ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
7. a / A
Figure 1: Multi-sensor Input Data (VIS, IR, Laser-scan) -
TopoSys
part is recorded with an infrared sensor and the lower part
of the image are height data, generated of laser-scan data.
Interpretation of remote sensing data means to transform
input data into a structural and figurative description of the
input data, which represents the result of the analysis (see
on right side in figure 2). The structural description gener-
ated from the generic semantic network has the same struc-
ture as the semantic network. This form of result descrip-
tion makes it possible to access information of the object
type, the geo-coordinates and all other attributes calculated
during the analysis. The result and all intermediate results
are stored in XML-descriptions and can be used for other
examinations.
The figurative results are thematic scene maps (see fig-
ure 2) existing for all abstraction levels. The results can
be iterative visualized for the existing abstraction levels.
The detail level of the structural presentment corresponds
with the figurative presentment. The selection of an object
in the figurative representation focuses the corresponding
object in the structural representation, so the attributes of
the object are directly available.
2 SYSTEM DESCRIPTION
In addition to the input data additional knowledge in terms
of a semantic network is provided to the system (see fig-
ure 2). This semantic network, exemplarily shown in fig-
ure 3 contains nodes and edges, whereat nodes represent
the objects and edges represent the relations between the
objects. Nodes of the semantic network are denoted con-
cepts. During the analysis, the generic semantic network
generates an associated network of instances, denoted in-
stance network. In the concept nodes the data necessary for
the analysis and an image processing operator are stored.
inhabited area agriculture
| settiement industry N
N
1
| house | | garden | | building | | parking area | acreage | | meadow | forest |
with holisiic TopDown-Op | P949 | without holistic TopDown-Op
Figure 3: Example for a Semantic Net
The stored data is e.g. the minimal and maximal number of
instances, that is necessary to build a higher instance. The
instance settlement must contain at least two component
parts of type house.
The semantics of the nodes of GEOAIDA is not ex-
act determined, as in other semantic networks (Tónjes ef
al, 1999), but can be particularized in the bottom-up-
operators.
The result of the interpretation is shown schematically on
the right side of figure 2 and contains a structural descrip-
tion of the result and a presentment of thematic maps.
2.1 Analysis
The middle part of figure 2 shows the analysis struc-
ture, which can be distinguished in two analysis steps,
the bottom-up-step and the top-down-step. The top-down-
analysis is model based and generates a network of hy-
pothesis based on the semantic network. The grouping
of hypothesis and the associated verification or falsifica-
tion of the hypothesis is task of the data driven bottom-up-
analysis. Result of the bottom-up-analysis is the instance
network shown in figure2.
2.1.1 Top-down-operator The top-down-operator has
the task to separate a region into subregions and to build
hypothesis for the expected objects. The task is realized
recursive from the upper nodes in the semantic network to
the lower nodes. For this purpose any segmentation op-
erator can be integrated, which creates hypothesis for the
subregions. The advantage of holistic operators are e.g.
splitting a region into subregions by means of a consis-
tency measurement. E.g., the contemplation of texture al-
lows only a few possible hypothesis for the investigated re-
gion, other possible hypothesis contained in the semantic
network are not possible.
For all generated subregions geo-references exist, so the
global reference is always given. The use of holistic oper-
ators enables a limitation of possible hypothesis, whereby
the complexity of the analysis and the calculation time is
reduced. Holistic operators can reduce the class affilia-
tion of a region, without knowledge of the exact compo-
nent parts of the region. During the top-down-analysis the
A - 63