calibration, depending on which model parameters are
treated as a priori known or unknown.
The functional model describing the relationship between
the derived and measured quantities consists of the well
known colinearity equations (Eq. 12). For each pair of im-
age coordinates (x, y) observed on each image the follow-
ing pair of equations is written:
Il
Il
U
E, (Am Xp) = Cy + AX
V (Eq. 12)
Frey ero FAV
with the auxiliaries:
U X-Xo
v| = DB(@, 9, K) |Y - Y, (Eq. 13)
W Z-Zo
with the image coordinates (x, y), the elements of interior
orientation (Xp, yp. C). functions of additional parameters
(Ax, Ay), the object coordinates (X, Y, Z), a3 by 3 orthog-
onal rotation matrix D with the three rotations Q) €, K and
the object coordinates of the perspective centre (Xo, Yo,
Z9). Starting from observed values the overconstrained
equation system (Eq. 12) can be solved leading to esti-
mates of the unknown parameters (cf. Eq. 5,6). In most
applications it can be assumed that the exterior and interi-
or orientation parameters of the camera are unknown.
In the next step the derived object coordinates are re-pro-
jected into each image and used to restart the 2D feature
extraction. This proceeds until the estimated unknowns do
not change significantly form on iteration step to the next.
The result of this process is then transferred with addition-
al information about precision and reliability to the CAD
environment, which allows the final judgement by the hu-
man operator.
4. PRACTICAL EXAMPLE
The work described in this paper was done as part of the
contribution to the CIPA project “Wagner-Pavillon”. The
idea and the initiative of this project belongs to P. Wald-
Figure 5: Otto-Wagner-Pavillon, Vienna.
552
háusl (Waldháusl, 1991) and the aim was to check the cur-
rent state-of-the-art in architectural photogrammetry.
The Otto-Wagner-Pavillon is one of Otto Wagner's Stadt-
bahn Station buildings (see Fig. 5) on the Karlsplatz in Vi-
enna, built in 1898/1899. The dimensions of the building
are 15 x 8 x 10 m?. The object coordinates of 44 non-sig-
nalized (but well defined in the majority) control points
have been determined geodetically with an accuracy of
2 mm.
4.1. Image acquisition
In this test the image acquisition was performed with an
inexpensive S-VHS camcorder. Although camcorders are
not intended for photogrammetric use, they exhibit many
useful characteristics. They are inexpensive and widely
used for other purposes as well, they are portable and free-
hand, they need no special equipment, they offer the abili-
ty of on-site quality control. Furthermore they provide
very inexpensively means for storage of huge amount of
video data on video tapes.
The JVC GR-S77E camcorder is a free-hand portable
camera, which allows on-site control for the acquired im-
agery via an internal monitor. The camera incorporates a
1/2" colour sensor (6.4 x 4.8 mm?). The analog images
are stored on a S-VHS video tape and have to be digitized
by a framegrabber. The digitized images have a size of
728 x 568 pixel, which results in a pixel spacing of 8.8 uum
in the horizontal and 8.5 jum in the vertical direction. The
focal length (zoom lens) can vary between 8.5 mm and
65 mm, which was fixed at its shortest focal length during
the image acquisition.
4.2. Object reconstruction
The image acquisition took place in a way that an image
sequence from the object was generated by a person walk-
ing with the camera around the object and filming all four
facades. From this video tape 38 single frames were ran-
domly digitized with a framegrabber. The mean distance
between the camera and the object was about 16 m, which
results in an image scale of 1:1800.
The measurement of image features, the estimation of ex-
terior orientation parameters and additional camera pa-
rameters for self-calibration as well as the determination
of the object coordinates was performed with the feature
measurement routine as described above.
A total of 1611 object coordinates was determined to de-
scribe the entire building. Therefore 2589 image points
have been measured semi-automatically by the feature
measurement routine. The normal equation system, which
has to be solved consists of 5178 observations and 1726
unknowns. The precision of the object coordinates was
determined as a part of the bundle adjustment routine. The
results indicate a precision in object space of 1.1 cm in X,
1.3 cm in Y, and 0.7 cm in Z, with X- and Y-axis in the
plane and Z-axis in height. This corresponds to 6.7 pm in
image space. A more detailed description of the numerical
results of this test is given in (Streilein, 1995).
These results are comparable to the results derived with
classical photogrammetric equipment in the CIPA test
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
Fi
(Pa
pro
pre
can
is
Th
DII
jec
env
tec
line
The
cap
ani
der
thre
Thi
pro
an.
of 1
usi
tim
aut
reli
fin:
ope
per
