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 206

strategies based on the available information. In order to abide by this philosophy
inherent to multi-dimensional space, we have made a series of choices which we will
justify as we go along.

10.2.1. Representation space and situated agents

We suggest focusing the design of our system on the concept of situated agents.
Such an agent is embedded in a three-dimensional space: the image space (the
“where”), the goal space (the “what”) and the method space (the “how”). A represen-
tation of the system is shown in Figure 10.2.

goals

image

each arrow gives the answer to a question
Where?
What?
How?

associated space

image

goal

knowledge base

agents store their results in the world model

Figure 10.2. General view of the multi-agent system. Each agent is embedded in a
three-dimensional space comprised of the image space, the goal space and the method space.
The agents are situated in the image, with a precise goal. They work locally using the data
from the knowledge base to produce partial results that will be stored in the world model.
This world model is then shared among all of the agents

The goal space is comprised of a set of concepts that represent the elements in
the scene we wish to analyze, such as roads, vegetation, the sky, the sea, etc. In the
detection phase, an agent for a given concept (what?) will select a method (how?) in
order to extract a region of interest (where?). The population of agents constantly uses
the information specified in the knowledge base or gathered in the world model to
work out its own strategy:
- an agent is always situated in an area of interest. For example, the system may
be searching for vehicles on the roads. Extracted from a previous process, roads then
constitute the areas of interest for vehicle tracking;