| In recent years the image based modeling technology has drawn the interests of many researchers and is becoming one of the hotspot for the area of computer graphics and virtual reality. Different from the traditional geometric modeling technology, the image based modeling technology get raw data from the original scene with cameras or other kind of optical instruments. Then, with some pre-known knowledge of scene and cameras, the IBM method retrieves the 3D scene from the information encapsulated in the image sequences. Compared with traditional method, the IBM method not only save a lot of human effort, but also enhance the photo-reality of the scene greatly. Up to now, many methods has been developed to study the field, our focus goes as following:1 In this thesis, a passive reconstruction method named of structure from motion is investigated. We realize the feature extraction, feature matching, calculation of the fundamental matrix and projective stratum reconstruction. In the process of the realization, we make some improvement to the RANSAC algorithm that is used to calculate the fundamental matrix. A new feature selection method is suggested based on watershed transform. This algorithm greatly solves the degenerate case in the previous method and results in more robust fundamental matrix.2 In this thesis, an active reconstruction method named of weakly structured lighting is reviewed. We implemented the camera calibration, source light point calibration, shadow plane estimation, triangulation. Near the end of the chapter, the advantages and disadvantages of the algorithm are evaluated.3 We further study the image rectification method and stereo matching method that are generally used in many reconstruction methods. We realize one kind of rectification method and stereo matching method respectively. During the realization of the stereo matching method, we proposed a new optical flow calculation method. It enhances the matching accuracy greatly with only a little more overheads.
|
PDF Full Text Request