. Istanbul 2004
is (upper) and
e
than 10 well-
order to model
ion parameters
inteed a well-
ck. The digital
therefore the
ied in a bundle
nultaneously in
Oto orientation
rs include the
tant, the radial
and the affinity
stment for each
of cach project
ident to approx.
ontrol points in
ect; while for
an 2mm. Each
a large number
um values
Hy Hz
nm] | [mm]
8.0 7.8
ck points
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BS. Istanbul 2004
of image point measurements (minimum 14 points per image in
the average) and cach object point was measured in at least 5
images on average guaranteeing high reliability in the overall
connection of the images. As an example, the empirical
accuracy of the signalised points (comparison of
photogrammetrically and geodetically determined points) in
project Celle is summarized in table 2. The accuracy was better
than 5mm on average, while the maximum value was 16mm.
Such results are more than sufficient for 3D building
reconstruction. However, it 1s not assumed that the same results
could be achieved with poorly-defined object points as would
be usual for historical buildings.
Fig. 5: Manual image point measurements supported by
epipolar lines for improved point finding in PICTRAN D
After the update of the camera file using the adjusted
calibration parameters and the exterior orientation parameters,
the 3D evaluation could be conducted by digitisation (manual
image point measurement) of all necessary points for the
reconstruction of the building. For reasons of reliability all
points were measured in at least three different images, which
were acquired from different positions to ensure optimal
intersection geometry. All manual measurements were
performed in PICTRAN D as points, lines or polygons. The
superimposition of the epipolar lines in the images after
measurement of each point (Fig. 5) offers significant support to
the operator in finding the points quickly and in avoiding point
mistakes. The 3D point coordinates of each measured point
were determined by spatial intersection and the standard
deviation of each point was shown in the user interface for
online quality control. Similar parts of the buildings, e.g.
windows, were measured only once in detail, while for all
varied details the position of this building part was measured
using only three points for the later fitting in the constructed
model. Due to the complexity of historical buildings some parts
of the building, e.g. bent walls or ornate entrance doors (Fig. 6)
were generalized.
The digitised 3D points were then transferred for further CAD
processing to AutoCAD via the DXF interface. As a strategy
the entire building was divided into certain object parts, which
were measured in PICTRAN and transferred afterwards to
AutoCAD for reconstruction.
5. CAD RECONSTRUCTION AND VISUALIZATION
5.1 CAD reconstruction of the castles
The detailed 3D reconstruction of all three castles in their
entirety was performed stepwise in AutoCAD: ie. each
building was constructed in the following sequence: ground
plan, walls, towers, roofs, windows, entrances (doors), and
finally the assembling of all objects to one complete volume
model. Additionally, the immediate terrain environs of each
castle, which were measured by geodetic methods, were
modelled with AutoCAD Land Development as a digital
elevation model (DEM). The corresponding DEM was later
integrated into the entire virtual 3D model of the castle to
produce a better visualization of the adjacent site. The result of
the construction is a 3D volume model of each castle. Fig. 7
shows a wire frame and rendered model of Celle castle. All
information in the 3D AutoCAD file is structured in layers; i.e.
the same type of object, e.g. walls, windows, frames or glasses,
etc., was saved in a special defined layer. The 3D model of
Ahrensburg consists of 78 different layers, while the CAD file
of Celle and Gliicksburg consists of 114 and 39 different layers,
respectively.
Some perspective scenes from different viewing positions were
computed in AutoCAD for each castle, in order to check the
quality of the 3D model by rendering the volume model. The
DWG file size (AutoCAD 2002) of the 3D models amounted to
243 Mbyte for Ahrensburg, 260 Mbyte for Celle, and 71 MB
for Gliicksburg. Using the new software release AutoCAD
2004, the file size can be significantly reduced by a factor of
three to five. Nevertheless, this amount of data caused problems
for the visualization due to the computer performance of a
typical workstation (two parallel processors of 2.4 GHz, an
internal 1 GB RAM, and a fast graphic card nVidia Quadro4
700 XGL), compared to the file sizes. The computation of one
perspective scene with 3D Studio VIZ took approx. 15 minutes,
Fig. 6: Measurement, reconstruction and rendering of a window of Celle castle (from left to right), ornate doorway as a photo
and as a generalized reconstructed doorway (right)
