ch, and virtual
f walls and are
, and a mixture
cting an entire
isualize in real
ured aerial and
je to assemble
Ig castles, give
ed.
to finish. The
oblems in 3-D
1 figure 1:
equate number
s.
e to obtain 3D
le. We clearly
symmetry.
ake it difficult
interaction.
r scanning and
ying, and other
to fully model
2 summarizes
; X
lines
etry from
jround
"s
LOT EE
zoe.
|
d
ential use.
conomical and
ollection time,
astle-modeling
laser scanning
y required to:
|. Identify the main problems in creating accurate and photo-
realistic castle models using solely image-based techniques.
. Evaluate using low-altitude aerial images for 3D modeling.
We need to determine the achievable accuracy at the rather
n3
long range from which the images are taken, the quality of
point extraction, and the effect of integrating the resulting
model with ground-based models. Accurately surveyed
points should be used to evaluate the accuracy.
3. Develop a technique to combine image-based models with
each other, and with models from other sources such as floor
plans. The technique must remove overlaps and fill gabs
between models to create a model suitable for visualization.
In the remainder of the paper we give an overview of methods
being used for modeling castles in section 2. This will lead to
analysis of the problems in section 3, then the details of our
approach in section 4. Results of modeling the Stenico castle
are given and evaluated in section 5, followed by conclusions.
2. PREVIOUS WORK
Digital recording of complex sites like castles belongs to a large
body of work in 3-D modeling. Organized by the technique,
related work, albeit not at all comprehensive, is reviewed next.
2.1 CAD with Architectural Drawings / Floor Plans
Traditional 3-D CAD techniques using architectural drawings
remain the most common [Forte et al, 1998, Haval, 2000].
Many use synthetic textures, which yield a computer-generated
look. Textures from images offer more realistic appearance
[DeLeon, 1999, Foni et al, 2002]. The approaches typically lack
automation, although Dikaiakou et al, 2003, modeled heritage
buildings in Nicosia using an automatic building generation
technique based on a library of predefined 3D building blocks.
2.2 Surveying and Interactive Photogrammetry
Hanke et al, 2002, modeled the medieval fortress Kufstein,
Austria, using images taken from a helicopter a by metric film
camera, ground images by non-metric camera, and surveying.
CAD software was used to fill in missing parts. Bacigalupo and
Cessari, 2003, used Photogrammetry and surveying to model
medieval castles in Western Sicily. The methods are still labor
intensive and the projects may take several years to complete.
2.3 Automated Image-Based Modeling
To our knowledge, no large complex-site model was completed
based purely on fully automated image-based techniques, but
sections of castles were modeled automatically [Pollefeys et al,
1999]. Closely spaced images, like low-resolution videos, are
required for robust matching. This may be difficult to acquire
for a complete castle. Even if accessibility is not an issue,
covering a large site with closely spaced image sequences is
time consuming. Also, accuracy becomes an issue on long
image sequences due lo error propagation. However, the
technique can be useful on parts of the site, like stonewalls with
well-defined features. Most of the efforts are focused on the
automatic recovery of internal and external camera parameters
and the stereo matching of extracted points. On the other hand,
acquiring points suitable for modeling and creating the model
itself, which involves segmenting the point clouds into
topologically meaningful groups, remain interactive [Gibson et
al, 2003]. Many techniques attempting full automation of the
modeling process [Werner and Zisserman, 2002, Wilezkowiak
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
et al, 2003], or improving the final models accuracy [Cantzler,
2003], were developed. The techniques rely on constraints of
surface shapes and assumed relationships between surfaces, or
require vanishing points from sets of parallel lines.
2.4 Laser Scanning
Several large-scale projects acquired all the geometry by laser
scanning [Cain et al, 2002, Allen et al, 2003, Fruth and Zakhor,
2003]. Laser scanners promise to provide highly detailed and
accurate representation of any shape [Blais, 2004]. Combined
with color information, either from the scanner itself or from a
digital camera, a realistic-looking model can be created. The
accuracy at a given range varies significantly from one scanner
to another. Also, due to object size, shape, and occlusions, it is
usually necessary to use multiple scans from different locations
to cover every surface. Aligning and integrating the different
scans will affect the final accuracy of the 3-D model.
2.5 Combination of Multiple Techniques
Borg et al, 2002, modeled parts of a temple in Malta using
Photogrammetry for outlines and laser scans for details, plus
surveying to register the data. Georgopoulos and Modatsos,
2002, utilized available tourist slides, and existing engineering
drawings and geodetic measurements to model the church of
Holy Sepulchre in Jerusalem. Bundle adjustment with self-
calibration on the digitized slides, with geodetic and
engineering drawings measurements as control, was used in that
7-years-long project. El-Hakim et al, 2003a, used aerial and
terrestrial images for the main shapes, and laser scanning for
fine geometric details to fully model the abbey of Pomposa in
Italy. Flack et al, 2001, developed tools specifically designed to
assemble models created by various techniques.
3. CASTLE MODELING: ISSUES AND ANALYSIS
The following sub-sections discuss the problems, in different
categories, of existing methods applied to castle modeling.
3.1 Data Acquisition
As mentioned above, castles setting and the multiple lines of
defense make it difficult to find suitable locations to take
images or scans. Also the size and complexity of the structures
will require a huge amount of data to cover all the details. Low-
flying helicopter may be the solution to cover outer walls, roofs,
and courtyards with a practical number of images. Ground
images or scans at selected locations can furnish the details of
some of the areas not visible from the helicopter.
3.2 Modeling from Floor Plans
After acquiring all possible sensor data, we may still have some
sections without coverage. These parts can be determined from
existing architectural drawings or floor plans. In the case of
drawings created directly in digital form from surveying data,
file formats such as AutoCAD or DXF allow the information to
be arranged in separate layers, each containing a different
element. Some adjustments to obtain consistent local geometry
and layout topology are needed. Creating 3-D models may be
done interactively on simple floor plans, but can be time
consuming and error prone on complex buildings. Thus, some
semi-automatic techniques were developed to create 3-D
models from floor plans [Lewis and Sequin, 1998]. Commercial
CAD software does not address the situation where a floor plan
