ranges of dimension cannot cover the range of the object
dimensions. In the database, only models 4 and 6 cover the
dimension of the object, while the sizes of the other models are
either too small or too large and therefore they are ignored.
1 2
Figure 8 : CAD models in the database
The graph searching process is then used to find the
correspondence between the nodes in the object and the nodes
in the model. Graph matching starts with one planar surface.
Once one match. is established, more matches can be added if
the resulting match meets the constraints of the node similarity
and topologic relations. After all elements of the object are
matched with those of a model, the object is transformed into
the coordinate system of the model. Finally, not only the object
is recognised as model 6, but also its position and orientation
are determined.
Figure 10 : Line segmentation of edges
6. CONCLUSION
This paper describes the elements of an automatic procedure
for object recognition. The digital photogrammetry system is
data-driven in that no a priori scene knowledge is required. The
descriptions of the objects are computed without any
knowledge about existing models, which is important when the
environment is unknown. The process of object reconstruction
reduces the image data to a few parameters of geometric
functions, which are more meaningful and reliable. However,
the system for the application of object recognition is limited to
industrial components with simple regular shapes. It is not
efficient for complicated objects or objects with occlusion. One
reason is that the processes of edge detection and line
segmentation cannot extract small detail features correctly,
since there is not enough edge information.
7. REFERENCES
Besl P. J. and Jain R. C., “Three-Dimensional Object
Recognition,” ACM Comput. Surveys, Vol. 17, No. 1, pp. 75-
145, 1985.
Brady J., Nandhakumar N. and Aggarwal J., “Recent Progress
in the Recognition of Objects from Range Data,” in Proc. 9th
Int. Conf. Pattern Recognition, pp. 85-92, 1988.
Chin R. T. and Dyer C. R., “Model-Based Recognition in
Robot Vision,” ACM Comput. Surveys, Vol. 18, No. 1, pp. 67-
108, 1986.
Fan T. J., “Describing and Recognizing 3-D Objects Using
Surface Properties,” Springer-Verlag New York Inc. 1990.
Flynn P.J. and Jain A. K., “CAD-Based Computer Vision: From
CAD Models to Relational Graphs,” IEEE Transactions on
Pattern Analysis and Machine Intelligence, Vol. 13, No. 2, pp.
114-132, 1991.
Forstner W., “A Fast Operator for Detection and Precise
Location of Distinct Points, Corners and Centres of Circular
Features,” ISPRS Intercommision Workshop on “Fast
Processing of Photogrammetric Data,” Interlaken, pp. 281-305,
1987.
Huang Y. and Trinder J. C., “A Feature-Based Approach to
Reconstruction of 3D Objects from Digital Images,”
International Archives of Photogrammetry and Remote Sensing,
Vol. 30-3, pp. 391-398, 1994.
Trinder J. C. and Huang Y., “Edge Detection with Sub-Pixel
Accuracy for a Flexible Manufacturing System,” SPIE, Vol.
2067, Videometrics II, pp. 151-161, 1993.
258
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
KEY WO
ABSTRA
It is hop
advisable
condition
1. Introd
The exp
place at
before |
possibilit
continuo
photogra
Procedur
operator
However
automate
analytica
televisior
access t
form of
facilitate:
autocorr
metry we
exclusive
ble to the
the mod
Unfortun
making :
Rapid de
was a I
nature o
PC, we «
of AGH
photogre
[Jachim:
content «
into vect
the PC r
Gagnon
À true
stereogr
place a
[Klaver '
Very co
simultar
still are
