Jochen Schiewe
In principle the integration of data can take place at all processing levels and it can be used for a variety of purposes
for example for |
automatically separating Digital Terrain Models (DTMs) from Digital Surface Models (DSMs);
improving the automatical extraction of objects from imagery;
detecting blunders resp. improving the reliability of elevation data;
substituting missing height information (e.g., in regions of clouds or low sampling density);
improving the extraction of morphological lines and points;
generating virtual (static or dynamic) landscapes at different scales (Fritsch, 1999).
Of course, new problems are also arising with the integration of different data sets. For example, contradictory height
or attribute information have to be handled. Hence, a key issue in DSM integration will be on blunder detection resp,
error handling.
Following the idea of this conceptual model, it will be necessary to develop algorithms for the specific purposes in
order to demonstrate or to disprove the desired profits from integrating elevation and image data. In the following we
will concentrate on the fusion of DSMs with imagery by dealing with the first three of the mentioned topics which are
actually closely linked to each other.
3 INTEGRATION TOPICS
In the following three closely related key topics with the integration of digital elevation and image data will be ad-
dressed: The derivation of the terrain surface from the DSM without additional information (section 3.2), which is for
instance mandatory for the extraction of objects that stand out against their surrounding (like buildings or wooded areas;
section 3.3). Vice versa these object information are also valuable for a correction of the approximated DTM as well as
for the detection of blunders within the elevation model (section 3.4). Previously, section 3.1 describes the used data
sets.
3.1 Test data
We will mainly rely on a data set covering parts of the City of Osnabrück (Germany) which was captured with the first
version of the digital air-borne scanner HRSC-A (Scholten et.al., 1999). Because this scanner was originally designed
for the use on the planet Mars, the spectral properties of the imagery especially in the red and infrared bands are not
optimal for earth observation purposes (figure 1). In our case the spatial resolution of the scanner leads to a ground pixel
size of 0.16 m for the panchromatic channel. The radiometric resolution of all channels amounts to 8 bit. In order to
evaluate the intended object extraction a reference has been generated through visual on-screen digitizing.
E | green | red | | infeared |
Landsat TM 7
[ pan
HRSCA blue | green | red | | ins |
{ : (um)
400 500 600 700 800 900 1000 1100
Figure 1. Design of spectral bands of HRSC-A scanner compared to those of Landsat TM.
Due to the nadir and the oblique looking (stereo) channels of the HRSC-A it is possible to acquire along-track stereo
scopic height data. A gridded Digital Surface Model (DSM) with a horizontal spacing of 0.5 m has been obtained b)
automatical matching using ISTAR's SPOT3D software (Renouard and Lehmann, 1999). The DSM represents heights
between 67 m and 100 m with an estimated accuracy of 0.1 m to 0.3 m (Scholten et.al., 1999).
In addition, also height data derived by automatical matching of imagery from the French satellite system SPOT cover
ing the Westbank (Palestine) and Israel will be used.
808 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
—7
32
[n or
tween
tional
accur:
3.2.1
1996)
mathe
value:
- also
above
Obvic
too hi
(see S
hand
(ie. t
322
follow
DSM
(resp.
Wsorr
height:
at ever
DTM ;
Figure
deed n
showir