Research On Key Technologies Of Image-based Automatic 3D Building Reconstruction

The 3D model of buildings is the basic data of real estate and smart cities in China.This thesis focuses on the study of dense reconstruction and texture mapping which are two important steps of image-based 3D building reconstruction.The main research contents of this thesis are as follows:(1)A hypothesis plane fitting algorithm based on 3D line model of buildings is proposed,which contains two parts: a 3D line segment clustering algorithm based on Manhattan-world assumption for clustering 3D line segments,and a multi-hypothesis plane fitting algorithm based on mean shift for abstracting hypothesis planes.The result of 3D line segment clustering is used to determine the direction of plane fitting.Experiments show that the proposed algorithm can obtain the building hypothesis plane model accurately and efficiently.(2)An improved PMVS algorithm based on spatial triangle mesh subdivision is proposed to improve the efficiency of PMVS algorithm.Through spatial subdivision and expansion of triangle meshes which are generated from the building hypothesis plane model,an accurate quasi-dense point cloud can be quickly obtained,which can be used as seed points of PMVS instead of the sparse and noisy seed points generated by PMVS itself.Experiments show that the proposed algorithm can effectively improve the efficiency of PMVS.(3)A multi-view seamless texture mapping algorithm based on MRF is studied.Markov random field is used to solve the optimization problem of texture selection under multi-view situation.Multi-band blending algorithm is used to eliminate the stitching seams in the texture.Experiments show that the final textured 3D building model obtained through the proposed algorithm is seamless and visually realistic.