Multiview Normalized Epipolar Constraint
Yi Mat, Shawn Hsut · 2001
In this paper, we study the structure from motion prob- lem as a constrained nonlinear least squares problem which minimizes the so called reprojection error subject to all con- straints among multiple images. By converting this con- strained optimization problem to an unconstrained one, we obtain a multiview version of the normalized epipolar con- straint of two views. Such a multiview normalized epipolar constraint serves as a statistically optimal objective func- tion for motion (and structure) estimation. Since such a function is dejined naturally on a product of Stiefel man- ifolds, we show how to use geometric optimization tech- niques to minimize it. We present experimental results on real images to evaluate the proposed algorithm.