33. Istanbul 2004
, blue- terrain,
yw hel LAN
Class. rate
terrain
class. rate
ea Salem
vegetation
Overall
class. rate
| Karlsruhe
erence operators
r for subsequent
defined features
hem 9 different
the influence of
1e independence
ymbinations and
es the individual
n rate has been
gnificant border
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
gradients which should separate terrain objects has evidently no
influence on the results in test site Karlsruhe. Comparing
first/last pulse differences and height texture which both
contribute to discriminate buildings and vegetation, it is obvious
that height texture is of less importance because the averaged
improvement of classification rate is only about 1% to 3%. For
first/last pulse differences this value is about 7% to 10%.
Adding the shape parameter to the feature combination only at
test site Karlsruhe a slight improvement of the results (about
2%) can be observed due to the higher amount of larger
buildings compared to rural region of Salem. The intensity
values — only available for test site Salem — contribute
significantly to the classification success. An increase of about
7% was achieved.
34 Maximum-likelihood classification
Besides the fuzzy logic approach with different inference
operators also a statistical classification method has been
applied to be able to compare the fuzzy logic results with a well
proven standard approach and to discuss the differences. A
maximum likelihood classification was chosen for this purpose.
To obtain reasonable results exactly the same training and
control objects has been used in this classification.
The results for both test sites Karlsruhe and Salem - based on
the combination of all parameters - are assembled in Table 5.
For reasons of comparison also the main classification rates of
fuzzy logic are included in this table.
Test site Class. | Class.rate | Class. | Overall
rate vegetation | rate class.
buildings terrain rate
Fuzzy | Salem 95 96 93 95
logic | Karlsruhe | __89 90 - 90
Max.- | Salem 96 96 93 95
lik. | Karlsruhe 92 86 - 89
Tab. 5 Comparison of main classification rates between fuzzy
logic and maximum-likelihood method
lt is obvious that classification rate of vegetation in test site
Karlsruhe is higher for fuzzy logic than for maximum
likelihood but contrary for building while the total classification
rate is the same. These differences are caused by the influence
of the definition of membership functions in the fuzzy logic
approach. Even a modification of the related membership
functions in order to increase the classification rate of buildings
would inevitably lead to an accordant decrease of classification
rate for vegetation, so the resulting overall classification rate
would remain nearly the same. The results of both methods are
in the same dimension if all available features are used. If
combinations of only a few features are applied no definite
assessment can be made. For test site Karlsruhe fuzzy logic
Seems to provide better results while it is a contrary situation
for Salem. The advantage of fuzzy logic may be that the
transferability to other locations seems to be easier especially
for applications where only a few training areas/objects are
available due to its robust membership functions.
4. CONCLUSION
Using a priori knowledge about the characteristics of 3D objects
In laserscanning data for definition and extraction of object-
413
relevant features suitable results can be achieved using fuzzy
logic or maximum likelihood classification. An improvement
may be possible by introducing a hierarchical classification
scheme based on a set of rules. Such a logical decision structure
will be implemented in the next phase of this project to
overcome some disadvantages of standard inference operators
like they were used in these investigations. Additionally a post-
segmentation has to be integrated in this approach to separate
different object types which are erroneously combined to one
segment, e.g. vegetation objects which are located directly
beside a building.
REFERENCES
Definiens, 2001. www .definiens.de
Douglas, D., Peucker, T., 1973. Algorithms for the reduction of
the number of points required for represent a digitized line or its
caricature. Canadian Cartographer, 10(2), pp. 112-122.
von Hansen, W. & Voegtle, T., 1999. Extraktion der
Geländeoberfläche aus flugzeuggetragenen Laserscanner-
Aufnahmen. PFG, Nr. 4/1999, pp. 229-236.
Hofmann, A. D., Maas, H.-G., Streilein, A., 2002 Knowledge-
Based Building Detection Based on Laser Scanner Data and
Topographic Map Information. ISPRS Comission III, Vol.34,
Part 34 “Photogrammetric Computer Vision”, Graz, Austria,
A169-174
Lohmann, P.: Segmentation and Filtering of Laser Scanner
Digital Surface Models, Proc. of ISPRS Commission ll
Symposium on Integrated Systems for Spatial Data Production,
Custodian and Decision Support, IAPRS, Volume XXXIV, Part
2, pp. 311-315, Xi'an, Aug. 22-23, 2002
Lohr, U., 1999, High resolution laser scanning, not only for 3D-
city models. Fritsch, D. and Spiller, R.: Photogrammetric Week
’99, Wichmann, Karlsruhe, Germany
Maas, H.-G., 1999. The potential of height texture measures for
the segmentation of airborne laserscanner data. In: Fourth
International Airborne Remote Sensing Conference and
Exhibition / 21st Canadian Symposium on Remote Sensing,
Ottawa, Ontario, Canada.
Schiewe, J., 2001. Ein regionen-basiertes Verfahren zur
Extraktion der Geländeoberfläche aus Digitalen Oberflächen-
Modellen. PFG, Nr. 2/2001, pp. 81-90.
Steinle, E. & Voegtle, T., 2001. Automated extraction and
reconstruction of buildings in laserscanning data for disaster
management. In: Automatic Extraction of Man-Made Objects
from Aerial and Space Images (III), E. Baltsavias et al. (eds.),
Swets & Zeitlinger, Lisse, The Netherlands, pp. 309-318.
Till, T. 1993.Mustererkennung mit Fuzzy-Logik. Franzis-
Verlag GmbH, München
Voegtle, T., Steinle, E., 2003. On the quality of obeject
classification and automated building modelling based on
laserscanning data. The International Archives of
Photogrammetry, Remote Sensing and Spatial Information
Sciences, Dresden, Germany Vol. XXXIV, Part 3/W13, 8-10
October 2003, ISSN 1682-1750
