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 219

are then propagated using the Bayesian network. This is described by the diagram in
Figure 10.9.

= T = F = T = F

smpl = T smpl = F

T
T

T

F

F
F

T
F

smpl

Figure 10.9. The computation of the confidence in a sample is conducted using a Bayesian
network. In this example, two methods are used to characterize a sample

The measure of confidence associated with the sample is given by the following
relation:

P ( Spl = T )= P ( Spl = T/Att 1= T,Att 2= T ) · P ( Att 1= T ) · P ( Att 2= T )

+ P ( Spl = T/Att 1= T,Att 2= F ) · P ( Att 1= T ) · P ( Att 2= F )

+ P ( Spl = T/Att 1= F, Att 2= T ) · P ( Att 1= F ) · P ( Att 2= T )

+ P ( Spl = T/Att 1= F, Att 2= F ) · P ( Att 1= F ) · P ( Att 2= F )

- additionally, each support has a lineage resulting from the focusing relations. For
example, a hypothesis of the type “vehicle on a road” will be reinforced, whereas the
probability of having “road in the sky” will be minimized. The possibilities associated
to each of the hypotheses are set beforehand and later evolve through the Bayesian
network;
- the movement information can also reinforce a hypothesis. For example, the
movement information will reinforce a vehicle hypothesis and not a road hypothesis.
The concept used for calculating the confidence measure is similar to that described
for the samples.

Depending on the information calculated by the agents, the probabilities will
evolve and be updated. Information is propagated through the network based on the
“causes-consequences” dependency relations that are set.

10.6. The results

Given an initial goal, the system's objective is to pursue the exploration of the
image or sequence as long as it is useful. This exploration is guided by focusing rules