ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
(EIRE TES INT
top-down operator |
e |
external operators
bottom-up operator
result
system control
mbole destartion ot scene
symbolic description of scene |
Figure 2: Components of GEOAIDA
restrictions of the concepts are checked and hypothesis,
which are not conform are deleted.
The depicted process chain is shown in figure 2. On the
left side in figure 2 the input data is shown, consisting of an
image database and a semantic network. In the middle part
of figure 2 the scene interpretation is illustrated: The top-
down-step generates a hypothesis network, shown in the
middle, the bottom-up-step generates an instance network
(see on right side). Under the three different networks the
corresponding regions in the image are illustrated in three
different levels of detail. On the right side in figure 2 the
result of the scene interpretation is illustrated.
2.1.2 Bottom-up-operator The
has the following tasks:
bottom-up-operator
e Extraction of object attributes and measurement of
single objects.
e Grouping of objects with guidelines of the user.
e Adaption of the label images to the new obtained in-
terpretation.
e Measurement of the new group, generated by the
bottom-up-step.
If the top-down-analysis reaches the leaf nodes, the anal-
ysis turns from model based interpretation to data based
interpretation (bottom-up). — The bottom-up-operators
can also be extern programs, designed by the user of
GEOAIDA. The user can also specify his special tasks in
a functional description language. This sequence is shown
in the right part of the scene interpretation in figure 2
corresponding to the top-down analysis. The top-down-
path can generate different hypothesis for one region. The
bottom-up-step has to decide for an explicit interpretation
for a region (see figure 2). The uncolored areas illustrate
not classified regions, GEOAIDA can make use of a rejec-
tion class. Same colors in the structural illustration and in
the figurative map denote the same object class.
3 EXAMPLES
Because it is not possible to explain the whole analysis of
a project, some significant steps of the analysis are exem-
plarily demonstrated. The first example shows, how a re-
gion can be classified by means of the extracted objects. In
the following the necessity of a quality factor for concur-
rent interpretation possibilities and the splitting of a region
in different classes is illustrated. For the initial segmenta-
tion of the complete scene streets are used for the region
borders. The streets can be taken from GIS or can be auto-
matically extracted (Aviad ef al., 1988) (Barzohar ef al.,
1993) (Biickner, 1998) (Baumgartner ef al., 1999) from
the image data. In figure 4 a part of the complete scene
is pictured. Figure 5 shows all objects detected during the
scene analysis of one region. In figure 5.a the first step,
the initial segmentation of the scene is shown, the streets
are marked with black lines. For the central region the hy-
pothesis inhabited area (compare figure 3) is examined. In
figure 5.b the visual aerial image is shown. The extracted
objects, like houses, are shown in the following figure 5.c
without image data, each object type has its own label. The
houses in figure 5.c are extracted with use of height data by
an building extraction operator.
The initial hypothesis for the region inhabited area can be
validated based on the existing component parts like the
extracted houses. The interpretation result of the examined
region framed by streets is the class settlement illustrated
in figure 5.d. The classification result of the region was
made with a quality factor, that is calculated of the com-
ponent parts included in the region in the following way: