Algorithm 1. Proposed tracking algorithm.

Require: An input image sequence and object target locations $P(x_1,y_1,w_1,h_1)$ in the first frame.  Ensure: Target object locations in the subsequent frames $P(x_t,y_t,w_t,h_t),$ where $t=2,3,\dots$   repeat //Translation calculation Calculate the correlation response map $R_z$ using (6) and estimate the target object location $(x_t,y_t)$ by finding the maximum response position from $R_z$; //Scale estimation Construct target feature pyramid around the position $(x_t,y_t)$ and extract the HOG features after resizing the image patch with same size and calculate the scale correlation response map $R_s$ using (6); Determine the optimal scale $s$ using (12); Calculate the PSR score using (8); Compute dynamic learning rate $n_t$ using (9) and (10); //Translation correlation model update if $PSR>T_1$ then Update the translation model parameters using (11); End //Scale correlation model update if $\max (R_s)>T_2$ then Update the scale model parameters using (13); End Until end of the image sequences;