Bolles, R.C.[Robert C.],
Horaud, P.[Patrice], and
Hannah, M.J.[Marsha Jo],
3DPO: A Three-Dimensional Part Orientation System,
IJRR(5), No. 3, Fall 1986, pp. 3-26.
BibRef
8600
And:
IJCAI83(1116-1120) reprinted in
BibRef
RCV87(355-359).
BibRef
And:
3DMV87(399-450).
Light stripe 3D data is used for input. Locate primitive features,
cluster these, generate and verify the
hypothesis of match, generate transformations.
BibRef
Horaud, P.[Patrice], and
Bolles, R.C.[Robert C.],
3DPO's Strategy for Matching Three-Dimensional Objects in Range Data,
Conf. on RoboticsAtlanta, March 1984, pp. 78-85.
BibRef
8403
Kahl, D.J.,
Rosenfeld, A., and
Danker, A.J.,
Some Experiments in Point Pattern Matching,
SMC(10), No. 2, February 1980, pp. 105-116.
BibRef
8002
UMD-TR-690, September 1978.
Point features are used to find a global transform (translation only)
between two images of the same scene. Different numbers of feature
points may be found in the two images, but the distortions and
rotations are small. For each pair of points in both images, a
translation is computed to map the first point in one pair to the
first point in the other pair. If the translation also approximately
maps the second points in the pairs then the rating of this possible
translation is incremented. The best global translation is indicated
by a high rating (or a cluster of high ratings) in the accumulation
space. This technique is sensitive to displacement noise, but
tolerates deletions or additions of points. Since the global
accumulation covers only translation, changes in orientation
(rotation) also cause problems. Some error tolerance is possible by
introducing labels (or property values) for each feature point.
BibRef
Moss, S.[Simon],
Hancock, E.R.[Edwin R.],
Pose Clustering with Density Estimation and Structural Constraints,
CVPR99(II: 85-91).
IEEE Abstract. IEEE Top Reference.
WWW Version.
BibRef
9900
Earlier:
Structural Constraints for Pose Clustering,
CAIP99(632-640).
WWW Version.
9909
BibRef
Ranade, S., and
Rosenfeld, A.,
Point Pattern Matching by Relaxation,
PR(12), No. 4, 1980, pp. 269-275.
WWW Version.
Relaxation.
The input is two sets of points, each corresponding to feature
locations in different views of the scene.
Like Kahl(
See also Some Experiments in Point Pattern Matching. ), the system
finds a global displacement (translation) that best fits the data, but
works better with small rotation and scale changes. Points are
matched with all points in the other image with the match rating based
on how many other point matches agree with the transform computed from
that match. The scores are computed (and updated) based on the scores
of the other point pairs to produce a highly rated consensus transform
for the set of points.
BibRef
8000
Wang, C.Y.[Cheng-Ye],
Sun, H.[Hanfang],
Yada, S.[Shiro], and
Rosenfeld, A.,
Some Experiments in Relaxation Image Matching Using Corner
Features,
PR(16), No. 2, 1983, pp. 167-182.
WWW Version.
BibRef
8300
Earlier:
UMD-CS TR-1-71.
Relaxation. A relaxation procedure is used to find matches between pairs of
images that differ in position and orientation. The matching is
performed on sets of feature points (corners), which have position,
orientation, contrast, and sharpness. After several iterations,
good matches are clustered which gives sets of transformations
(translation and rotation). The best transformation can be
selected from these likely ones. This extends an earlier method
See also Point Pattern Matching by Relaxation.
BibRef
Ogawa, H.,
Labeled Point Pattern Matching by Fuzzy Relaxation,
PR(17), No. 5, 1984, pp. 569-573.
WWW Version.
BibRef
8400
Ogawa, H.,
Labeled Point Pattern Matching by Delaunay Triangulation and
Maximal Cliques,
PR(19), No. 1, 1986, pp. 35-40.
WWW Version.
BibRef
8600
Kitchen, L.,
Relaxation for Point-Pattern Matching:
What it really Computes,
CVPR85(405-407).
Univ. of Massachusetts. Preliminary.
BibRef
8500
Sang, N.,
Zhang, T.X.,
Rotation and Scale Change Invariant Point Pattern Relaxation
Matching by the Hopfield Neural Network,
OptEng(36), No. 12, December 1997, pp. 3378-3385.
9801
BibRef
Chapter on Registration, Matching and Recognition Using Points, Lines, Regions, Areas, Surfaces continues in
Long Sequences, Motion .