Browse Prior Art Database

Segmentation of a Three-Dimensional Scene Into Free Areas And Object Surfaces Using Occlusions of Trinocular Stereo

IP.com Disclosure Number: IPCOM000120571D
Original Publication Date: 1991-May-01
Included in the Prior Art Database: 2005-Apr-02
Document File: 3 page(s) / 94K

Publishing Venue

IBM

Related People

Echigo, T: AUTHOR

Abstract

A program is disclosed for segmenting a three-dimensional (3D) scene into free areas, in which there are no objects, and visible object surfaces, using (1) sparse data obtained from trinocular stereo and (2) occluded edges that cannot be seen from at least one of three cameras.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 58% of the total text.

Segmentation of a Three-Dimensional Scene Into Free Areas And Object
Surfaces Using Occlusions of Trinocular Stereo

      A program is disclosed for segmenting a three-dimensional
(3D) scene into free areas, in which there are no objects, and
visible object surfaces, using (1) sparse data obtained from
trinocular stereo and (2) occluded edges that cannot be seen from at
least one of three cameras.

      The proposed method employs trinocular stereo in order to
obtain 3D edge segments, not only because trinocular stereo is more
reliable for finding correspondences than binocular stereo, but also
because it clearly classifies edge segments into 3D edge segments
that can be seen from all three cameras and ones that cannot be seen
from at least one of three cameras.

      The proposed approach consists of three steps. In the first
step, free areas are obtained only from 3D edge segments which can be
seen from all three cameras, for which correspondences are found
under the constraints of trinocular stereo. Fig. 1 shows that the 3D
edges A and B can be obtained from edge images {a, b}, {a", b"}, and
{a', b'} on the image planes of three cameras. When the edge images
{a, b}, {a", b"}, and {a', b'} are in the same order between all
images and there is no edge between any two edge images, the hatched
areas in Fig. 1, which are enclosed by the lines of sight from all
cameras, are free areas.

      The second step is to expand the free areas by using occluded
edges...