Probabilistic and Statistical Methods - Information Fusion in Signal and Image Processing - page 72

conditional probabilities P(w/Ci)

P(w/Ci) P(Ci)

P(w/river)

P(w/city) P(city)

P(w/city)

P(w/river) P(river)

gray level space gray level space

Figure 6.2. Left: conditional probabilities of the cities and rivers in the texture image on XS2.
Right: probabilities conditional to the two classes multiplied by their a priori probability
P ( city ) = 11% and P ( river )=2%

Figure 6.3 shows the fusion result, for an a priori maximum criterion. The river is
superimposed in white over the original image, as well as the contours of the urban
areas. The rest corresponds to the class C 3 .

6.9. Probabilistic fusion methods applied to target motion analysis

After seeing an example of Bayesian fusion in image processing, we are now going
to see other probabilistic methods based on different concepts applied to a traditional
problem in signal processing, i.e. that of target motion analysis. After a general pre-
sentation of target motion analysis, we will examine in detail some aspects that are
more relevant to data fusion. In particular, we will focus on showing what the basic
techniques of target motion analysis can contribute to a detection system.

6.9.1. General presentation

First of all, let us point out the goal of target motion analysis: determining, based
on measurements (observations), the trajectory of a moving object. This essentially
involves a framework in which a dynamic system is partially observed. A traditional
and historical example is the estimation of a planet's trajectory, based on optical obser-
vations (angle measurements). Below are the most widespread types of measurements
in target motion analysis:
- angles (sonar, ESM, infrared, etc.);