Research On Compression And Fast Reconstruction Of 3D Cultural Relic Point Cloud Model Based On Compressive Sensing

The digital protection of cultural relics is the extension of traditional cultural relics in the dimension of time and space,which is of great significance in the heritage and protection of cultural heritage,cultural relics display and research and service system.The 3D point cloud model is widely used in the field of computer-aided heritage preservation with its quick access and precise texture details.Because the point cloud data of 3D artifacts obtained by laser scanner is huge and dense,it will lead to a large amount of resources being consumed in data storage,transmission and processing,so it is necessary to compress the point cloud data of 3D artifacts.Aiming at the problems of lossy compression based on geometric topology,long time-consuming reconstruction and low reconstruction precision,this paper transforms the 3D point cloud compression reconstruction problem into digital signal processing field,uses compression perception theory to compress and reconstruct it,and proposes two fast reconstruction methods.Specific research contents and innovations are described as follows:(1)Because the the geometric information of 3D cultural relic point cloud model can be regarded as discrete geometric signal,but the cultural relic point cloud is scattered and dense,and the original point cloud signal is non sparse signal,so it can not be compressed directly by using the compression perception theory,so it is necessary to project the geometric signal of cultural relic point cloud sparsely to make it sparse signal in the change domain.In this paper,three-dimensional discrete Laplace matrix is used as sparse basis.Firstly,we use the octree method based on hash function to construct neighborhood constraint relation for 3D cultural relic point cloud,and then construct 3D discrete Laplace base through point cloud adjacency matrix.compared with discrete fourier bases,discrete wavelet bases,discrete cosine bases,and spectral transformation bases,the discrete laplacian sparse bases constructed by this method have smaller coherence coefficients,so they are more suitable for sparse representation of 3d cultural relic point clouds.(2)Due to that the essence of signal reconstruction is the inverse problem solving process,aiming at the problem that the cloud sensing matrix of 3D cultural relic point is too large,which leads to the longer reconstruction time.This paper uses the method of truncation singular value decomposition to reduce the dimension of sensing matrix and sampling signal,and compares the commonly used truncation value determination algorithm experimentally.The experimental results show that this method has achieved good results in the standard data set Stanford model,the experimental data Terracotta Warriors model and the three color figurines of Hu people in the Tang Dynasty model,which can quickly reconstruct the original 3D cultural relic point cloud model under the condition of ensuring the reconstruction quality.(3)In order to improve the reconstruction effect of the model and reduce the reconstruction error,a method based on sparse automatic encoder and compressed sensing is proposed for the rapid reconstruction of cultural relics dense point cloud model.In view of the bottleneck of slow recovery caused by the large scale of sensing matrix in the inverse problem based on compressed sensing theory,a method of reducing the scale of inverse problem and accelerating the reconstruction speed by sparse automatic encoder is proposed.In order to test the performance of the method,the Stanford rabbit point cloud model and the terra cotta warriors head point cloud model are used for simulation experiments.The experimental results show that this method can obviously speed up the reconstruction of 3D cultural relic scattered dense point cloud model with high accuracy.(4)Finally,this thesis compares the reconstruction time and accuracy of four classical algorithms based on greedy method on the point cloud model of the terracotta warriors' head and the point cloud model of the Hu terracotta warriors in the Tang Dynasty.The experimental results show that the greedy algorithm is effective for the reconstruction of 3D cultural relic point cloud model,and the reconstruction results of these four greedy algorithms are comprehensively evaluated.