Geometrical Fitting of Missing Data for Shape from Motion Under Noise Distribution.

When converting image sequence to 3D, several entries of the matrix have not been observed by occlusions and other entries have been perturbed by the influence of noise. In this paper, we propose a method for fitting geometrically missing data in noisy observation matrix with iterative SVD factorization. The main idea of the proposed algorithm is that the orientation and distance of noisy vector ca be handled directly by geometrical properties between 2D image plane and 3D error space under noise distribution. To confirm the recoverability of missing data, we carry out the experiments for synthetic and real sequences. The results in practical situations demonstrated with synthetic and real video sequences verify the efficiency and flexibility of the proposed method. [ABSTRACT FROM AUTHOR]