PHOTO-REALISTIC 3-D RECONSTRUCTION OF CASTLES WITH
MULTIPLE- SOURCES IMAGE-BASED TECHNIQUES
an . * . . > A : . m" :
- Lorenzo Gonzo *, Sabry El-Hakim "', Michel Picard ", Stefano Girardi *, Emily Whiting *
* Center for Scientific and Technological Research, ITC-irst, Trento, Italy - (lgonzo; girardi)@itc.it
^ Visual Information Technology (VIT), National Research Council, Ottawa, Canada -
(Sabry.El-Hakim; Michel.Picard)@nrc-cnre.ge.ca
* Department of Architecture, Massachusetts Institute of Technology (MIT), Cambridge, MA, USA
Commission V, Working Group V/2
KEY WORDS: Architecture, Cultural Heritage, Integration, Modeling, Virtual Reality, Photogrammetry, CAD.
ABSTRACT:
Digitally reconstructing a large and complex heritage site, particularly medieval castles, for documentation, research, and virtual
reality simulation is a challenging task. Castles by design contain many lines of defense such as inner and outer rings of walls and are
often located on high grounds, surrounded by water or forests. They also have protective entrances, towers, courtyards, and a mixture
of large complex buildings. Although laser scanners have many advantages, they may be impractical for reconstructing an entire
castle. They are not portable or fast enough and can generate a huge amount of data that is difficult to register and visualize in real
time. Therefore, we set out to fully explore the potential of only image-based techniques for modeling castles. We captured aerial and
ground-based images and used existing floor plans and limited surveying, which compelled us to develop a technique to assemble
and integrate models made from different data sets. We will discuss the issues and problems associated with modeling castles, give
details of our approach, and present and evaluate the results of modeling the Stenico castle in Trentino, Italy, as a test-bed.
1. INTRODUCTION
Castles are unique structures that were assembled over different
periods and follow no conventional architectural design. They
were built for defensive purposes and had to make the most of
readily available natural protective settings like hilltops and end
of ridges. They were fortified with multiple rings of high walls,
gates, and towers. As a result, data acquisition and modeling of
castles are complicated. Two types of model are desired, one
for a photo-realistic walkthrough animation or a movie that
shows 'every attribute, and one for interactive visualization
where trade-off between true-to-life details and speed (limited
by computer hardware) must be made. The latter model is the
most challenging and also it is the most demanded by end users.
Figure 1: Aerial views of Stenico castle over Stenico village.
We initiated a castles documentation and visualization project,
beginning with the Stenico castle (figure 1). The castle is one of
the oldest and most important medieval castles in Trentino, and
is an interesting mixture of styles of buildings added over
several centuries. A view of the castle is depicted in the January
panel of the "cycle of months" frescoes in Aquila Tower at
Buonconsiglio castle in Trento, which we modeled in another
project [El-Hakim et al, 2003b.] The goal of the new project is
to reconstruct all castle buildings, courtyards, protective walls,
and interiors particularly frescoed rooms. The cost and data
collection time must fit the limited budget and granted access
* .
Corresponding author
time, noting that comparable projects took years to finish. The
planning phase of the project revealed several problems in 3-D
reconstruction of castles, as can be perceived from figure 1:
—
. Due to castles location, it is hard to find an adequate number
of places from which to capture images or scans.
. The assumptions made on standard architecture to obtain 3D
data and increase automation are not applicable. We clearly
cannot assume parallelism, perpendicularity, or symmetry.
3. The complexity and variety of architecture make it difficult
to capture all details without extensive manual interaction.
N
Previous work suggested.that combination of laser scanning and
Photogrammetry both aerial and terrestrial, surveying, and other
data such as maps or floor plans, are all needed to fully model
large complex sites [EL-Hakim, 2001]. Figure 2 summarizes
available techniques and their anticipated use.
A
Reference frame, major elements and outlines
Photogrammetry from
aerial and ground
images
Existing drawings
and floor plans Surveying
|
Y
Eme geometric details
Close range
Laser scanning |
Image sequences |
1
v
Figure 2: The available techniques and their potential use.
Since image-based techniques entail the most economical and
portable sensors and the shortest on-site data collection time,
we decided to carry out the first phase of the castle-modeling
project with these techniques. We expect to add laser scanning
in the next phase. In the first phase we specifically required to:
International
Aie HOT ny
|. Identify tl
realistic Ce
. Evaluate 1
We need
long rang
point. extr
model w
points sho
3. Develop :
each other
plans. Th
between n
na
In the remai
being used f
analysis of 1
approach in
are given an
Digital recor
body of woi
related work
2.1 CAD wi
Traditional :
remain the
Many use sy
look. Textui
[DeLeon, 19
automation,
buildings in
technique ba
2.2 Surveyii
Hanke et al
Austria, usir
camera, gro
CAD softwa
Cessari, 200
medieval cas
intensive anc
2.3 Automa:
To our know
based purely
sections of c
1909].. Clos
required for
for a compl
covering a ]
time consun
image sequ.
technique ca
well-defined
automatic re
and the stere
acquiring po
itself, whicl
topologically
al, 2003]. M
modeling pre