which have high photorealistic representation,and maintain the angle and proportion relation with the physical sample.However,some defects such as holes will occur in the process of reconstruction,which might be caused by insufficient information and lower resolution of images.These holes not only affect the quality of reconstructed model,but also many subsequent operations,such as texture mapping,triangulating process,etc.So,in order to improve the reconstruction accuracy,it is necessary to detect model boundary and fill holes.For the questions of boundary point recognition and restoration of holes of the model,this paper emphatically studies the algorithms of boundary extraction and hole repairing based on scatted point cloud,and improves and puts forward the algorithm on the above basis.In this paper,simulation experiments achieve the desired results and the experimental models are mainly based on non-closed models.the main work of this paper is as follows:(1)This paper briefly analyzes several three-dimensional reconstruction techniques based on image,and introduces the three-dimensional reconstruction technology based on PMVS in detail,such as its main theoretical basis and algorithmic process.Before modeling,we use foreground and background separation method to preprocess the input images,and remove reconstructed model noise.(2)Propose a new boundary feature recognition algorithm based on second extraction of sampling points.This paper presents a multi-stage random sampling method to obtain the initial edge length of the grid on the above basis,and get the spatial raster index and topology structure of each spatial discrete point.According to the number of points in the neighboring grids of each point to extract boundary feature of model preliminarily,and obtain the initial boundary feature points.We put forward the concept of grid proxy point,and create a micro-cut plane with the initial boundary point and their neighborhood mode points by using least-square fittingmethod,and project them to the plane.We filtrate the initial boundary points according to the angle between the corresponding projection points of the initial boundary point and its neighbor mode points in the plane,then obtain the correct model boundary characteristics.(3)Present a new boundary extraction algorithm based on space partition.In this algorithm,we get the spatial raster index and topology structure of each spatial discrete point on the above basis,and according to rasterization results,divide the bounding box respectively along the X-axis,Y-axis,Z-axis,and in the way we convert the 3D points to the 2D plane.Then achieve the reasoning algorithm of planar point cloud topology structure to identify the boundary grid of slice layer and filter pseudo edge points.Finally,obtain the model boundary features directly.(4)On the basis of boundary extraction,this paper designs a new repair algorithm based on normal vector constraint.This method uses the relation of the spatial coordinates and the normal vector of the boundary points and its neighborhood points to fill the missing part of the model through the curve fitting method along two axes direction.Then the obtained initial fill points set is processed with the filter function to get the filter mean point,which is the final fill point.