Browse Prior Art Database

Object Localization System Using Divided Model Edges

IP.com Disclosure Number: IPCOM000101818D
Original Publication Date: 1990-Sep-01
Included in the Prior Art Database: 2005-Mar-16
Document File: 2 page(s) / 79K

Publishing Venue

IBM

Related People

Iwai, S: AUTHOR

Abstract

Disclosed is a system for localizing randomly oriented 3-dimensional objects. Vision is indispensable for intelligent robots or flexible manufacturing systems. An important role of vision is to locate a specified object that may be randomly oriented or occluded by other unknown objects. 3-dimensional edges are derived from the scene, and are matched with the 3-D edges of the object model. To find the best match is a combinatorial problem, and it requires a lot of computation time. In this invention, model edges are divided into some number of groups in order to improve reliability and performance. Key points of the disclosure are: 1) Model edges are divided based on view points. 2) Division of model edges makes it possible to implement a matching process in parallel processing hardware.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Object Localization System Using Divided Model Edges

       Disclosed is a system for localizing randomly oriented
3-dimensional objects. Vision is indispensable for intelligent robots
or flexible manufacturing systems. An important role of vision is to
locate a specified object that may be randomly oriented or occluded
by other unknown objects. 3-dimensional edges are derived from the
scene, and are matched with the 3-D edges of the object model. To
find the best match is a combinatorial problem, and it requires a lot
of computation time. In this invention, model edges are divided into
some number of groups in order to improve reliability and
performance. Key points of the disclosure are:
1) Model edges are divided based on view points.
2) Division of model edges makes it possible to implement a matching
process in parallel processing hardware.

      A 3-dimensional model for an object is a collection of
3-dimensional edges. A 3-dimensional edge is represented as the
position of its center point, its orientation, and its length. The
edges are divided up into six groups. This grouping is done by
selecting visible edges from six different directions: front, back,
top, bottom, right, and left. One edge may exist in more than one
group. It does not have topological data.

      The algorithm of localization is based on a "pose-clustering
method", where the sensed edges are matched with model edges a
many-to-many way, and a transformation histogram is calculated by
calculating the transformation from two matching pairs.  The
following is the algorithm of the localization system:
for k=1 to 6 do in parallel (1)
  for each pair of pairs ( Si1, Mkj1 ) and ( Si2, Mkj2 ) (2)
begin (3)
       check calculability of transformation parameters
       if calculable then
         begin
           attempt to compute transformation parameters (4)
if transformation exist then place it (5)
                      in parameter space
         end
   ...