004
98).
Ray
‘arth
1 the
nced
tion
o of
| for
idian
-224
«ture
ig of
'nsor
tical
mote
P.M.
iolite
x ot
ssive
mote
vious
lysis.
493-
992).
scape
apper
JPL
ation
FUSION AND PERCEPTUAL ORGANISATION OF FEATURES
FROM MULTI-SENSOR DATA: GENERAL CONCEPTS AND NEW DEVELOPMENTS
J. Schiewe
University of Vechta, Research Center for Geoinformatics and Remote Sensing (FZG), PO Box 1553, 49364 Vechta
*
Germany — jschiewe@fzg.uni-vechta.de
Commission IV, WG IV/7
KEY WORDS: fusion, interpretation, recognition, feature, laser scanning, multi-sensor
ABSTRACT:
While the available, eventually multi-source data from remote sensing systems yield the advantage of producing more features, the
crucial point in the scene interpretation workflow is still the big, application dependent gap between these features and the related
object characteristics. The necessary bridging process of perceptual organisation as well as the fusion of features have been
neglected .too much in the past. In this contribution we will present our methodological developments concentrating on the
perceptual organisation of features that have been derived from multi-sensor data aiming at the extraction and classification of
topographic surface edges. The algorithm we are proposing consists of a segmentation and classification of the raster-based
elevation and image data, transforming the resulting polygon outlines into vector format and performing some post-processing
steps (dilation, smoothing).
1. INTRODUCTION
The various and heterogeneous user demands concerning
properties of remotely sensed data have led to the development
of a couple of new sensors with advanced spatial, spectral,
radiometric or temporal characteristics. Furthermore multi-
sensor systems have been designed for the simultaneous
acquisition of image and elevation data, in particular by
electro-optical cameras and laser scanning systems. Section 2
gives an overview of such multi-sensor data sources in general,
and the FALCON system, which will be used in this study, in
particular.
However, despite these technical developments the user's
acceptance for remote ' sensing products has not been
significantly increased. This is mainly due to the fact that
respective automatic processing methods — in particular for
interpretation purposes — are not operational or have not
revealed an additional value of the advanced data sources yet.
To be more specific: While the available, eventually multi-
sensor data yield the advantage of producing more scene
features, the crucial point in the interpretation workflow is still
the big, application dependent gap between these features and
the related object characteristics (Schenk, 2003). The
necessary bridging process of perceptual organisation has been
neglected too much in the past. Furthermore, a fusion of
features coming from different sensors has hardly been applied
within the perception step. Section 3 will elaborate on the
aspects of perceptual organisation and fusion in more detail
from a theoretical point of view.
In the context of these deficiencies we will demonstrate in the
second, practical part of this contribution our methodological
developments concentrating on a fusion of features that have
been derived from multi-sensor data. As an example we will
605
focus on the extraction of topographic surface edges
(commonly known as breaklines). The task from a
methodological point of view consists not only in delineating
the edges as such but also in their classification (e.g. walls,
embankments). The algorithm we are proposing consists of a
segmentation and classification of both, the raster-based
elevation and image data, transforming the resulting polygon
outlines into vector format and performing some post-
processing steps (dilation through matching with image edges
and smoothing). This hybrid feature level fusion process as
well as first qualitative results will be presented in section 4.
2. MULTI-SENSOR DATA SOURCES
2.1 General remarks
Due to the enhanced performance of GPS/IMU (Inertial
Measurement Units) solutions for capturing position and
orientation data of the associated moving platforms airborne
laser scanning systems have reached maturity in the last ten
years. Today standard systems capture both, multiple
reflections (first and last echo data) and intensities of the
reflections (generally, in the near infrared portion of the
spectrum).
Enabling furthermore the simultaneous acquisition of multi-
spectral image data, one yields high potential multi-sensor
systems that on one hand deliver more accurate and reliable
elevation data compared to stereo image matching solutions,
and on the other hand by-pass the disadvantage of the “blind”
laser scanning information. The presently most prominent
examples of these multi-sensor airborne systems are FALCON
(TopoSys company), which will be used within this study (see
section 2.2), the laser scanner ALS 50 in combination with the
