anbul 2004
rate on the
comparison
up-to-date
npleteness,
ain interest
] that are
reas. Other
if they are
the ATKIS
1 reality by
we look at
the quality
n the chain
Airborne or
S can be
e reference
1racy, too.
rmine the
es like the
| attributes
ses quality
Yformation
*
Spatial
as been a
| computer
Heipke et
18 to enter
articularly
ion areas.
wledge in
easily be
tic feature
has to be
isions to a
flow, the
id reliable
(trate on
er (2004),
perator in
TKIS by
satellite
g them to
formation
9 training
howing a
lasses are
lized for
ysis. GIS
of prior
' stabilize
/stem for
proven to
ge about
cess. Our
networks
d to land
“Ms, and
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
In our system the comparison of the ATKIS DLMBasis and
reality, also called update, comprises two steps, namely
verification and the acquisition of change. Verification is
characterised by the following features:
e The image analysis process is guided by information from
a GIS about the object to be verified, i.e. the algorithm
makes use of the information stored in the GIS to detect
the image object.
e If there is a certain degree of consistency of image features
and information from the GIS, the object is accepted.
e Otherwise the object is labelled as not accepted.
Thus, verification is suited to determine specific quality
measures. For updating, in contrast, information about a new
object not yet stored in the dataset or information about changes
of the old object have to be extracted from the image, too.
Hence, a reasonable process chain starts with a verification step.
In the event of any inconsistency other feature extraction
algorithms can be triggered to derive more detailed information
that is of value for the following update process.
2. IDEAS BEHIND THE SYSTEM
2.1 Concept of the Prototype
The concept of the research and development project of BKG
and the University of Hannover is characterized by the
following main ideas:
* Transfer of knowledge-based
techniques to an operational
applications (cf. Sec. 1.3).
* Development of a prototype for comparing the model
ATKIS DLMBasis to reality given in form of digital
orthoimages.
* [Efficient integration of the prototype into an interactive
workflow.
interpretation
practical
image
solution for
The system development is embedded in a broader concept of a
knowledge-based workstation, which provides functionality
from photogrammetry, GIS, and cartography for the acquisition,
and maintenance of geoinformation. A major goal of this
concept is to integrate several components performing different
tasks within the framework of a knowledge-based system.
Currently, we are automating the interactive quality control step
by step using procedures that have been developed in the
research and development project with the University of
Hannover. The automated procedures consist of automatic steps
that are started by an operator and return a result that requires
further interaction of the operator.
The fully automatic part attains to solve the bigger part of the
quality control unassistedly and to focus the human operator to
those objects where the algorithms detect ambiguous situations.
Thus, the goal is to reduce the amount of human interaction by
automatically completing routine work which is a time
consuming part in the quality control process chain.
Challenging situations are afterwards analysed and solved by
the human operator in a separate step.
The results of the automatic procedure are passed to the human
operator in the form of a so-called traffic light diagnostics, i.e.
the results are displayed by means of red and green colour. An
attribute corresponding to the traffic light diagnostics and
737]
indicating the result of the automatic procedure is attached to
each inspected object. If the algorithm is able to detect and
locate the corresponding image object without observing
inconsistencies, the ATKIS object is marked with green colour.
Otherwise the object is labelled red. i.e. not accepted, since the
algorithm was not able to establish full correspondence. The
human operator can access more parameters of the diagnostics
whenever necessary.
Since the human operator decides on acceptance or rejection in
case of the red objects only, the decision of the automatic
procedure has to be reliable in particular for objects labelled as
green. The different situations that can occur when comparing
decisions from a human operator and diagnostics from
automatic procedures are classified in Table 1.
Automatic Green Red
Human Operator
Green
Red
True Positive
False Positive
False Negative
True Negative
Table 1. Confusion matrix of decisions: human operator vs.
automatic procedure, terminology.
2.2 Components of the System
The system is designed as a knowledge-based workstation,
which provides functionality from knowledge-based
photogrammetric image analysis and cartography for the
production of geoinformation. It consists of three major parts:
the GIS component, the knowledge-based component, and the
image analysis component (q.v. Figure 1).
| 1) Automatic Pre-Processing
GIS Component
3) Interactive Post-Processing
E
whe
Knowledge-Based
Component
e Control of Processes
* Object Extraction
e Verification
Image Analysis
Component * Update
Figure 1. The components of the system for quality control
The G/S component is based on ArcGIS. It acts as an automatic
pre-processor of the ATKIS data, as an interface to the database
and to the image processing system, as the environment for
interactive post-processing of automatically derived results, or
generally spoken as the user interface for handling the whole
workflow and for visualising the overlay of orthoimages and
ATKIS dataset.
The knowledge-based component is designed for making
knowledge about topographic objects available, for transferring
it to the image analysis component in a suitable way, for
handling the results from the image analysis component, for
deriving scene descriptions, and for controlling the complete
automatic workflow.
The image analysis component comprises the automatic image
feature extraction modules and the comparison with the original