Visual Tracking Using Max-Average Pooling and Weight-Selection Strategy

Algorithm 1

The proposed algorithm with max-average pooling.

Input: An image from the video sequences, the center point of the bounding box, the noise coefficient

𝜆, and the number of particles 𝑁.

(1) Sample 𝑁 particles around the initial point of the bounding box, and the distribution of the particles

 is the normal distribution, which is implemented by the function randn.

(2) Initialize the dictionary using tracking results of the first ten frames.

(3) Compute the formula to obtain the sparse codes: min_𝑊,𝐷∑^𝑁_𝑖=1𝑥_𝑖−𝑤_𝑖𝐷²+𝜆𝑤_𝑖

 Note that 𝑑_𝑘≤1 and 𝑘=1,2,…,𝐾, where 𝐾 is the size of the patch’ width by height.

(4) Pool the features by computing: 𝑤_𝑖=(max{𝑤_𝑖1,𝑤_𝑖2,…,𝑤_𝑖𝑗,…,𝑤_𝑖𝑁})/(∑𝑤_𝑖𝑗)

(5) Compute the error of reconstruction:

               𝑒𝑟𝑟𝑜𝑟=∑𝑥_𝑖−𝑤_𝑖𝐷²

(6) Finally, obtain the particle which is corresponding to the minimum error.