International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B6. Istanbul 2004
In order to minimize the photogrammetric expertise during the
plotting phase, the use of knowledge is made in three different
steps in the photogrammetric data processing.
2.1 The first step: geometrical knowledge
Geometrical knowledge can be used in order to help the
homologous point determination. We know that in the field of
architectural or archaeological photogrammetry we can often
have a geometrical approximation of the measured surface. In
this context the I-MAGE process (as Image processing and
Measure Assisted by GEometrical primitive) (Drap,
Grussenmeyer, Gaillard, 2001) is a convenient help for the user.
After the manual measurement of some points on the object-
surface, the user can focus his attention on the semantic way,
observing only one picture when the system makes
automatically 3D measurements. In this paper we present
namely DSM tools based both on a surface approximation and
on image processing considerations.
2.2 The second step: archaeological knowledge
The second step is using archaeological knowledge to obtain a
complete representation of a measured artefact. This work,
published last year, (Pierre Drap, Julien Seinturier, Luc Long.
2003), is managed in three phases:
1) Development of the theoretical model: for each identified
object, a geometrical description offers a whole of geometrical
primitives, which are the only objects to be potentially
measured, and a theoretical representation of the object.
2) As photogrammetric measurements are highly incomplete
(the object is seen only partially or may be deteriorated), the
Expert System determines the best strategy to inform all the
geometrical parameters of the studied object, starting from the
measurement process and handling the default data as defined
in the architectural model and the geometrical model. The
expert System used is Jess, for more details you can refer to
(Jess, 2001)
3) The resulting object is thus based on a theoretical model,
dimensioned more or less partially by a photogrammetric
measurement. During the exploitation of the photographs the
operator can choose the number of attributes of the object
which are relevant to measure. The choice of attributes is
revisable in time, as for example during a second series of
measurements. The system can be used to position in space
some objects from a catalogue after a scaling process. For an
extended measurement process, in addition to positioning in
space, the system allows analysing how the measurements vary
from the theoretical model (e.g. deformations or erosion). After
all the system allows checking the model relevance.
This approach has been used in underwater archaeological
surveys, during the excavation of the Grand Ribaud F Etruscan
wreck with more than 1500 amphorae of the same typology.
(Grand Ribaud F, 2000-2004)
2.3 The third step: knowledge to manage results
In an archaeological context, a lot of data is produced from
various sources. Heterogeneity is one of the main problems of
this kind of applications. It can be found at several levels from
the given data to the target format. We mainly focus on three
kinds of heterogeneity due to multiple data sources, differences
between objects of study and changes between versions.
2.3.1 Heterogeneous data sources
The first kind of heterogeneity comes from the differences
between data sources. In our application, they can be divided in
124
two main families whether they are calculated or produced by
an expert. The first ones are often well structured (i.e. vector or
matrices) whereas the second ones are less structured (natural
language). We call them semi-structured data and it is important
that they can be more structured than full text. We are dealing
with multimedia data, mainly pictures and semi-structured text.
That is why the data model must provide a way to federate
heterogeneous data across multiple data sources. The user must
be provided with a unified view of data.
2.3.2 Dealing with objects specificity
The second kind of heterogeneity comes from the differences
between studied objects, for instance between amphora and
parts of amphora or even between amphorae themselves. Many
data come from pictures, thus if not enough pictures are
available, some pieces of information can be missing. Another
problem comes from the fact that some objects may be
incomplete (i.e. broken amphorae).
The data model must also provide tools to express variations
between objects from the same class and different level of
details in the description of an object.
2.3.3 Temporal heterogeneity
And last but not least, we have to deal with changes. The
changes can take place at several levels from values to data
structure. For instance, the values can change between
underwater and surface measurements; this kind of changes can
be managed with modern database systems. But in our
application, the classification of data (i.e. relation between
families of objects) and the inner structure of those objects
depend on the knowledge of experts and this knowledge can
change when new measures are done. The data model must
provide an easy way to change the schema of data and
application build on top of it must be able to deal with change.
2.3.4 Using XML for modelling photogrammetric Data
For all the reasons presented in the previews sections we have
chosen XML as our data model. XML is de facto the standard
for federation and exchange of data between heterogeneous
applications. We use SVG (SVG, 2001) to generate a view on
the result and to visualize the data fusion operation. This will be
described in section 4 of this paper.
3. AUTOMATIC GENERATION OF 3D MEASURES
Whereas in aerial photogrammetry automatic generation of
DSM is becoming mastered, it remains a research topic in close
range photogrammetry because of the greater complexity of the
scenes. A relevant result can be obtained provided that a pre-
existing network of 3D-measures is to be densified. With the
aim of measuring new points automatically, two multi-images
approaches have been explored: on the one hand the area-based
matching and on the other one the feature-based matching.
Five different steps follow the previous principles considering
simple algorithms based on one approach to algorithms that
combine both. All these methods have been implemented in
ARPENTEUR and complete the I-MAGE module.
3.1 The first phase: Roma, 3-D Automatic Measurement
Principles
ROMA, Representation of Oriented Model for Arpenteur, is the
first tool built on the I-MAGE method developed in the
framework of the Arpenteur Project (Drap, Grussenmeyer,
Gaillard, 2001). Roma allows automatic measurement using a
set of oriented photographs and a mesh visible on these
photographs.
In
Th
res
of
tri
tri
rel
pn
As
gri
lo
Or
im
the
ho
Se:
lin
Col
be
Me
ref
