You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Fusion of sensor data, knowledge sources and algorithms for extraction and classification of topographic objects
Baltsavias, Emmanuel P.

International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
M. Hahn ', F. Samadzadegan 2
Department of Surveying and Geoinformatics, University of Applied Sciences Stuttgart, Schellingstr. 24, D-70174 Stuttgart,
Germany, m.hahn.fbv@fht-stuttgart.de
2 Faculty of Engineering, University of Tehran, P.O.Box 11365-4563 Tehran, Iran
KEYWORDS: Wavelet Analysis, DTM Integration, Fusion, Multiscale Processing, Filtering
The number of sensors and techniques for the acquisition of Digital Terrain Models (DTMs) and the variety of applications based on
DTMs have grown significantly during the last decade. As a result of that, DTMs exist with different spatial resolution and accuracy.
For some areas, quite a number of DTMs is available or there is the task to update existing DTMs with new measurements, e.g. of
higher accuracy. For merging given DTMs of a certain area, different possibilities exist, even though from a theoretical point of view
not much attention has been paid to this topic so far. Probably the simplest procedure is to introduce the raster points of different
DTMs jointly as observations into an interpolation process.
In this paper, we present a new concept for the integration of DTMs based on wavelets. Wavelets can contribute to tasks for which
multiscale aspects are essential. The concept considers two or more DTMs which may differ in resolution and accuracy. The three
most important components of our DTM integration approach are the following: by wavelet decomposition and multiscale processing
a DTM and its accuracy can be represented in a series of scales. Merging of two DTMs takes place on the same scale level, e.g. by a
least squares fit. Finally, an integrated high resolution DTM is obtained by a wavelet reconstruction process. First experiments have
been carried out to illustrate the procedure and gain insight into the achieved accuracy improvements of integrated DTMs.
In recent years, wavelets have taken over a leading role in
analysis and synthesis of signals. The wavelet transform is
widely applied, e.g. in image processing for data compression,
data de-noising and feature extraction but also in many other
Wavelet-based approaches show some favourable properties
compared to the Fourier transform. The Fourier transform uses a
complex exponential function to transform a signal into the
frequency domain. The plane waves oscillate infinitely with the
same period and do not allow localisation. Changing a signal in
just one data point results in a change of the whole spectrum of
this signal without a possibility to estimate the location of the
change in the signal from the spectrum. A wavelet can be
interpreted as a "small wave” or a generalised oscillation. In the
wavelet transform, a shifted and dilated wavelet is used to
analyse the signals. Thus, the wavelet transformation can be
interpreted as a convolution or correlation of the signal with
scaled versions of the wavelet. The degree of similarity between
a part of the signal at a certain location and the scaled version of
the wavelet gives the key input for feature extraction. Threshold
strategies for the wavelet coefficients with small absolute values
lead to efficient data compression with high compression rates,
as well as to noise reduction schemes. For examples, please cf.
Thierschmann, et al., 1997 and Pan and Schaffrin, 1999.
Another important issue for the integration of DTMs are the
conceptual developments in the area of data integration or data
fusion. It is helpful to have an understanding of the methods
and goals of integration or fusion. The following table is a
shortened excerpt taken from Abidi and Gonzalez (1992):
Feature level
estimation /
logical /
of expected
increased feature
increase in
truth or in
increase quality,
value of addi-
tional features
Table 1. Data fusion (adopted from Abidi and Gonzalez,
1992, p. 37).
On the feature level, the goal of fusion is to increase the feature
measurement with regard to quantity or diversity. The symbol
level pursues decision-related goals by aiming at an
improvement of probabilistic decisions based on inference
techniques. The situation on the signal level can be
characterised by a wide range of estimation techniques which
typically aim at improving the accuracy or more generally the
quality of a state.
The integration of DTMs falls under the category of fusion on
the signal level. Our integration method consists of multiscale
signal processing based on wavelets. Localisation provides the
theoretical background that DTM information of multiple scales
is combined according to its local spectral property. The goal is