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 221

first
agent in
the image
the first letters indicate
the agent type:
C: concept agent
L: localization agent
Dt: detection agent
Dc: decision agent
the latter specifies the objective:
V: vehicle
indicates the probability for
a support to be a member of
the vehicle concept
Figure 10.11. Agent graph generated for detecting vehicles using the direct strategy; four
types of agents are implemented: the Concept, Localization, Detection and Decision agents.
All these agents have the same objective, namely the vehicle concept.
In this example, nine vehicle-type supports are detected
This population is organized around four types of agents:
- the Concept agent is the first agent that is created. Its role is to initialize the
system and activate the exploration loops according to the information at its disposal
in the knowledge base. In this example, it requests the activation of spatial exploration
using a localization agent, since localization criteria are specified. On the other hand,
no movement information is described in the base. As a result, the loop related to
temporal exploration is not activated;
- a Localization agent searches the image for pixels that satisfy the set of localiza-
tion criteria specified in the knowledge base and associated with the vehicle objective.
After a labeling phase, this agent provides elementary areas referred to as samples,
which are associated with statistical information. As soon as a sample is identified,
the Localization agent requests the creation of a new Detection agent. The nine sam-
ples that were found led to the creation of nine Detection agents;
- each Detection agent works with the associated sample and performs a local
statistical analysis to extract the object's support. No focusing strategy is specified
and therefore the spatial exploration comes to an end;
- the Decision agent has no complementary information regarding movement
since no knowledge on that subject is specified. Therefore, it transmits the data ob-
tained without requesting complementary analysis and provides a decision.