• DLR (GE)

• SENER (SP)

• SIC (BE)

• DLR

• SENER

• SIC

When one looks at the evolution in the field of earth observation data acquisition systems, it becomes obvious that the flux of aerial and satellite imagery will grow exponentially, whereas the number of trained photo-interpreters will not. Therefore the real challenge for computer system developers in the field of scene analysis is to build a semi-automatic system which relieves the human photo-interpreter from the routine part of his work, so that he can focus on those parts of the job for which his human intelligence and intuition are indispensable. SAHARA system is a possible layout for such a system.

The aim of the project was to show the potential usefulness of image interpretation techniques to help photo-interpreters. Typically, this could include drawing a sketch of an airport, looking for changes with respect to a reference geographic database, etc.

As a test case, the project was dedicated to the interpretation of airports. In this scope the main objects of interest are runways, taxiways, shelters, airplanes and buildings.

Due to the complexity of real scenes, object-related detectors are not able to solve the problem on their own. It is necessary to incorporate high level expert knowledge into the system. The proposed system makes use of a strategy to efficiently build an incremental interpretation. It also uses global rules to ensure a coherent interpretation (i.e. no building in the middle of a runway, …). Even with this high level knowledge, the system will make mistakes and it is of prime importance to allow the user to intervene at any moment to help the interpretation, by correcting errors, stopping the search in a non promising area, making the final decision on the affectation of a building, etc.

Each type of knowledge requires a different representation method.

Therefore, those knowledge sources were encoded in independent modules communicating through a blackboard as shown on following figure.

During the project, an important effort as been invested in the development of an efficient Graphical User interface (GUI). This GUI allows the user to follow the interpretation and to intervene at any moment. Further developments are needed to allow the user to easily modify the strategy or the rules. Diagnostic tools to analyze the effect of a rule and eventually correct that rule are also needed.

The database contains multi-sensor/multi-spectral images of 5 airports. The images were made by degrading airborne images to simulate earth observation images that will be available in the near future for commercial satellites:

• Vis 1m

• SAR 5m

• IR 5m

DLR provided all images for the project. The images shown here are down-sampled versions

• The strategy module first requests the detection of the runway area.
• Rules (length, width, …) come into play to remove false alarms.
• An airport area is then defined around the detected runways.
• Taxiways are searched in that region.
• This search is carried out in several band (VIS and IR) and the detected objects are fused.
• Rules come into play to transform the big taxiway candidate into a runway (red), to remove some false alarms based on their geometry or on radiometric measures in the different spectral bands (SAR, IR). A typical radiometric based rule states that a taxiway should be smooth and thus sufficiently dark on the SAR image. Note that a good candidate (lower right taxiway) is also removed. This typical example shows that the results could be improved by introducing more geometry based rules (the confidence of the removed taxiway could be increased by a rule because it connects to taxiways.
• The operator interacts with the system to remove some false alarms (the road above), to add undetected objects (the lower right taxiway).
• The search for an object may also be initiated by the user for a specific region of interest. Here, taxiways are requested and the system returns the precise outline of the taxiway. In this way the operator wins time because he doesn’t have to manually vectorize the precise border of the object.
• Shelters are finally searched for in a region of interest around the detected taxiways.

During the SAHARA final meeting, the demonstrator was presented to the Photo-Interpreters (PI’s) of the WEU (note that some of them were involved from the beginning on to provide some user requirement and to provide the needed expert knowledge) satellite center. Discussions showed that the most promising use of the system is linked to the detection of changes (surveillance of critical areas, …).

SAHARA was funded by the Western European Union (WEU) Satellite Center.

P. Druyts, W. Mees, D. Borghys, C. Perneel, M. Acheroy, J.L. Valero, "SAHARA: Semi-Automatic Help for Aerial Region Analysis'', in ISPRS Joint Workshop, Hannover, Vol 17, pages 267-274,1997

W. Mees and C. Perneel, "Advances in Computer Assisted Image Interpretation", in Informatica - International Journal of Computing and Informatics

W. Mees, "Scene analysis: fusing image processing and artificial intelligence", in IEEE-CS looking .forward, 1997