3D Reconstruction Based On Multiple View Geometry

Multiple view3D reconstruction is an important research topic in computer vision. In this study,3D geometric constraint information is extracted from multiple views of the object and used to realize3D reconstruction. The early3D reconstruction methods are limited in practical application. They canonly be achieved under static and known environment as they need the calibrated camera or thespecific movement of camera. In this paper, the3D reconstruction method based on multi-viewgeometry doesn’t need the specific moving trajectory of the camera or the camera’s calibration. It canbe used in the case when the camera parameters are unknown. The main research contents are asfollows;1)Study the theory and related methods of the camera model and multi-view geometry incomputer vision;2)Study the method of corner detection and corner matching. In corner detection,discussed the advantages and disadvantages of the current methods, and proposed an improved Harriscorner detection method, which can effectively avoid the situation of corner clustering and cornerundetected; In corner matching, the initial match based on gray correlation values and the fine matchbased on the fundamental matrix by introducing the robust estimation, can effectively eliminate thefalse matches. Finally, we can obtain more correct match points by the guided matching;3)Study theprojective reconstruction and metric reconstruction of multiple views. In the projective reconstruction,we use the simplified reconstruction theory and the duality six points algorithm to achieve theprojective reconstruction, which has the stability and geometry effectiveness; In the metricreconstruction, we use stratified reconstruction method to realize the affine reconstruction and themetric reconstruction. The multi-view reconstruction method studied in this paper can be used wideras it can achieve self-calibration, without requirement of camera pre-calibration, specific trajectory ofcamera, vanishing points or vanishing lines in the reconstruction;4. In the end, we developed a set ofautomatic3D reconstruction system on the VC++6.0platform.