USE OF HYPERSPECTRAL AND LASER SCANNING DATA FOR THE CHARACTERIZATION
OF SURFACES IN URBAN AREAS
Dirk LEMP, Uwe WEIDNER
Institute of Photogrammetry and Remote Sensing, University Karlsruhe, Englerstraße 7, 76128 Karlsruhe, Germany
{lemp,weidner}@ipf.uni-karlsruhe.de =
KEY WORDS: Hyper spectral, LIDAR, Reconstruction, Classification, Urban
ABSTRACT *
A recent project of the Engler-Bunte-Institute (EBI), chair of water chemistry, and the Institute of Photogrammetry and
Remote Sensing (IPF) aims at the quantitative assessment of pollutants on urban surfaces by chemical analysis and
image processing methods. The motivation of this project is the fact that nowadays a better part of the rain water from
sealed urban surfaces is treated in sewage plants, although this might not be necessary, because the load of pollutants
of the first flush is much higher than in the following run-off. Therefore, the dimensioning of sewage systems may be
adopted to this observation and costs may be reduced. In the project, the research focus of EBI is the chemical analysis
of washed off pollutants and modelling of the resulting pollution (run-off), whereas the research at IPF deals with the
characterization of urban surfaces, namely their geometry (slope, exposition, size) and their surface material. For this
purpose two different types of data are used: hyperspectral and laser scanning data with 4 and 1 m planimetric resolution
respectively. We combine these data sets of high geometric and spectral resolution to create a detailed map of sealed
urban surfaces. The laser scanning data will not only be used to derive geometric properties of the surfaces, but also to
improve the classification of materials as it helps for the discrimination of roof and ground surface materials with similar
spectra. The paper will present first results of data analysis, which will be focussed on roof surfaces in a first step.
1 INTRODUCTION
In the year 2000, the European Union implemented the
water framework directive. This regulation oblige every
member state to review the impact of human activity on
the status of surface waters and on groundwater. In a recent
research project we focus on a small, but nevertheless im-
portant topic in this context: the assessment of pollutants
on urban surfaces and their impact on the pollution load.
Thus, one aim of the project is not only to derive informa-
tion on the amount of sealed surfaces in an urban area (cf.
(Butz and Fuchs, 2003)), but also to derive a detailed sur-
face material map. Therefore, the work package consists
of five subtopics — chemical measurements for the charac-
terization of the chemical processes on reference roof sur-
faces, determination of surface geometry, classification of
surface materials, modelling of the resultant pollution, and
model verification. In this paper, we describe our work on
two of these subtopics, namely the information derivation
of the surface characteristcs, i.e. geometric and material
properties.
Urban areas are characterized by their complex geomet-
ric structure and their heterogenity concerning the occur-
ing surface materials. The appearance of surface patches’
materials in the data is influenced by the acquisition and
object geometry. Furthermore, the age of the material and
environmental conditions, e.g. by weathering and humid-
ity, also have impact on their appearance. All these facts
lead to the necessity of high resolution input data to solve
the tasks — high resolution with respect to the geomet-
ric resolution, but also to the spectral resolution in order
to discriminate the various surface materials. Therefore,
we combine data derived from laser scanning, which pro-
vides the necessary geometric information, and hyperspec-
1011
tral data for the classification of surface materials.
In the following, we give a short overview on related work
dealing with the combination of laser scanning and hyper-
spectral data. Section 3 introduces the input data. Our ap-
proach for the characterization of surfaces in urban areas is
presented in Section 4 focussing on roof surfaces in a first
step, followed by a summary of recent results in Section 5
and the conclusions.
2 RELATED WORK
Up to now, the two data types were often used exclusively,
either to derive the geometry based on laser scanning data
(cf. (Vogtle and Steinle, 2003)) or to derive material maps
based on hyperspectral data (cf. (Heiden et al., 2001)). The
improvement of reconstruction from laser data by addi-
tional image information is discussed, but mainly to reject
vegetation areas. (Gamba and Houshmand, 2000) use hy-
perspectral data (AVIRIS) in order to improve reconstruc-
tion results based on IFSAR, namely to mask vegetation
areas, but the used data has only limited resolution. (Mad-
hok and Landgrebe, 1999) integrate DSM information in
order to improve the results of hyperspectral classification
based on HYDICE data. In their research the DSM, de-
rived from aerial imagery, is applied for the discrimination
of roofs and ground surfaces. The materials may have a
similar spectrum, but they can be discriminated based on
the height information. (Simental et al., 2003) combine
hyperspectral (HyMap) and laser data to derive a mobility
and trafficability map in an open area, thus the require-
ments seem to be less strong than in our application.
The approach of (Bochow et al., 2003) is the closest related
work to our approach. They use a normalized Digital Sur-
