An algebraic/analytic method for reconstruction from image correspondences
Gunnar Sparr · Series in machine perception and artificial intelligence · 1992
A method for the reconstruction and motion problems is presented, under the assumption that a number of point correspondences in a pair of images are known. A geometric and algebraic theory is based on invariance properties of point configuration under affine and projective transformations. In particular, a characterization of those image pairs which correspond to the same point configuration in an unknown scene is given. An algorithm along these lines is presented, yielding complete sets of solutions to the reconstruction problem. A main idea is to exploit the affine structure of the problem, and work with relative locations, before turning to the metrical structure and absolute locations. It turns out that relative depth information can be achieved at a low computational cost