Algorithm 3: Dictionary update.

for $t=t_0+1$ to the end of the frame sequence

1. if $p(X_t^j|Y_t,X_{t-1})>{\theta }_p$

1-1. add the target descriptor $v_t^{tg}$

and background descriptors ${\overrightarrow{v}}_{t,(a_x,b_x)}^{bg}$ to F

1. else

1-2. $k_p=k_p+1$

for every k frames

2. build the new metrics $D_w$ and $C_w$ using the vectors in F

by structured sparse PCA

3. update classifier $\mathsf{\Phi }$ using $D_w$ and $C_w$

4. compute $\mathsf{\Phi }(\overrightarrow{v})$ for $\overrightarrow{v}\in F$

5. keep the $k_0$ largest target descriptors in F,

and delete the rest descriptor from F

6. $k_p=0$

end