Research On The Key Technologies Of Three-dimentional Reconstruction From Multiple Images

The key technologies of three-dimentional reconstruction from multiple imagesinclude camera calibration, stereo matching and visualization of3D model. One of thecore problem is to acquire the corresponding image elements automatically anddensely, which means the obtainment of identical points in wide baseline imagematching and stereo matching. Another key is to build the right geometric model andsolve it robustly. Based on those analyses, the main works in this paper are listed inthe following part.①Reasearch on the basic multi-baseline geometry. Before introducing the epipolargeometry, some basic concepts and transformations of projective geometry arepresented. They are the cornerstone of the following multi-baseline geometry.②As a key step of3D reconstruction from images, image matching holds greatimportance. Therefore, SIFT, milestones in the field image matching areintroduced clearly and the robust matching based on fundamental matrix is alsopresented here.③After an introduction of camera models’ different appearances in Photogrammetry,Computer Vision and Computer Graphics, a series of transform equations of themodels’ parameters between those different appearances are given. On the basis ofthese, a camera simulation software is implemented and applied in the display ofphotos in virtual scenes. After this, the camera self-calibration algorithms arestudied. Considering the purpose of this research, this paper claims that calibrationfrom vanishing points is the most appropriate calibration approach.④Stereo matching solves the stereo corresponding problem, it distincts the differentmodeling algorithms for calibrated images. After a short review on classicalproblem, space-patch model reconstruction algorithm is studied in detail; itgradually reconstructs the model by expanding and filtering step by step. It is proved by the experiments based on real datasets that these algorithms arereasonable and can achieve a nice reconstruction. A modeling system based on MFCframework and other open source packages such as OpenGL is presented here, it hastwo main functions in the moment, namely planar camera calibration andmulti-basement images reconstruction based on SPM.