background image
DigiCULT 51
of cultural heritage were not properly designed for
its requirements. Piva expected that the `design and
standardisation of secure data hiding technologies for
IPR protection' might be achieved in 2010, while
advanced image processing tools, developed in coop-
eration between heritage experts and ICT specialists,
could appear in 2008.
Fabrizio Cardinali, Fabrizio Giorgini and David
Fuschi (Giunti Labs, Italy) saw major limitations in
the performance of current 3D capturing systems as
well as shortcomings in the languages used to code
3D objects. They expected that improved 3D acqui-
sition systems could become available in 2007, bet-
ter algorithms for 3D reconstruction in 2008, and a
new efficient language for coding 3D objects (beyond
X3D) around 2010.
Actually, the RTD strand of 3D objects, models
and environments received most attention. Sofia Pes-
carin (Fellowship Researcher, CNR ITABC, Italy)
saw a need for more powerful 3D laser scanners with
a capture rate of 25 scenes/s and with integrated GPS.
A major gap she considered to be the fact that the
parameters of the commercial market would not suit
applications for the cultural heritage sector. Howev-
er, Pescarin also saw major difficulties in establishing
interdisciplinary groups that would share and exploit
3D objects and other advanced objects among insti-
tutions and projects. Therefore, appropriate conditions
would need to be set and mandated for, e.g. EU-
funded RTD projects.
Kirk Martinez (Senior Lecturer, University of
Southampton, UK) also urged that 3D scanners
and AR/VR software would need to be developed
to allow for a much easier deployment in heritage
organisations. He expected that this could be achieved
within 3 to 5 years if funded, but up to 10 or more if
not. As the RTD target, he saw European manufac-
turers being able to provide the heritage sector with
3D scanning technology specifically adapted for cul-
tural heritage applications.
Last but not least, Luc van Gool (Computer Vision
Laboratory, ETH Zurich, Switzerland) provided us
with a detailed roadmap for RTD towards the future
generation of 3D scanning, modelling and automat-
ic registration. Gool described the current main lim-
itations of RTD in this area, which included a fairly
low scanning speed (although fast structured light and
stereo methods were becoming available). Wide base-
line matching was considered difficult in general, but
would need to be solved, e.g. for automatic crude
registration. Furthermore, improving 3D acquisition
systems would need a combination of multiple strate-
gies into a single system; the same would be required
for combined geometric­radiometric analysis. Howev-
er, Gool thought that the following might be achieved
10 to 15 years down the road:
`1/ On-line scanning: 3D models should be built
up during the scanning. The model is being shown
on the screen while performing the scanning, as it
grows. One can immediately inspect the completeness
and the quality of the result. Off-line data processing
has the disadvantage that the lack of certain data (i.e.
the presence of holes in the model) or erroneous data
only become clear when it is too late.
2/ Integrated shape and surface reflectance model-
ling. 3D scanners do yield 3D shapes, but not the pre-
cise radiometric properties of the surfaces, which are
nevertheless crucial to fully visualise the true appear-
ance of the object.
3/ Opportunistic 3D modelling: currently there is
no single 3D acquisition technique that can deal with
all types of shapes and surfaces. Methods may have
difficulties with narrow structures, deep cavities, spec-
ular reflectances, etc. Future systems could be hybrids
that deal with much broader classes of objects by
(automatically) adapting their modelling strategy.
4/ Automatic registration: typically, 3D data come
in as partial patches. These then have to be glued
together to form complete models. There are good
solutions for fine-registration, but these require good
initialisations, which are usually obtained by first man-
ually puzzling the pieces together in more or less cor-
rect relative positions. Fine docking is then done
automatically. This crude puzzling also ought to be
done automatically, i.e. there still is a need for robust,
crude registration algorithms.'
As areas where breakthroughs in RTD could be
achieved in the shorter term, Gool considered stable
online scanning, for which so far only academic
prototypes have been demonstrated (around 2007);
and combined geometric and radiometric analysis ­
`to a large extent an issue of efficient optimisation
algorithms' (around 2008). These, and possible RTD
breakthroughs considered to be more demanding, are
included in the table below.
Professor Gools' analysis of 3D scanning, modelling
and registration shows that there are still major
bottlenecks for the technology to be declared fully
mature. We would also like to point out that many
experts and practitioners from the heritage sector stat-
ed a considerable demand for high-quality 3D scan-
ning technology, less so for AR/VR, and they urged
that this needed to be adapted specifically for the sec-
tor's purposes and budgets.
DCTHI7_271104.indd 51
06.12.2004 8:38:04 Uhr