Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!purdue!tut.cis.ohio-state.edu!zaphod.mps.ohio-state.edu!think!snorkelwacker!ai-lab!tmb From: tmb@wheaties.ai.mit.edu (Thomas M. Breuel) Newsgroups: comp.graphics Subject: Re: how to match 2 views to an object Message-ID: <5404@rice-chex.ai.mit.edu> Date: 8 Dec 89 05:28:36 GMT References: Organization: MIT AI Lab, Cambridge, MA Lines: 21 given two sets of 2D points, with known one-to-one correspondent relationship {(Pi,Qi), i=1,2,...,n}, find out whether they could be two perspective projections of a set of 3D points in the space. Put it in another way, given two pictures each having a line drawing structure isomorphic to the other, find out if they could be taken from a 3D object in the space. This problem has been addressed in work on feature based visual object recognition and structure-from-motion. I remember that Shimon Ullman has proven a number of results about the number of views and number of points in each view that are needed to determine the 3D coordinates of the points given correspondences and the 2D positions of the points in each view. Ullman and Basri have recently shown that the 2D coordinates of points in new views can be written as the linear combination of coordinates of points in three given 2D views. For feature based object recognition, the standard problem is that the 3D position of the points are known but the correspondences are unknown, although some work has also been done on the 2D/2D case without known correspondences. Thomas.