Full text: Fusion of sensor data, knowledge sources and algorithms for extraction and classification of topographic objects

International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
M. Honikel 
Institute of Geodesy and Photogrammetry, ETH Zürich-Hönggerberg, CH-8093 Zürich, marc.honikel@geod.ethz.ch 
KEYWORDS: DEM, ERS, SPOT, Data fusion. 
Generation of digital elevation models (DEMs) is a main issue of remote sensing, as precise DEMs are needed in a wide range of 
remote sensing applications. The most widespread techniques for DEM generation are stereoscopy for optical sensor images and 
interferometry for SAR images. Both techniques suffer from certain sensor and processing limitations, which can be overcome by the 
synergetic use of both sensors and DEMs respectively. In this paper, different strategies for fusing SAR and optical data are 
combined to derive high quality DEM products. Quality in this context means accuracy of heights. An estimation of the error 
properties of both datasets is derived both from the data processing and the single DEMs themselves. The stereo-optical and In SAR 
techniques have distinct, partially complementary, error properties. Two techniques, which take advantage of the complementary 
properties of InSAR and stereo-optical DEMs, will be applied for the fusion process. First, as cross-correlation plays a major role for 
the height measurement in both techniques, the cross-correlation of phases, respectively grey values, are determined for each DEM 
point and used as weight for the DEM fusion. In the second technique to be applied, by taking advantage of the fact that errors of the 
DEMs are of different nature, affected parts are filtered and replaced by those of the counterpart. The results of both techniques will 
be fused in a final step. This procedure is tested with two sets of SPOT and ERS DEMs of regions in south-central Catalonia, 
resulting in a remarkable improvement in DEM accuracy and representation of the terrain. 
The two main techniques for DEM generation from satellite 
imagery are stereoscopy with optical sensors and interferometry 
with SAR sensors. The height is measured in different ways in 
both techniques. InSAR computes the height from the phase 
difference of the point backscatter in two passes. Since phases 
in an interferogram are wrapped in an 2n interval, the critical 
step in the generation of elevation models with SAR 
interferometry is phase unwrapping, i.e. the solution of these 
phase ambiguities. Phase unwrapping becomes extremely 
difficult in cases of low signal to noise ratio due to 
decorrelation of the phase measurements. In the optical case, the 
coordinate difference of conjugate points in both images must 
be measured for the height determination. Therefore, the correct 
identification (matching) of conjugate points is essential for 
accurate stereo height measurements. Again, decorrelation 
handicaps the matching or even leads to point mismatching. 
As both techniques are based on different principles and 
sensors, the question arises, how these independent height 
measurements can be fused to obtain DEMs, which do not 
suffer from the single sensor limitations and consist only of 
valid measurements of each single source. Obviously, this task 
goes beyond the straightforward (yet not simple) identification 
and exchange of data in regions, where the height measurement 
of one of the contributing sensors fails (e.g. occlusions, 
layover), as it would not take advantage of the independent 
measurements as a whole. An estimation of the height errors is 
needed for any fusion process, in order to profit from both 
sensors. Provided a measure is found, which is directly related 
to the height error, it can be used to define a weight according 
to which the DEMs can be fused. As correlation is a critical 
factor for both techniques, the correlation coefficient of both 
measurements will be introduced as weight for the proposed 
fusion method. Although the correlation coefficient has 
different meaning in both techniques, it still relates to the 
measurement error. 
Like all measurements, the measurement of topographic heights 
is error corrupted for various reasons. According to the error 
model introduced by Baarda (1967), three general types of 
errors have to be distinguished: 
• systematic errors, which occur in all measurements due to 
measurement inconsistencies. 
• random errors, which are inevitable and nonpredictable. 
They can be described as random variables. 
• blunders, which are large and occur due to an erroneous 
handling of the measurement process. 
This error model, originally designed for geodetic networks, 
applies to redundant measurements of the same parameter. As 
we deal here with single measurements with different sensors, 
the model must be modified, as an assumption of redundancy 
would be invalid in this case. 
The systematic error in this context applies to a global error, 
which affects all measurements within a DEM (e.g. baseline

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.