SVD-based camera self-calibration and 3-D reconstruction from single-view

摘要

Due to the difference between single-view and multi-view, the homography matrix, epipolar constraint and fundamental matrix in the case of model-based single-view are first defined. Then, inspired by the SVD method of two-view problem, and observing the similarity of our problem to two-view problem, we prove that, in order to uniquely determine homography matrix, a 4-dimensional mid-parameter vector can be optimally estimated from the data transformed by the corresponding left singular matrix of SVD analysis of the fundamental matrix. At last, the intrinsic parameter matrix, the 3-D motion as well as the 3-D reconstruction can be straightforward calculated. So, a new algorithm to self-calibrate the intrinsic parameter matrix of a camera and to reconstruct the 3-D shape of the target in the single-view is successfully developed. The experiment has demonstrated that its performance is fairly satisfactory.