ISPRS
2000
be overcome in future, others are inherently related to
the difficulty of accurately modelling the sensor’s inte-
rior orientation. The conclusion from papers related to
this subject and presented at the Congress are similar:
direct orientation is not (yet) reliable or, in more chal-
lenging cases, accurate enough. Thus, a combined
adjustment is recommended.
- New applications, most notably city modelling,
increase the need for generating DEMs and DTMs
quickly and economically. Airborne laser ranging has
become a very viable option to traditional photogram-
metry methods. The workshop in LaJolla demon-
strated the potential of laser altimetry, but also the
need for standardising calibration procedures. Certain
applications, for example extracting man-made
objects, require the combination of laser ranging with
imaging methods. This is another example of a chal-
lenging fusion problem; increased research activities
are expected in this area, as well as in developing suit-
able algorithms for processing laser datasets (thinning,
merging, segmenting, etc.).
- The majority of research related to Commission Ill is in
the area of object recognition and image understanding.
Since these are hard problems, progress is incremental.
The workshop in Munich reflected the state-of-the-art.
There is an increasing trend towards the utilisation of
multispectral data for the recognition process. However,
most approaches are still ad-hoc. More theoretical inves-
tigations into object recognition with multispectral and
multisensor data are required. As an example, models of
real world objects must include spectral aspects to the
extent that they are recorded by multispectral/hyper-
spectral sensors.
- Grouping is essential because feature extraction alone
cannot be expected to result directly in parts of objects.
Particularly, three trends can be noticed: First, grouping
uses more and more attributes, such as the strength of
the gradient or colour values of the features themselves,
as well as of adjacent features. The second tendency is
that grouping is done in three-dimensional object space
using photogrammetric camera models and constraints
on two or more images. Third, there is a trend towards
interleafing grouping and matching processes.
- Almost all object recognition systems developed so far
contain a problem-specific control structure. Therefore,
the adoption of these systems to, even slightly, changed
conditions or new applications remains very difficult. As
a consequence of this research should focus more on the
general strategies instead of solving specific problems.
- There is a definite trend towards real incorporation of
interaction due to the so far limited success rates of so-
called “fully automatic” methods; this holds for object
recognition in images and digital surface models. This
will have the consequence that in coming years more
research results will become available for practical
applications.
- Newest investigations concern the quality and effi-
ciency of image understanding algorithms, and results
on building and road extraction were presented.
- The use of multispectral/multisensor data increases
the potential for solving the object recognition and
scene classification problem more effectively. How-
ever, exactly how to solve this difficult multi-stage
fusion problem is not clear and only little progress has
been achieved. It remains a major research topic.
- A sensor combination that delivers independent infor-
mation about the object space comprises laser rang-
ing, imaging, and multispectral/hyperspectral systems.
It is even conceivable to have this combination on one
platform.
- Traditional object recognition approaches must be
extended to include classification techniques that have
been successfully used in remote sensing for many
years. Another “burning” issue is object modelling: it
ought to include information that the new sensors deliver,
such as spectral aspects of objects.
- Fusion becomes increasingly important and must be
addressed on different levels. The trend of using several
sensors on the same platform requires establishing a
common reference system for the sensors (fusion on the
physical (sensor) level). Similarly, data obtained by dif-
ferent sensors, perhaps not on the same platform, must
be merged (fusion on the data level). Not all multiple sen-
sor data can be merged on that level, however; it may be
necessary to extract features independently and merge
them on the feature level.
- The test site Ocean City has been established by WGIII/5
during the 4 year period. Aerial photography (b/w), multi-
spectral and hyperspectral data, and several laser altime-
ter flights are available (see respective reports).
Technical Commission IV: Spatial Information Systems
and Digital
by Commission IV President:
Mapping
Dieter Fritsch (Germany)
and Commission IV Secretaries: Monika Sester (Germany)
Markus Englich
Report of Outgoing President
In the year 2000 we can review the position of the Commis-
sion in the light of the changes introduced after the Con-
gress in Vienna. It was decided during the Vienna Congress
that in future this commission should be a "homebase" for
Geographical Information Systems, in particular dealing
-
œ
©
with fundamental theoretical developments, operational
aspects and GIS applications. The review today clearly con-
firms that this move was very. Co-operation has started with
the International Cartographers Association (ICA) and with
the Spatial Data Handling Expert’s Group of IGU - the
incoming meeting of SDH in 2001 will be a Joint Meeting
International Archives of Photogrammerty and Remote Sensing. Vol. XXXIII, Part A. Amsterdam 2000.