Abstract
First, in this thesis, we will introduce the stratified 3D reconstruction method. Neither the internal parameters of the camera nor the motions of the camera are known. The only assumptions are the object of interest is rigidity and the internal parameters are invariant. The sequences are generated by circumnavigating the object of interest with a camera. First, the suitable feature points are located and matched using an image matching algorithm. Then the epipolar geometry is computed and the initial projective reconstruction is performed. After that, self-calibration of camera is achieved by employing the concept of stratification and thus the reconstruction structure is upgrade from projective stratum to affine and metric stratum. However, the implement of the stratified 3D reconstruction proposed by Marc Pollefeys [23] still have some steps which difficult to solve. In viewing of this, we make some assumptions and describe a method to overcome these difficulties.