3D Reconstruction Method Based On Weak Texture Multi-View And Its Application In The Field Of Cultural Relics

Multi-view 3D reconstruction is a research hotspot in the field of computer vision.It has broad application scenarios in the areas of cultural relics protection,virtual reality,military operations,medical diagnosis and treatment.Common physical objects usually contain weakly textured areas.These weak areas have the characteristics of single color,repeated textures,few dotted features,and similar gray values.When the proportion of weakly textured regions of the reconstruction object is high,many problems existing in the traditional methods are exposed.For instance,low feature registration accuracy,inefficient incremental solving process,redundant calculation of bundle adjustment,and errors that remain after model correction are found in the traditional incremental structure from motion algorithm.Therefore the existing 3D reconstruction methods are difficult to reconstruction 3D modeleffectively.This thesis focuses on 3D reconstruction algorithms Based on Weak Texture Multi-view.The main work is as follows:(1)Aiming at the problems of difficult extraction of feature points and low accuracy of feature registration in weakly textured images,thefeature point extraction and matching algorithm based on Distance and Texture combining Normalized Cross Correlation(DTNCC)is proposed.First,the Euclidean distance and texture features are combined to obtain the normalized cross-correlation number of DTNCC.Secondly,using the self-similarity of weakly textured regions,The optimal radius feature point extraction algorithm based on DTNCC is designed to extract denser feature points of weakly textured targets.Finally,DTNCCwas used to classify the diffusion feature queue to improve the number and accuracy of feature matching point pairs.(2)Aiming at the problems of model drift and low optimization error efficiency during Structure From Motion(SFM),this thesis proposes an SFM optimization algorithm based on weak texture multi-view.First,a registration method based on Regional Curvature Binary Descriptor(RCBD)was used to register the local error point cloud.After the reconstruction reaches a certain scale,the global bundle adjustment is used to eliminate the bias in the point cloud generation process,which significantly improves the efficiency and effect of correcting model drift.(3)We achieve 3D reconstruction based on weak texture multi-view,and verify its feasibility in the field of cultural relics protection through experiments.The DTNCC-based feature point extraction and matching method is used to provide high-quality initial parameters for SFM.The SFM optimization algorithm based on weak-texture multi-view can solve the error accumulation of the camera pose,reduce the optimization error time,and retain rich point cloud details.The system provides new research ideas and technical support in the digital protection of cultural relics.