Digital Signal Processing Reference
In-Depth Information
adaptive decision based on the mean and variance of target's gray level valueit's
decision criteria is given in formula (8) shown below
Ixy MeanI w Th
(, )
−
( _ )
>
∗
σ
(8)
k
k
2
Where,
w
is the mean value of pixel's intensity in tracking window,
M ean I
(_)
k
σ
is the variance.
Th
is an adjustable parameter that its parameter's selection is
related to the fluctuation degree of background pixel's gray level value.
Step2
2
take the binarization operation on the raw measurements detected by thre-
sholding the pixels in time differenced image and adaptive decision in the tracking
windowand store the location and gray level values corresponding to those mea-
surements
Step3
Clustering the measurements of time domain differenced frame of
imagefind its cluster center
and also clustering the measurements of adaptive
Z
_
c
detectionfind its cluster center
Z
_
a
Step4
Compare the coordinates of
c
and
a
:
Z
_
Z
_
1
If
≤
compute average coordinate position
cent
of
|
ZcZa
_
−
_|
1
Z
_
α
in
c
and
regard it as position of fusion center. According to variance
Z
_
Z
_
a
formula (6)
take
values in [1.5, 2.5]. Within the tracking window,
divide the raw measurements detected by parallel
way of decision in differenced frame of image and
adaptive thresholding into two parts, the Figure1
given for this purpose. Within the elliptical area, will
carry out logical “or” operation to measurements
those came from differenced image detection and
adaptive decisionand carry out logical “and” oper-
ation to measurements those of outside elliptical
area.
2If
|
set the measure
λα
usually
λ
Center of
tracking
window
Ellipse area
Center of clustering fusion
Tracking
window
Measurements
− >
in tracking window,
carry out logical “or” operation to measurements
those came from differenced image detection and
adaptive decision.
Step5
ZcZa
_
_|
1
Fig. 1.
Measurements in tracking
window
By decision fusion of Step4according to the advantages of local intensi-
ty of target in tracking window and the statistics of gray valueset the thresholdget
the measurements at time k, that is
() (, ,)
. The effective measurement detection
zk
=
xyI
algorithm is shown in Figure 2:
4
The Tracking Algorithm Based on Gaussian Particle Filtering
For nonlinear filtering problemparticle filter can be adapted well to Gaussian, or
non Gaussian noise environments. The N sampling points {x
j
, j=0,1,2,…,k}
