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 218

This description of the scene is constructed incrementally. The update of the world
model is conducted throughout the analysis process. The information gathered is char-
acterized by a confidence measure that makes it possible to evaluate the membership
coefficient of an entity to a given concept. This parameter evolves over time and prop-
agates through the information structure using Bayesian networks. The networks were
constructed with the “Bayes Net Toolbox 3 ”.

For a given support, its confidence depends on three sources of information:
- the samples that have contributed to constructing it;
- the additional information, if there is any, such as the presence of movement;
- the other supports that have induced its construction by focusing (the concept of
lineage).

The calculation of confidence associated with each support can be shown in the
form of a causality graph. It depends on several sub-networks (Figure 10.8).

smpl N

smpl 1

measurements taken
from the image

parent

sup

mov id

the movement focusing strategy

Figure 10.8. Description of the Bayesian network used for calculating the confidence
of a support. This confidence depends on the confidence obtained for each sample,
on focusing relations and on the movement

We are going to describe each one of these sub-networks in detail:
- for each sample (Spl1, . . . SplN) that belongs to the support, a statistical analysis
is conducted in order to calculate the measure of confidence for each attribute (Att1,
. . . AttN). The system calculates this measure of confidence as a gap between two
probability distributions originating from the measure and the model. These measures

3. www.cs.berkeley.edu/%7Emurphyk/Bayes/bnt.html.