Gaussian Particle Filter Based Algorithm for Tracking of a Dim Moving Point Target in IR Image Sequences - Multimedia and Signal Processing - page 221

Experimental data setup: In the case of the strong background fluctuation, we add
white noise on the IR image sequences. its deviation is 0.0001, SNR is 7dband are
tracked by the algorithm of paper [1] and of this paper separately; then add random
speckle (non Gaussian noise) to IR image sequences, the mean is 0deviation is
0.0005SNR is 7db, use the algorithm presented in this paper, and the adaptive thre-
shold is taken to Th=2, initial state of target is [39 83 0.8 0.3 0 0]. The comparisons
of trajectory of tracking algo-
rithm between paper [1] and this
paper are shown in Figure 3 and
Figure 4 respectively.
The comparisons of absolute
tacking error under different
noise types and different tracking
algorithms presented in this
paper and paper [1] is shown in
Figure 5. After 50 times of
Monte-Carlo experiments, the
average detection and tracking
time are given in Table 1.

True trajectory
PF tracking

Fig. 3. Whole trajectory of tracking algorithms

5.2 Analysis of the Results

We can see from the Figure 3
and Figure 4, the algorithm pre-
sented in this paper can effective-
ly tracking the dim moving target
in IR image sequences. Figure 5
shows that the absolute tracking
error is much smaller, the error
fluctuation is smaller, the stabili-
ty is better, the noise type's
influence to target tracking accu-
racy is smaller than paper [1],
and no error drift phenomenon
occurred. But all have better real-
time efficiency.

True trajectory
PF tracking

Fig. 4. Local trajectory of tracking algorithms

Table 1. Comparison of single frame detection time and tracking time

Single frame
Detection time (s)

Tracking time of
180 frames (s)

PF 0.003386 13.459407

G-GPF 0.006666 13.480746

S-GPF 0.006658 13.480316