Surface Reconstruction From Point Cloud Based On Normal Constraints And Screened Poisson Equation

With the continuous improvement of living standards,the demand for refined three-dimensional models is increasing.At the same time,the increasing variety of three-dimensional point cloud acquisition methods has also enabled us to significantly improve the digital expression of the surrounding world.Therefore,in the background of industry development,it is important to use point clouds to efficiently and easily generate high-precision 3D mesh.As a typical method of 3D mesh,surface reconstruction technology has gradually been widely used in many industries due to its convenient and efficient characteristics.This paper mainly focuses on the surface reconstruction based on the screened Poisson equation,and proposes to use normal constraints to perform accurate and efficient surface reconstruction.This paper studies the content of normal estimation and surface reconstruction.(1)for point cloud with noise,non-uniform sampling and sharp features,a robust normal estimation algorithm combining distance,residuals and normal difference is proposed.The normal difference is used to adaptively select the neighborhood size of the sample points.Further,the distribution interval under the Gaussian distribution of all sample points is obtained according to the estimated difference of the initial and final normal.And it determines the accuracy of the estimated normal.(2)to solve the problem that the screened Poisson surface reconstruction is prone to generate wrong surfaces in areas with inaccurate normals,it is proposed to use the accuracy of estimated normal for targeted constraints.On the one hand,the weights are allocated according to the constraints,so that they have more accurate position constraints at each resolution;on the other hand,it is also used to limit the expansion of the octree node with multi-grid algorithm.Moreover,in the isosurface extraction,asymptotic processing and consistency octree optimization are carried out for the connection ambiguity and inefficient retrieval of moving cubes,and finally the surface reconstruction is realized.We compared the normal estimation algorithm and the surface reconstruction algorithm on the 3D scan dataset,multi-view dense reconstruction dataset and standard simulation dataset.For normal estimation,the quantitative analysis and qualitative comparison at different levels of noise,non-uniform sampling and sharp features show that our robust normal estimation algorithm can accurately estimate the normal of point cloud with different defects.For surface reconstruction,the multi-dimensional comparative analysis also shows that our surface reconstruction algorithm can effectively solve the problem of pseudo-surfaces generation caused by inaccurate normals,thereby reconstructing a more accurate surface.In addition,we also makes a comparative analysis of the efficiency,and the experiment shows the high efficiency of our algorithm.Finally,by virtue of the parameter selection analysis of the confidence threshold,it also reflects the generalization of our algorithm.