or DTM
[e DTM;
"asoning.
inference
for point
n Overall
This first
in fuzzy
| surface
ork.
dings of
or height
Iamburg,
ners and
immetric
-hniques.
iction on
Jochen Schiewe
IMPROVING THE INTEGRATION OF DIGITAL SURFACE MODELS
Jochen Schiewe
University of Vechta, Germany
Institute for Environmental Sciences
jochen.schiewe @uni-vechta.de
KEY WORDS: DTM/DEM/DSM, Integration, Quality control, Formalization, Image matching, Object extraction
ABSTRACT:
Mutual benefits can be expected from an integrated processing of elevation and image data in order to improve the
quality of automatically derived valued-added data. On this background new and modified algorithms will be described
and empirically tested for the following closely related important tasks: The generation of Digital Terrain Models
(DTMs) from Digital Surface Models (DSMs), the detection of buildings and wooded regions in the course of an object
extraction, and the identification and revision of blunders within elevation models.
1 MOTIVATION
The user's acceptance of remotely sensed data mainly depends on the corresponding availability, quality and costs. The
current status can be characterised in such a way that tremendous efforts have been made to deliver more data in shorter
üimes by using new sensors and the benefits of the internet, while on the other hand a couple of important automatical
data processing methods are neither reliable enough nor operational yet.
Additionally or complementary to traditional data acquisition methods electro-optical space sensors with improved
spatial resolutions and along-track stereoscopic properties (e.g., with the Ikonos system), digital airborne scanners (e. f
the announced systems of Leica or Z/I Imaging), radar-interferometric sensors (e.g., the Shuttle Radar Topographic
Mission) or laser scanners (e.g., the operational systems TopoSys or ALTM) became available in the last few years or
will be on the market very soon. It is well known that the mentioned data sources are showing advantages and disad-
vantages which are in some cases complementary to the features of other sensors. For instance, laser scanning methods
produce "blind data", i.e. no. semantical or image information is associated with the elevation values - in contrast to
optical sensors, whereas laser scanners are much better suited for capturing and processing heights in wooded or urban
areas. Consequently, Ackermann (1999) forsees multi-sensor-systems that will represent a new development stage in
the field of photogrammetry and remote sensing.
On the other hand, important tasks within the data processing chain like the automatical derivation of elevation models
(e.g., by stereo matching) or the extraction of objects from imagery very often do not lead to satisfying and reliable
results as obtained with human operators. Considering the above mentioned integration of multiple data sources new
chances and challenges can be identified.
In this context, this paper wants to contribute to an improvement of the integration of image and elevation data — with
the emphasis on the latter one, by presenting new and modified methodologies for key tasks. After a general integration
concept is outlined (chapter 2), we will demonstrate and prove ideas and solutions for the generation of Digital Terrain
Models (DTMs) from Digital Surface Models (DSMs), for the extraction of objects (with the emphasis on separating
buildings from wooded areas), and for the detection and revision of height blunders (chapter 3). Chapter 4 will evaluate
the achieved improvements and will give hints to future research and development needs.
2 CONCEPTUAL MODEL OF DSM INTEGRATION
Despite of the lack of actual multi-sensor-systems - with a very few exceptions for testing purposes (e.g., Hug, 1997),
an integration of multi-sensor data from different sources and dates can already be performed. Following and general-
izing the cited idea of Ackermann (1999), our definition of a multi-sensor integration including height information
means
* either the fusion of DSMs of equal or different sources with equal or different spatial and/or temporal properties,
* erthe fusion of DSMs with image data (especially in the visible and infrared electromagnetic spectrum).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 807