nbul 2004
omatische
sätzen am
ıland. In:
erseminar,
Imaging.
itm
3. Object-
| Planning
of the
formation
9, pp. 50-
). Textural
7 Systems,
r 1973, pp.
yads from
formation,
K. 2003.
in Städten
ldem. In:
ngewandie
fräge zum
nn Verlag,
an Object-
med Area
ng Landsat
esearch &
02. p. 79
undeswehr
‘many, and
rocurement
Support.
MODELLING THE EXTRACTION OF FIELD BOUNDARIES
AND WIND EROSION OBSTACLES FROM AERIAL IMAGERY
M. Butenuth
Institute of Photogrammetry and Geoln formation, University of Hannover
Nienburger Str.1, D-30167 Hannover, Germany - butenuth@ipi.uni-hannover.de
Theme Session 11
KEY WORDS: Modelling, Integration, GIS, Image, Analysis, Vegetation, Extraction
ABSTRACT:
In this paper work on image analysis methods extracting field boundaries and wind erosion obstacles from aerial imagery is
presented. Describing the objects of interest and additional GIS-dat
overview of the numerous relations between the different objects
à together in an integrated semantic model is essential to get an
and how to exploit the prior knowledge. The strategy is derived
from the modelled characteristics taking into account an automatic processing flow. The field boundaries and wind erosion obstacles
are first extracted separately: A segmentation within selected regions of interest in the imagery leads to field areas, which are split, if
necessary, to preliminary fields. Furthermore, a snake algorithm is initialized to correct the geometric inaccuracies in some parts
yielding final field boundaries. Wind erosion obstacles are derived using DSM-data in addition to the imagery to verify search areas
from the prior GIS knowledge, for example parallel and nearby roads, or to extract wind erosion obstacles without prior information
about their location. Finally, a combined evaluation of the different objects is accomplished to exploit the modelled geometrical
similarities resulting in a refined and integrated solution. Results of the different steps prove the potential of the proposed solution.
I. INTRODUCTION AND MOTIVATION
In this paper work on image analysis methods extracting field
boundaries and wind erosion obstacles from aerial imagery is
described. This data is needed for various applications, such as
the derivation of potential wind erosion risk fields for geo-
scientific questions, which can be generated with additional
input information about the prevailing wind direction and soil
parameters, as described in (Thiermann et al. 2002). Another
arca is the agricultural sector, where information about field
geometry is important for tasks concerning precision farming or
the monitoring of subsidies (Anderson et al. 1999, Grenzdórffer
2002).
In the past, several investigations have been carried out
regarding the automatic extraction of man-made objects such as
buildings or roads, see for example (Baltsavias et al. 2001) and
(Mayer 1998). Similarly, investigations regarding the extraction
of trees have been accomplished, sec (Hill and Leckie 1999) for
àn overview of approaches suitable for woodland and (Straub
2003) for a method to extract trees not only capable for
woodland but also in the open landscape. It has to be
investigated, to what extent these approaches are usable for
extracting wind erosion obstacles such as hedges or tree rows,
and also which enhancements arc necessary to realize the
proposed strategy here.
In contrary, research with respect to the extraction of field
boundaries from high resolution imagery is still not in an
advanced phase: (Lócherbach 1998) presented an approach to
update and refine topologically correct field boundaries by a
fusion of raster-images and vector-map data. Focusing on the
reconstruction of the geometry and features of the land-use
units, the acquisition of new field boundaries is not discussed.
In (Torre and Radeva 2000) a so called region competition
approach is described, which extracts field boundaries from
acrial images with a combination of region growing techniques
and snakes. To initialize the process, seed regions have to be
defined manually, which is a time and cost-intensive procedure.
In (Aplin and Atkinson 2004) a technique for predicting
missing field boundaries from satellite images is presented,
using a comparison of modal land cover and local variance. The
approach involves manual post processing, because only fields
with a high likelihood of missing boundaries are identified, not
field boundaries directly. The aim of the solution, presented in
this paper, is a fully automatic extraction of field boundaries
from high resolution aerial CIR-imagery. Consequently, the
proposed strategy differs from the mentioned approaches. In
addition, relationships between the objects of interest — field
boundaries and wind erosion obstacles — will be exploited to
improve the results.
In general, the recognition of objects with the help of image
analysis methods starts frequently with a modelling of the
objects of interest and the surrounding scene. Furthermore,
exploiting the context relations between different objects leads
to a more overall and holistic description, see for example
(Baumgartner et al. 1997) and (Butenuth et al. 2003). The use
of prior knowledge (e.g. GIS-data) supporting object extraction
can lead to better results as shown in (Baltsavias 2004) and
(Bordes et al. 1996). These aspects are incorporated modelling
the extraction of field boundaries and wind erosion obstacles
and are reflected in the derived integrated strategy.
Initially, the integration of vector and raster data in one
semantic model is briefly described in the next section to obtain
an overview of the numerous relations between the objects to be
extracted and the prior knowledge. Afterwards, the strategy and
approach to extract field boundaries and wind erosion obstacles
is explained, followed by results to demonstrate the potential of
the proposed solution. Finally, further work required is
discussed in the conclusions.
1065
