A Study Of The 3D Reconstruction Based On Multiple View Geometry In Computer Vision

The 3D reconstruction is a long term research, and the early approaches have to achieve the camera calibration firstly and then can complete the residual work. Therefore, it can deal with the reconstruction in still or the known conditions. But Faugeras and Hartley put forward a new theory (i.e. also called multiple view geometry in computer vision) in 1992 which can use the un-calibrated image sequence, improving observably in 3D modeling. In this theory framework, it can deal with the 3D projective reconstruction of unknown image sequence. Comparing to the traditional Photogrammetry, 3D reconstruction of the multiple view geometry has the advantages of low-cost, high-reality and high-automatization, therefore it has broad prospective. Currently image based 3D reconstruction has become a hot topic of research.This paper goes deep into the research of the multiple view geometry, but mainly on two view 3D reconstruction. The main work of this paper consists of following three parts:1, it introduces the basic theory and method of the multiple view geometry, and it uses the method of the Harris corner detector and gray correlation coefficient to achieve the auto-obtaining and auto-matching the corners.2, it designs a highly automatization 3D reconstruction system which using two correspondent view. We use the triangle method to compute the characteristic points, and then updating the projective construction to measurement construction which using the theory of stratum.3, it achieves the RANSAC robust algorithms estimating the fundamental matrix, and taking effect of the non-linear aberrance into account.