| | |
Summary: A Probabilistic Framework for Correspondence and Egomotion
Justin Domke, Yiannis Aloimonos
Computer Vision Laboratory, Dept. of Computer Science
University of Maryland, College Park, MD 20742 USA
{domke,yiannis}@cs.umd.edu
Abstract
This paper is an argument for two assertions: First,
that by representing correspondence probabilistically, dras-
tically more correspondence information can be extracted
from images. Second, that by increasing the amount of cor-
respondence information used, more accurate egomotion
estimation is possible. We present a novel approach illus-
trating these principles.
We first present a framework for using Gabor filters to
generate such correspondence probability distributions. Es-
sentially, different filters 'vote' on the correct correspon-
dence in a way giving their relative likelihoods. Next, we
use the epipolar constraint to generate a probability distri-
bution over the possible motions. As the amount of cor-
respondence information is increased, the set of motions
|
