Studies On 3D Reconstruction From Image Sequences

With the development of Computer Graphics,Virtual Reality and Multimedia Communications in the past few years,highly realistic 3D models are increasingly needed to visualize and simulate the real world. However,Despite the constant progress in the traditional surface modeling techniques,creation of complex 3D models remains a painful and time-consuming task,moreover,the rendered images are mostly quite different from the real objects.It's simple in modeling and convenient in operating in contrast to traditinal metheds,so it will be used in wides areas.This dissertation mainly focuses on the following problems :1)A survey on the recent 3D reconstruction from image sequences techniques is given from the point of view of Computer Graphics.Some typical 3D reconstruction methods are summarized,the fundamental principles behind each methods and the state-of-the-art progress are also presented;2) Given a stratified reconstruction method based on rational B-spline method.The method takes advantage of the projective invariance of rational B-splines,given feature correspondences in multiple views,we first have a 2D B-spline fitted in each view.then 2D control points in each view are recovered independently,after this,the 3D B-splines and projection matricies can then be extracted from 2D B-splines using factorization methods,finally we given a method to upgrade this projection reconstruction to Euclidean reconstruction. Compared with previous stratified reconstruction works , The method given by our dissertation fully takes advantage of B-spline's character, simplifies matching in the algorithm , so very high reconstruction precision could be achieved at relatively low memory cost .