ibul 2004
in at least
ring algo-
point, that
point, is
ed image
the side).
art, where
ints might
stering al-
All points
ed (Fig. 4
its on the
stering. À
92
The right
ure shows
S.
ts, the use
ality anal-
[herefor a
equal to a
ints found
2$.
ight of the
ate of the
e approxi-
lata before
; using the
d determi-
adjustment
he clusters
nts, which
is clusters
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
d
i 4
SIE TO TTT TTT TTT TT RT RTT
Figure 5: Quality analysis 3: comparison of the point prediction
between the strict and the approximated mapping. Left figure:
number of points found as equal, depending on the used threshold
e. Right figure: histogram of a x4, depending on threshold e.
Table 4: Quality analysis 4: Comparison of the final estimation
using the strict and the approximated model
| | Test 7]
number of points (e € 0.001) 149
Zmas(€ € 0.001) 1:0 - 10^ [ciii
YmazlE < 0.001) 3.0 - 107? [em]
Zmanle < 0.001) 2.0 10 [em]
number of points (€ < 0.1) l
number of points (e € 0.15) 2
number of points (€ > 0.15)
or not found 4
(quality analysis 4), the corresponding object points were deter-
mined by fixed parameters of the orientation of the imaging sys-
tem. A cluster is defined as equal, if the difference between the
reconstructed points is smaller than 0.001 cm. The results are
given in Tab. 4. 149 clusters are identical, 1 object point has
a difference which is smaller than 0.1 cm, 2 points smaller than
0.15 em and 4 points have a bigger difference than 0.15 cm or
were not found by using the virtual projective camera as an ap-
proximation.
For quality analysis 2 we compare the final estimated 3D points
using the strict model of the multi media mapping and the approx-
imation. The error is given in Fig.fig:histogram. The differences
are normal distributed.
Fig. 7 shows the digital terrain model of the sediment surface
resulting from the estimated object points.
number of ports
Figure 6: Quality analysis 2: comparison of the final estimated
3D points using the strict model and the approximation.
6 SUMMARY
In this paper we introduced a classification of optical mappings
based on the geometry of the imaging system having a single
viewpoint or a non single viewpoint. From this classification
we got different kinds of image distortions: image space based
and object space based. The models for optical mappings be-
longing to the second kind of mappings need information about
611
Figure 7: Reconstruction of the sediment surface resulting from
the matched points using the virtual camera.
the scene structure and special complex algorithms for the pro-
jection between object and image space. Under this background
we surveyed the multi media geometry. We presented a method
to calculate a virtual projective camera which approximate the
strict non projective mapping. The approximation was used for
a point matching process using multiple views of a sediment sur-
face with multi media geometry. We introduced a new matching
process for multiple views based on geometric constraints alone,
which is usable for projective mappings and the approximation
of non projective mappings. Different quality tests show, that the
approximation is sufficient for the reconstruction of a sediment
surface.
ACKNOWLEDGMENT
This work results from a interdisciplinary project Geometric Re-
construction, Modeling and Simulation of Fluvial Sedimental Trans-
port in the Special Research Centre (Sonderforschngsbereich) SFB
350 Continental Mass Exchange and its Modeling, at the Insti-
tute of Photogrammetry, University Bonn, Germany. The author
wishes to express her gratitude to the Institute of Geodesy and
Photogrammetry, ETH Zurich, Switzerland to make it possible to
present this work at the ISPRS Congress 2004, Istanbul.
REFERENCES
R. Hartley and A. Zisserman. Multiple View Geometry in Com-
puter Vision. Cambridge University Press, Second Edition, 2003.
H.-G. Maas. Complexity analysis for the determination of image
correspondences in dense spatial target fields. In /4PRS, Vol 29,
Part BS, 1992.
H.-G. Maas. New developments in Multimedia Photogramme-
try. In Proc. Optical 3D Measurement Techniques. Wiechmann-
Verlag, 1992.
H.-G. Maas. Mehrbildtechniken in der digitalen Photogramme-
trie. ETH Zurich, Institut ff Geodásie und Photogrammetrie, Mit-
teilungen Nr.62, Habilitationsschrift, 1997.
S. K. Nayar and A. D. Karmarkar. 360 x 360 Mosaics. In Proc.
CVPR, pages 1:388-395, 2000.
R. Swaminathan , M.D. Grossberg, S.K. Nayar. Caustics of Cata-
dioptric Cameras. In Proc. ICCV, pages IT:2-9, 2001
R. Swaminathan, M. D. Grossberg and Shree K. Nayar. A Per-
spective on Distortions. In Proc. CVPR, pages 594-601, 2003
K. Wolff and W. Fôrstner. Exploiting the Multi View Geome-
try for Automatic Surface Reconstruction using Feature Based
Matching in Multi Media Photogrammetry. In 19° ISPRS
Congress. Amsterdam, 2000.
K. Wolff and W. Förstner. Efficiency of Feature Matching for
Single- and Multi-Media Geometry Using Multiple View Rela-
tions. Ih Proc. Optical 3D Measurement Techniques. Wien, 2001.
