Robust Scale Adaptive Tracking by Combining Correlation Filters with Sequential Monte Carlo

. 2017 Mar 4;17(3):512. doi: 10.3390/s17030512

Algorithm 1 Proposed Tracking Algorithm

1:
Initialize the model coefficients $\hat{α}$ and the target appearance $x$ with the bounding box $B_{1}$ given in the 1-st frame.
2:
for i = 2 to end of the sequence do
3:
Sampling image patches $z^{s}$ in different scale $s \in S$
4:
for each scale in $s_{i}^{n} \in S$ do
5:
Calculate the response map $R_{i}^{s^{n}} = F^{- 1} ({\hat{k}}^{x z} \cdot \hat{α})$
6:
$p (o_{i} | s_{i}^{n}) \leftarrow \max (R_{i}^{s^{n}})$
7:
$p (s_{i}^{n} | o_{i}) \leftarrow p (o | s_{i}^{n}) \times \int p (s_{i}^{n} | s_{i - 1}) p (s_{i - 1} | o_{i - 1}) d s_{i - 1}$
8:
end for
9:
$s^{'} = a r g \max_{s_{i}^{n}} (p (s_{i}^{n}))$
10:
calculate the $P S R (i)$ by Equation 12 in the most likely scale $s^{'}$
11:
if $P S R (i) > T_{u}$ then
12:
update the model coefficient $\hat{α}$ and $x$
13:
end if
14:
if $P S R (i) > T_{d}$ then
15:
target position $p_{i} = a r g \max (R_{i}^{s^{'}})$
16:
target scale $s_{i} = s^{'}$
17:
else
18:
target position $p_{i} = p_{i - 1}$
19:
target scale $s_{i} = s_{i - 1}$
20:
end if
21:
end for