Digital Signal Processing Reference
In-Depth Information
Step1:
Initialize the particle samples: Get target information at time k=1 by detec-
N
1
X
,
tion procedurecreate initial sample set
n
k
in which
N
is particle number,
n
n
=
1
particle sample obey Gaussian distributionand are sampled in random
n
=
1, 2...
N
way
Step2
By use of the effective measurement detection algorithm presented in this
paperget measurements
zk
at time k;
()
zk
()
assign the weights to the particles
based on the Gaussian posterior probability densityits density function described as
the formula (11) shown below:
Step3
according to the measurements
1
(
SO
−
)
2
WX Z
(
|
)
=
p
m
=
1, 2, 3,...
N
n
m
(11)
k
k
1
k
2
C
2
π
C
2
Where,
is the innovation of particle m;
O
,
C
are the clustering center of measurements and covariance matrix of relative
clustering center of particles;
Step4
is the coordinates of particle
m
,
S
m
SO
−
m
Estimate the target position by using normalized weightand estimate the
speed and speed innovations of target in next frame by using prior two frames and
current frame tracking information;
Step5
If tracking is not over, then go back to step1.
5
Experimental Results and Analysis
5.1
Experimental Environment and Results
This paper uses the infrared image sequence for validation, Experiments completed
by Matlab7.0 on the PC with Intel(R) Pentium(R) processor 1.80GHz 512MB storage.
The SNR of image sequences defined as formula (12) shown below:
μ
−
M ean
SNR
=
10 log
(12)
σ
In which
u
is the average grayscale value of target,
σ
is a standard deviation of
target area
Mean
is mean of target area. Tracking performance is validated by us-
ing absolute error which defined as formula (13) shown below:
_
Error
=
TrackPo
−
X
true
(13)
Where,
TrackPo
is estimated target position,
X r e
is true target position. The
target take uniform movement in 1-40 framesaccelerated movement in 41-70
framesdecelerated movement in 71-120 framesaccelerated movement in 121-180
frames. Then, we collected a total of 180 infrared image sequences, and set the target
size of 3x3 pixelsrectangular tracking window size of 9x9 pixels. Target image
occupied a circular area, set
λα
=2, the particle number is 200.
