Multi-Agent Methods: An Example of an Architecture and its Application for the Detection, Recognition and Identification of Targets - Information Fusion in Signal and Image Processing - page 229

method space 4 . The method we propose complies with a series of fundamental prin-
ciples such as:
- focusing and adaptation;
- distribution and co-operation;
- uncertainty management;
- incrementality and learning.

Each concept allows the system to navigate within a space, but also to switch from
one space to another. The suggested system designs its own strategy depending on the
information available and proceeds in an incremental fashion. Over time, the system
makes hypotheses that are confirmed or not, later on, depending on the information
that is gathered. Thus, the suggested method follows an anytime philosophy, which
means that there is always a possible decision and it is refined over time.

The system is equipped with various adapting capabilities. Here are the major
ones:
- local adaptation, which is present, among other places, within the Detection
agent, since it uses the statistics associated with the current sample to adjust the seg-
mentation parameters;
- global adaptation which is present in two different forms:
- adaptation of the agent population: the presence of movement in the image
can force the system to request a complementary analysis of the areas in question, thus
new agents are created to answer to this request,
- adaptation of the data structures: over a sequence of images, the system is
capable of managing the updates of the knowledge base and the evolution of the
world model by using in particular a set of probabilities to assess the confidence in
the results.

Exploration and analysis strategies are not set in stone. They depend on decisions
made locally by agents and the modification of knowledge allows the systems to
develop different strategies. This operating mode described for the detection phase
applies in the same way to the recognition and identification phases [EAL 01]: in this
work, the characterization phase was completed by a discrimination phase, since the
adaptability relies on the concept of the usefulness of attributes.

10.7. Bibliography

[BAJ 88] B AJCSY R., “Active Perception”, Proceedings of the IEEE , vol. 76, p. 996-1005,

1988.

4. This study was conducted as part of a CIFRE convention between the E/SCS/V of EADS, the
TIMC/IMAG/SIC laboratory and the DGA/CTA/GIP laboratory.