Research On The Application Of Line Feature In Image-Based Three-Dimensional Reconstruction

Three-dimensional(3D)reconstruction is an important research field in photogrammetry and remote sensing.The point-based reconstruction methods are difficult to preserve the geometric and structural information of the scene accurately.In order to generate the 3D model with high structural representability,this dissertation explores and discusses the application method of line features in image-based 3D reconstruction from four aspects: line-based model generation from a single image,line matching in stereo image pair,multi-view matching and 3D line reconstruction,constrainted triangular mesh generation.In general,the main work and innovations of this dissertation are as follows:1.A comprehensive summary of the existing point-based matching and 3D reconstruction methods is given.The contents that need to be optimized and breakthroughs in existing methods are analyzed.Various application methods of the line feature in 3D reconstruction are compared,and the feasibility of the line-based 3D reconstruction and high quality digital surface model generation combining point and line features is discussed.2.In order to generate line-based model from a single image,A 3D line reconstruction method based on structural analysis and compensation is proposed.The method first extracts the line features in the image and detects the dominate points on the line.Then,the spatial coordinates of the dominate points are solved by using the existing high-precision DSM.Finally,the line-based model is generated by cubic spline interpolation.Experiments show that the method can obtain dense line reconstruction results and effectively strengthen the structure description of the terrain and ground feature in the scene.3.Many unfavorable factors may hinder the line matching,such as fracture,deformation or occlusion occurred in line extraction.In order to overcome these problems,a novel method for line matching based on discrete description and corresponding points constraint is proposed.The initial corresponding lines are obtained by the distribution statistics of the corresponding points on the line.And the false matching in matched line pairs is eliminated by the similarity verification of line-point affine invariant.Better performance of the proposed method as well as its superiority in curve matching.4.To handle the problem of line matching in multiple images,this chapter extend the line matching method in image pair to multi-view level.Firstly,the adaptive geometric constraint multi-view matching strategy is used to choose candidate lines.Then,the corresponding lines in multiple images are obtained by the distribution of the corresponding points on the line.Finally,the line-based model is generated by discrete point precise matching together with B-spline in-terpolation.The experiment shows that the aforementioned line matching algorithm can obtain semi-dense matching results from multiple images,and the reconstruction algorithm is better to restore the spatial shape of the line feature with highly reliability.5.Combining both point and line features,a high quality triangular mesh generation method is designed.The method utilizes the point and line matching results to obtain point cloud,and then,to construct the initial mesh.The boundary of the influence domain in the mesh is determined by the position of the constraint line,and the mesh is modified by the attribute relation between discrete points on the line,so as to endow the mesh with the best structural precision and visualization performance.Extensive experimental results show the validity of the method.Line features can be effectively used to compensate the structural information lost in point-based model,and thus,the accuracy and visualization ability of the 3D reconstruction model is significantly improved.