Nonlinear Modeling and 3D Reconstruction from Un-calibrated Multiple-view
Yang Zhong-gen, Yuhong Liu · 2006
The constraints on the target shape model were firstly explicitly built up. Then, based on the rank-3 constraint on the centralized multi-frame digital shape matrix, a nonlinear optimization criterion function about the set of the multi-frame depth vectors was suggested. In the iterative update of the nonlinear optimization process, the left singular transformation matrix of the SVD of the multi-frame centralized digital shape matrix was firstly used to self-calibrate the intrinsic parameter matrix of the camera, then a generalized eigen value analysis process was used to optimally update the set of the multi-frame depth vectors. As soon as the nonlinear optimization iteration was completed, multi-frame 3D reconstruction, shape modeling and multi-frame motion recovery can be carried out one by one. The theoretic analysis and experimental demonstration have shown that the developed nonlinear algorithm is fast, accurate, efficient and practical.