Image Processing Reference
- in the confidence in the information since the system reinforces hypotheses or
not through time and weights the uncertainty depending on the information obtained;
- finally, in the knowledge base, which gives an operational description of the
scene elements, since the system can refine over time its knowledge that was set
beforehand. The term learning is used in this sense.
This incrementality is essential in the context of our application, where the relevant
information can be summed up to a few pixels in the detection phase. Therefore, the
objective is both to progressively take advantage of the environment and to authorize
the system to be adaptive.
We have just described the fundamental points of the design we are proposing.
They are associated with a set of keywords: focusing, adaptation, distribution, co-
operation, uncertainty, decision, incrementality and learning. These concepts are the
basis of the multi-agent architecture we will now describe.
10.3. The multi-agent system: platform and architecture
We propose a decentralized architecture to solve the problem of the detection,
recognition and identification of targets. We will focus our attention on the detection
phase. We will first present the system's architecture, i.e. the agent in its environment.
We will then discuss further the multi-agent platform that allows the exchange of
10.3.1. The developed multi-agent architecture
The agents behave independently. They work in an area of the image with their
own objective, in competition or in parallel. The knowledge necessary to achieve an
objective is specified in the knowledge base shared by all of the agents. Each one of
them gathers information and stores it in a world model, which is also shared. This
architecture is shown in Figure 10.2.
10.3.2. Presentation of the platform used
A community of agents cannot exist without a set of generally complex models
defined within a platform. This platform specifies the fundamental elements such as
the hardware and software architecture, or also the communication modes used by the
agents to communicate with each other and with the data. The different features pro-
vided by this structure are summed up in Table 10.1, according to the plan suggested
by the ASA (Agent System Architecture) workgroup, a group supported by the AFIA 1
and the GDR-I3 2 .
1. French Association for Artificial Intelligence.
2. The Research Group for Information, Interaction and Intelligence of the CNRS.