A Fast Adaptive Multi-Scale Kernel Correlation Filter Tracker for Rigid Object

. 2022 Oct 14;22(20):7812. doi: 10.3390/s22207812

Algorithm 1: MMKCF

Input:
The video frame, $I_{t}$ ; Initial bounding box of the target, $b b$ ;
Output:
The target position predicted by the tracker, p;
1:
while $I_{t}$ do
2:
if $t = = 1$ then
3:
$[{\hat{a}}_{0}, X_{0}] = i n i t i a l i z e (I_{1}, b b)$
4:
else
5:
if $1 < t \leq γ$ then
6:
for $i = s i z e (I_{t}) \times (1 - ε)$ to $s i z e (I_{t}) \times (1 + ε)$ do
7:
$[m a x_r e s p o n s e (i), p (i)] = k e r n e l_c o r r e l a t i o n_f i l t e r (I_{t}, {\hat{a}}_{t}, X_{t})$
8:
end for
9:
$\{\begin{matrix} {\hat{a}}_{t + 1} = (1 - μ) {\hat{a}}_{t} + μ {\hat{a}}_{t} \\ X_{t + 1} = (1 - μ) X_{t} + μ X_{t} \end{matrix}$
10:
$C + = m a x (m a x_r e s p o n s e)$
11:
else
12:
for $i = s i z e (I_{t}) * (1 - ε) : s i z e (I_{t}) * (1 + ε)$ do
13:
$[m a x_r e s p o n s e (i), p (i)] = k e r n e l_c o r r e l a t i o n_f i l t e r (I_{t}, {\hat{a}}_{t}, X_{t})$
14:
end for
15:
if $m a x (m a x_r e s p o n s e) \leq (β_{t} = θ \cdot \frac{C_{t}}{t})$ then
16:
$\{\begin{matrix} {\hat{a}}_{t + 1} = {\hat{a}}_{t} \\ X_{t + 1} = X_{t} \end{matrix}$
17:
else
18:
$\{\begin{matrix} {\hat{a}}_{t + 1} = (1 - μ) {\hat{a}}_{t} + μ {\hat{a}}_{t} \\ X_{t + 1} = (1 - μ) X_{t} + μ X_{t} \end{matrix}$
19:
$C + = m a x (m a x_r e s p o n s e)$
20:
$p = max (p)$
21:
end if
22:
end if
23:
end if
24:
end while