International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
in the mobile phone market which were pushing the
development of automated methods rather hard and many pilot
projects were performed, but with some few exceptions the
major parts were performed with classical stereo measurements.
We can see applications of semi-automatic methods in the
private sector focusing on small building projects in central city
areas for planning purposes or on larger downtown areas for
animation purposes, but very few cities have so far gathered
their own 3D city models for planning purposes. It is often
bound to the mayor or a certain city department to push this
development. In this regard it is a difficult business rather than
a common trend on the geoinformation market.
Some state authorities focus on parcel and road extraction in
high resolution satellite data or on building extraction in aerial
imagery and are using building models as control structures for
automated image orientation. The good news is that automated
feature extraction is increasingly used in education on
university level.
2.2 Commercial systems and research
Most of the approaches presented in the first "high" time of
research on feature extraction from Aerial and Satellite data and
Laser scanning DSMs (e.g. Grün et. al. 2001) have not reached
a practical level, but have disappeared completely due to the
end of a research project, change of staff or interest.
Other reasons were the often not very practical requirements of
some research prototypes, like e.g. 6-8 overlapping excellent
photographs in colour. This is simply not realistic in practice
and these huge extra costs were never really justified by
overwhelming results and empirical proof. Another reason was
in most cases the lack of reliability. Algorithms which fail
when changing to a new set of real input data are not useful for
cartographic feature extraction at all. There was quite late an
insight of Computer Vision people, that full automation does
simply at the moment not work for real world applications, so
they changed to semi-automatic approaches, but adapted very
quickly and showed very promising results. Another
observation is still the lack of editing tools in the so-called
automatic approaches to finalize the job, when automation did
not succeed.
LESSON 2: Full automation of cartographic feature
extraction is not yet possible. Semi-automatic systems
assisted by an operator seem to be the best solution for the
near future. Editing capabilities of results are an
indispensable requirement.
2.3 Modelling 3D worlds — Standards?
Other major obstacles encountered were the problems of
missing standards for 3D modelling in GIS and CAD, but
simply also the lack of standards for the orientation data of
input images.
The standardization in this field is something which must be
urgently attacked by international organizations and vendors. It
is only economically acceptable if there a handful of standards
for image orientation data to be addressed by the vendors. In
this respect EuroSDR has taken an initiative and opened a
project on InterOCI - Interoperability for Orientation and
Calibration data of Photogrammetric Images (EuroSDR, 2004).
Concerning the output the implemented GML functionality is
regarded as the most feasible solution in the future as it allows
the storage of geometry, topology and thematic features in a
common framework, which is not possible by DXF and other
CAD formats. Especially for the integration into a distributed
424
web architecture GML is the preferred exchange standard for
services like the Web Feature Service (WFS) of the Open GIS
Consortium (OGC). In this context a Special Interest Group 3D
of the GDI in Northrhine-Westphalia (SIG3D, 2004) is
elaborating on 3D standards and visualization, which is
increasingly asked for by many users. Detailed geometrical-
topological base models and LoD definitions have been
elaborated and are tested in 3D pilots. An interesting extension
is the automated linkage with terrestrial sensors (cameras and
laser), which should be one of the major research issues for the
forthcoming years.
2.4 Automated modules for feature extraction in digital
aerial imagery
inJECT (Gülch et al. 2000, Gülch and Müller, 2001) was
originally designed for semi-automatic feature extraction in
aerial imagery. It has been substantially improved not only to
be able to support parametric 3D building models, but also
polyhedral objects. Figure 1 shows the Senaatti test site of
EuroSDR's project on building extraction. Two students have
derived the complex buildings in 3D using a pre-release version
of inJECT 1.9. The result is shown in Figure 2 as a textured
VRML model.
inJECT has been lately enhanced by automated modules for
linear and area features specifically in orthophotos and
geocoded satellite imagery. They are described in more detail
in (Ohlhof et al., 2004). In addition to this the introduction of
attributes is now available. inJECT is used by companies and
administrative authorities as well as in academia for teaching
purposes.
RB
EuroSDR's building
ES ERES bier E : CE
Figure |. Senaatti test site of
extraction project.
Concerning area features an approach has been implemented in
a special version to automatically derive the outline of areas in
orthophotos by giving only a start polygon (e.g. triangle) inside
such an area. The approach is a combination of deformable
models and region growing techniques in a statistical
framework under the Minimum Description Cost environment.
This approach is called region competition. In a final
processing step the parcel contours are smoothed. An example
of a start situation, the result of region growing and a final
generalization of the contour is given in Figure 3. Other parcels
are shown in Figure 4.
The tool for the measurement of linear features (in particular
road networks) is based on a line tracking algorithm, where the
user first defines a starting point and the measurement direction
in the image. After that the procedure starts and measures
automatically points along the middle axis of the particular line.
Interne
Existir
genera
is very
The us
inform
The li
segmei
topolo;
operate
Án ex
orthoph
vector |
import
contain
read by
The fea
digital
describ
capture
automa
