3D Reconstruction Of Tangible Cultural Heritage Based On Multi-views

With photogrammetry and computer vision development,the technology of 3D reconstruction based on multi-view has attraceted many researchers' attention.The technology generates the virtual model from the scene and the target object by the image sequences,which is an important means of realistic 3D modeling and virtual reality.Currently,the research results have been applied in the field of digital city,medical imaging,machine vision,and tilt photogrammetry.However,tangible cultural heritage usually performs complex geometry features with curved and straight contour lines and convex shapes,which leads to diffcultly generating high-fidelity and high detail 3D model based on multi-views 3D reconstruction.Therefore,focused on the complex geometry features of tangible cultural heritage,the paper carried out the method of structure from motion to reconstruct 3D point cloud,proposed improved poisson algorithm to realize surface reconstruction,used MRF to realize texture mapping in best views,and finnaly achieved a realistic 3D model reconstruction.The specific contents and results are as follows:(1)Based on the analysis of the research status of 3D modeling methods and the contrast among multiple methods of 3D multi-view reconstruction,we proposed the research idea and the technical route of 3D multi-view reconstruction and prototype system's design.(2)We carried out reconstruction based on 3D point cloud tangible cultural heritage of multiple images.The paper mainly took out the method of structure from motion to achieve 3D point cloud reconstruction:firstly,we used scale-invariant feature transform algorithm to extract feature points,and applied nearest neighbor matching algorithm combined with RANSAC method to achieve correct matchs of the feature points among images;secondly,we made use of the method of layered gradually camera self-calibration to realize the reconstruction from the projective space to the metric space,and obtain internal and external camera parameters;then,we employed patch-based multi-view stereo algorithm to estimate image depth.Finally,we took the sculpture,the ancient and modern architecture of the actual scene as experimental objects,and verified the feasibility and effectiveness of the algorithm framework of 3D point cloud reconstruction for tangible cultural heritage based on multi-views.(3)We proposed an improved poisson algorithm based on Gaussian filter.Since poisson surface reconstruction algorithm still couldn't effectively meet the balance between details keeping and noise smoothing,this paper presented an improved poisson algorithm based on gaussian filter.Through using gaussian filter in the vector field estimates of point cloud isosurface,it achieved a more accurate estimate of the point cloud topology and noise effectively smoothing.On the other side,by adjusting the standard deviation of the gaussian filter,it realized a fine control of point cloud between details keeping and noise smoothing.The paper took the statue of a form complex sculpture as the experiment object,took point cloud from 3D multi-views reconstruction as input data,and carried out the point cloud gridding with improved Poisson algorithm.The results showed that improved poisson algorithm raised the accuracy and integrity of the mesh model and realized the approximation of the details keeping of the truth ground,which verified the effectiveness of the improved algorithm.(4)We proposed a texture mapping scheme based on Markov Random Field.There exit two problems in the texture mapping:firstly,obstruction and self-obstruction lead to invisible patches;secondly,the change of lighting conditions' among images leads to different brightness in the different images.Therefore,we proposed and implemented a texture mapping scheme based on Markov random field.From a global perspective,the scheme could solve the color and light texture discrepancies problems between adjacent patches in the grid model,which relized high realistic texture mapping.(5)We researched and developed a prototype system of multi-view 3D reconstruction for tangible cultural heritage.The system not only reconstructed 3D model of the target scene,but also supported interactive editing and the rapid extraction of model parameters.Finally,the paper took the stone statue and the TuLou building among Fujian Province tangible cultural heritage as the experiment object,and carried on the practical application of the typical objects in the tangible cultural heritage based multi-views 3D reconstruction.