Istanbul 2004
es and, when
sensor data.
m floor plans.
teresting parts,
coverage.
ections.
> unknown and
elf-calibration,
needed. Since
1akes imaging
quality camera
Many modern
then calibrate
figure 4).
ir and ceiling.
rrain and roof
: helicopter at
roofs, outside
ng walls, and
ng geometric
base-to-height
urveyed points
scale.
- Plans
Irposes:
ning.
surface shapes
are often not
site.
eed to provide
nd connecting
heights. The
dels. Window
-automatically
| as entrances,
or spaces. We
with selected
ve select two
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
images with the best geometric configuration, and apply our
hierarchical matching procedure developed in El-Hakim, 1989
(similar idea is proposed by Ferrari, et al 2003). Corners are
automatically extracted followed by stereo matching procedure,
constrained by the epipolar line and disparity-range determined
automatically from the seed points. Finally, 3-D points on
regular shapes like columns, arches, doors, and windows are
created semi-automatically using also selected seed points, as
described in El-Hakim, 2002.
Figure 5: Top pair with manually picked seed points, bottom
pair with automatically matched points on a castle entrance.
Figure 6: Two adjacent models connected by a portal.
4,5 Model Assembly
Combining models created by different data sets, such as the
two models shown in figure 6, must address several issues so
that the final model is appropriate for 3-D visualization:
Relative scale and orientation must be correctly determined.
Joint primitives, specifically surfaces, edges, and vertices,
from adjacent models must match perfectly. However, it is
unlikely that we have the same primitives between the
adjacent models. For example an arch may have 50 vertices
in a detailed model but only 8 in a general model.
Sensor-based techniques produce actual wall surfaces rather
than perfect planes. In contrast, wall surfaces from floor
plans are extruded using user provided heights, thus they are
exactly planes. This can be visually noticeable for adjacent
surfaces between the two types of model.
* No gaps, redundant or intersecting edges are acceptable.
Commercial modeling tools do not deal with these issues, thus
special techniques and software tools were needed. The models
can be registered manually in modeling software such as 3ds
max” using common points between them. The process will be
unnecessary if the individual models were directly created in
the same coordinate system using control points. Many models
however will not have any access to control points and will
need interactive registration. Once all models are registered, the
123
integration procedure begins. The models are organized in a
hierarchical manner where the top model contains the least
details (Figure 3). We import points from the detailed model
along the perimeter of common surfaces into the less-detailed
model. Then we adjust the latter's mesh with the new added
points to create a hole into which we insert the detailed model
without overlaps. Any remaining gap between models is filled
from the floor plans. Then points from adjacent models on the
borders of the gap are used to re-triangulate it so that we have
realistic surfaces rather than perfect planes in the filled gap.
5. MODELING OF THE STENICO CASTLE
The above procedure is applied to modeling the Stenico castle,
which consists of the following components (figure 7):
* Buildings of mixed styles organized around 4 courtyards. The
buildings include the 12th-century House of Nicholas and
Walled House, the 13" century Court of Justice and Council
Hall, and the 15" century House of Johan Hinderbach.
Inner and outer tall thick walls with arched gates.
A Renaissance loggia built by Clesio.
The clock tower and the Fune or Bozone's tower.
Ramps and staircases to the multi-level building entrances.
The 13" century San Martino's chapel with medieval frescos.
Rooms with Renaissance frescos.
External
Wall
aci
Aerial View
Entrance San Martino Chapel
Figure 7: Sample project images of some castle elements.
5.1 Data Acquisition
We used a 5-mega-pixels Olympus“ E-20 digital camera. Figure
8 shows camera positions from the helicopter over the castle.
e
e
o S «
>
$
- * aic b £e
e»
Figure 8: Actual locations of the used aerial images.
The same camera was used to take several sets of ground
images for outside detailed and room models. The imaging,
both aerial and terrestrial, took a total of 4 hours in two visits to
the site. Two camera settings were used during the project, one
