CSE 252 B : Computer Vision

Last lecture we learned how to find interest points using the Förstner corner detector. The goal of this lecture is to find correspondences between these points automatically. This is known as the correspondence problem. The correspondence problem is arguably the hardest problem in structure from motion (SFM). For example, one must deal with pointsets of different length, with spurious coordinates and positional noise. All of SFM relies on the success of finding correspondence between points in the two images. If you have two views that are too widely separated, it can become difficult to match interest points for purposes of estimating a homography or fundamental matrix. Correspondence is an absolutely critical part of the vision problem.