International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
determination of the orientation of each used photo. The
approx. 10cm large signals (see Fig. 3) were self’ constructed
and printed on white paper. Each castle was supplied with
several targets (Ahrensburg 22, Celle 47, Gliicksburg 28),
which were well distributed over the object and fixed mainly on
windows.
Fig. 3: Geodetic control point determination and signalisation
of control points (upper left and right corner)
For the determination of the coordinates of the signalised
control points a local geodetic 3D network (see Fig. 3) was
designed and measured with a Leica total station. All control
points were measured in the geodetic network. The coordinates
of the control points were determined in a geodetic 3D network
adjustment using the software PANDA from GeoTec, Laatzen,
Germany. At Ahrensburg castle 10 natural points, clearly
identified and well distributed over the four facades, were
measured as additional control points. A standard deviation of
better than Smm was achieved for the coordinates of all control
points.
4.2 Photogrammetric Data Processing
The determination of the orientations of each photo was
performed by multi image triangulation. To connect all photos
in an overall triangulation photo block, all control points and
selected natural tie points were measured manually in the
digital images. 37 images of block Ahrensburg, 254 images of
block Celle and 188 images of block Glücksburg were used for
photo triangulation. In Fig. 4 the image point measurements for
tie and control points are illustrated exemplarily using the
software PICTRAN D (Release 3 and 4) from Technet GmbH,
Berlin, Germany.
Fig. 4: Image point measurements of tie points (upper) and
control points (lower) in one image of Celle castle
In all three photo triangulation blocks more than 10 well-
distributed points per image were measured in order to model
the lens distortion as one of the camera calibration parameters
by bundle block adjustment. This also guaranteed a well-
controlled connection of all images in the block. The digital
SLR camera Fuji SI is a non-metric camera; therefore the
interior orientation parameters must be determined in a bundle
block adjustment, which could be performed simultaneously in
PICTRAN B with the determination of the photo orientation
parameters. The interior orientation parameters include the
location of the principle point, the camera constant, the radial
distortion, the decentering distortion, the shear and the affinity
parameters. The results of the bundle block adjustment for each
castle project are summarized in table 2. The o; of each project
was 4.5micron or cven better, which is correspondent to approx.
half a pixel in image space. The RMS of the control points in
XYZ was better than 5mm for each project; while for
Ahrensburg and Celle the RMS was better than 2mm. Each
triangulation block was sufficiently fixed using a large number
AR CD > ine * n , (d 78 ag
Cal ve # # s # Rays | #IP/ 6. RNC # empirical accuracy maximum values
CP | OP perpt | Ph | [um] Cup |," | ur [He y MN) ME |
[mm] [mm] | [mm] | [mm] | [mm] | [mm] | [mm]
Ahrensburg 37 | 22 | 155 | 780 3 21 2.8 1.6
Celle 254 | 47 | 247 | 3530 14 14 4.1 1.6 27 4.8 27 2.4 15.6 | 8.0 7.8
Glücksburg 188 | 28 | 543 | 4793 9 25 4.5 4.6
Ph... Number of photos IB. eee Number of image points chP Number of check points
CP eer Number of control points Gu ee pns Sigma aposteriori from the adjustment
OP. Number of object points RMS............ Root Mean Square of the control points
Table 2. Results of the bundle block adiustment
Internation
Le
of image px
the average
images on
connection
accuracy
photogramr
project Cell
than 5mm -
Such resul
reconstructi
could be a«
be usual fot
Fig. 5: M
epipolar lin
After the
calibration
the 3D eva
image poin
reconstructi
points were
were acqui
intersection
performed |
superimposi
measuremet
the operator
mistakes. T
were deter
deviation o
online qual
Fig. 6: Me
an
