An Image Sequence Based 3D Reconstruction Algorithm

3D reconstruction is an important research area in computer vision. There are many applications of 3D reconstruction, such as object recognition, robot navigation, virtual reality, medical image reconstruction, terrain measure, industry inspection. The methods of image based modeling can be divided into two groups: active algorithms and passive algorithms. Both of them have their own advantages and disadvantages respectively, and they can be applied to different cases.This thesis mainly discusses stereo vision based algorithm, which is a kind of passive methods. Generally speaking, a stereo vision system consists of camera calibration, feature detecting, feature matching and 3D model calculation. The key issues in 3D reconstruction are how to find the good features and match them, i.e. to set up the relationship between the different images.This thesis presents an image sequence based algorithm. Fist, Harris detector is used to detect feature points. After that, the idea of multiresolution is introduced to match these feature points and build 3D mesh model. The speed of the algorithm presented in this thesis is very fast, but the precision of result is not satisfactory. The algorithm will be improved in future.There are five chapters in this paper. In chapter one, the review of background, previous works and some applications of 3D reconstruction are presented. Chapter two introduces the basic idea of binocular vision. In chapter three, feature detecting and matching are chiefly discussed, which are the key issues in stereo vision. In this chapter, three classical point detectors, Moravec detector, SUSAN detector and Harris detector, are introduced first. Then other part of this chapter gives a similarity and correlation based feature point matching algorithm. In the last chapter, a method combining mesh interpolation and multiresolution has been implemented to obtain 3D mesh model from coarse to fine. Some results are presented in the last chapter.