Research On The Theory And Algorithms Of 3D Reconstruction From Multiple Images

Reconstructing 3D models from 2D images has been a hot research topic in the field of computer vision. It is a comprehensive and cross research area and has a very wide application prospects in the fields of aerial mapping, visual navigation, medical diagnosis, electronic commerce and virtual reality, etc. This paper provides a deep research on the theory and algorithms of 3D reconstruction from multiple images and a corresponding reconstruction software system is developed based on these knowledge.Some basic concepts in 3D reconstruction, such as camera model, coordinate system transformation, epipolar geometry, camera calibration, are first introduced. Then a feasible solution of the 3D reconstruction from multiple images is proposed and a detailed analysis of each step is presented. The Structure from Motion(SFM) method based on SIFT feature detection and bundle adjustment is introduced in the sparse point cloud reconstruction step. The Patched-based Multi-View Stereo(PMVS) algorithm is introduced in the following dense point cloud reconstruction step. For the possible point clouds registration issue, the classic Iterative Closest Point(ICP) registration algorithm is introduced and a novel two-stage registration algorithm based on the local invariant feature is presented. An incremental surface reconstruction algorithm based on the breadth-first-search is proposed for the point cloud meshing step. An automatic texture mapping method is presented for the last texture mapping step. Finally, the modular architecture and implementation of our software system is introduced and experimental results demonstrates that satisfactory 3D models can be reconstructed, which verifying the feasibility and efficiency of the algorithms mentioned in this paper.