Abstract:
A representation of polyhedral environment in a stereo pair of images is proposed
and its three-dimensional recovery is presented. The polyhedral environment is
represented as a number of image-to-image mappings in the form of matrices,
each corresponding to a planar surface in the environment. Unlike a mere depth
map, such a representation is segmented, in the sense that different surfaces in
the environment correspond to different matrices and are separated. A mechanism
is proposed to recover the representation even for scene that is not densely
featured. Experimental results on typical polyhedral scenes are given.