Research Of Image Based 3D Reconstruction

With the e-Manufacturing developing, e-business is getting regard, how to reconstruction the 3d models of merchandise is very important. Image based 3D reconstruction is to recover the 3D models of objects and scenes using image sequences. It is a cross field of image processing, computer graphics, computer vision, Pattern Recognition and so on. Currently Image based 3D reconstruction has become a hot topic of research. Comparing to other ways to construct 3D models, such as using special software and range scanners, Image based 3D reconstruction has the advantages of low -cost, high-reality and high-automatization, therefore it has broad prospective. Theoretically, Image based 3D reconstruction is the inversed aspect of rendering of 3D models, so how to recover the 3D information according to the incomplete or noise-corrupted 2D information is the hard problem of this technique as well as computer vision. The future research into Image based 3D reconstruction will push the development of related subjects.This thesis goes deep into the research of Image based 3D reconstruction, mainly on manual method and stereo vision method, the main work of this thesis consists of following three parts:1. Developed a manual 3D reconstruction system based on the theory of single view metrology. Single view metrology compute the 3D coordinate of a point based on the projective geometry's characteristic of cross-ratio invariability. In this system the input only one image, by manual we can reconstruction the 3D model of scene.2. Improved the camera calibration method of Tsai's, take into account the radial aberrance and tangent aberrance of lens. Because the aberrance is little, we get a calibration method in which the interior and exterior parameter are apart from the revise parameter of binocular disparity.3. Developed a matching software based on the literatures. By selecting the feature points, correlation matching, computing the intensity of matching, relaxation method, Lmeds method we can get matching points.