© -^ N oO £& O
T
1
© -— N 0 2 On
T
—
T -220 : + T
j -221
] -222
| -223
j -224
| -225
-226
0 2 7465 145-19 "4 3
X
e
6
oo}
o
Oo
NE
A
o
ot
10
Figure 4: Results of the estimation process. Calibrated values are 0 mm for z and y, -220 mm for z. View size is
approximately 5 cm, and camera distance 20 cm.
depending on estimation level). The needed time on a
Sun SS2 workstation is in the range of 0.2-12 seconds,
depending on the estimation level. The used image size
is 64 x 64 pixels. In figure 4 are the results shown for
the estimation process. The raw image data comparison
level is shown overhere.
Extrapolating these results to the presented case study,
convergence of the estimation process is likely in a few
seconds, using a normal workstation, when estimating
at the level of segmented images. When estimating at
raw image data longer run-times are anticipated, be-
cause R. S. images generally have a higher resolution,
and the calculation of the individual pixel values is the
time limitingstep. Expected precision of the 3D param-
eters is at pixel accuracy for the associated 2D shifts.
5 Discussion
The approach to apply model based knowledge engineer-
ing to hypotheses generated from a 3D GIS and to use
RS data as evidence for the updating of the (likelihoods
of) 3D GIS in terms of object classes and parameters
has been shown to be a robust one. Like in object ori-
ented software, the method hides ” information” about
the actual representation of 3D objects. Only the object
parameters are accessible (exported) for each instantia-
tion of a class of primitive objects.
Further work is needed for the determination the com-
putational feasibility of the method by applying the
method to complex problems like building a GIS of the
University of Twente campus.
References
[1] Detecting Building in Aerial Images, A. Huertas
and R. Nevatia.
Computer vision, graphics, and image processing
41, 131-152 (1988)
[2] Towards Automatic Cartographic Feature Extrac-
tion, David M. McKeown, Jr.
Mapping and Spatial Modeling for Navigation,
NATO ASI Series, Vol F 65, Springer-Verlag Berlin
Heidelberg 1990
968
[3] Model Construction and Shape Recognition from
[4
Sd
Occluding Contours, Chiun-Hong Chien and J.K.
Aggarwal.
IEEE transactions on pattern analysis and machine
intelligence, voll1, no 4, April 1989
Using Perceptual Organization to extract 3-D
Structures, Rakesh Mohan and Ramakant Nevatia.
IEEE transactions on pattern analysis and machine
intelligence, volll, no 11, November 1989
A Spatial Structure Theory in Machine Vision and
its Application to Structural and Textural Analysis
of Remotely Sensed Images, He Ping Pan.
Ph.D. Thesis, Enschede 1990, ISBN 90-9003757-8
Evaluation of comparison levels in iterative estima-
tors, A.J. de Graaf, K. Schutte.
Proceedings ICPR11, August 30 - September 3,
1992, The Hague (to be published).
ABS'
This ]
rior o!
ing m
and r
stitut
as the
tify a
and s
image
image
searcl
descri
KEY
Featu
In re
togra
aerial
els, c
[Sche
1991]
corre
Using
that |
indee
Anot
ages
prog:
ones
minii
betw
a mo
tatio
patcl
tion
with
road:
cont:
This
ages
(lanc
the ¢
meas
easy
The
tiona
the «
scrip
weat
scrip
mati
