Ulrich Thoennessen
SEGMENTATION OF INTERFEROMETRIC SAR DATA FOR BUILDING DETECTION
Uwe SOERGEL, Ulrich THOENNESSEN, Hermann GROSS, Uwe STILLA
FGAN-FOM Research Institute for Optronics and Pattern Recognition
D 76275 Ettlingen, Germany
{soe,thoe,gross,usti | @fom.fgan.de
Working Group II/4
KEY WORDS: Visualisation, Interferometry, Data fusion, Segmentation, Buildings.
ABSTRACT
The improved quality of InSAR data suggests to utilise such data for building detection. But, the phase information
from which the height data is calculated, is often severely disturbed, depending on the signal to noise ratio. In this paper
we refer to investigations to stabilise and improve the InSAR height data.
After speckle filtering, a segmentation of the intensity data is carried out. With these segments height data are masked
and an average height is calculated using the related intensity values as weights. In a post-processing step, possible
existing over- and under-segmentation are corrected. Adjacent objects with similar heights are merged and objects
including shadow areas are split.
Different tasks are distinguished for utilisation of the derived 3D information. A test site including the airport of
Frankfurt (Main) was chosen. For the visualisation purpose, the segmentation result is shown. The results are compared
to a vector map and differences are depicted and discussed.
1 INTRODUCTION
Photo interpretation of remotely sensed data can be supported by fusing data from different sensors and knowledge
sources together with an appropriate visualisation of this data. Such a visualisation can be e.g. mapping of imagery on a
3D-model of urban areas. Multi spectral aerial images (Figure 1) can be combined with laser altimeter data.
Additionally, vector maps and laser elevation data may be used to generate prismatic models as a coarse 3D-description
of buildings [Stilla & Jurkiewicz, 1999]. In this paper we focus on improved visualisation of InSAR data and combining
InSAR data and vector maps.
Figure 1: Knowledge Sources: a) Map, b)Optical Image, c) IR Image, d) SAR Image, e) Laser Altimeter Data
The improved quality and ground resolution of InSAR suggests to use this data for supporting the interpretation of man-
made objects. But, in the case of SAR interferometric data many additional problems have to be taken into account, e.g.
radiometric correction, layover effects, multi-backscattering, speckle, phase-unwrapping errors, relations and
dependencies of intensity and noise in the height information [Schreier, 1993] [Bamler et al., 1998].
Even small errors in the phase lead to big errors in the height values. Thus, the phase information has to be prepared for
further processing. Smoothing of the height data, like phase multilook processing [Goldstein et al., 1988], blurs edges of
man-made objects (e.g. building walls) and as a consequence, details may be lost. On the other hand, object borders can
often be extracted in the intensity data because of the different backscattering behaviour of different materials.
Additionally, in many cases areas with comparatively homogeneous intensity distributions correspond to regions,
respectively objects, with same height information. Such regions are extractable by edge-preserving region-oriented
image processing algorithms. In our approach we segment regions with similar intensity values in the speckle-filtered
intensity data using an edge-sensitive region growing algorithm [Levine & Shaheen, 1981]. We are interested in man-
328 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
In
pr
the
int
TI
ph
SA
CO
SZ
va
Wi
c
19
In
ne.
as
Th
If
Th
