Adaptive Correlation Model for Visual Tracking Using Keypoints Matching and Deep Convolutional Feature

. 2018 Feb 23;18(2):653. doi: 10.3390/s18020653

Algorithm 1: Proposed tracking algorithm.

Input : Image I; initial target position

p_{0}

and scale

s_{0}

; previous target position

p_{t - 1}

and scale

s_{t - 1}

Output : Estimated object position

p_{t}

and scale

s_{t}

Initialize correlation filters

H_{1}^{t r a n s}

H_{1}^{s c a l e}

and set

K_{1}

D_{1}

Foreach

I_{t}

Extract multiple

f_{c}

generated by VGG-Net;

Compute the translation correlation

y_{t r a n s}

using Equation (5) and Equation (6)

Set

p_{t}

to at the maximum of

y_{t r a n s}

Compute the translation correlation

y_{t r a n s}

using Equation (9)

Set

s_{t}

to at the maximum of

y_{s c a l e}

Compute discounting factor

ρ_{t}

using Equation (12) to Equation (15)

Update

A_{t r a n s}

B_{t r a n s}

A_{s c a l e}

B_{s c a l e}

, K, D

End