Correlation Filter Based Visual Tracking With Circular Shift On Local And Semi-Local Domains

Abstract

In Parallel Consideration Of Spatial And Appearance Selective Attention Like Human Visual
Perception, A Correlation Filter (CF) Can Be Estimated Based On The Association Between Object
Region (Local) And Background Region (Semi-Local). In The Proposed Work, Initially, An Efficient
Correlation Filter Is Estimated With The Selective Circularly Shifted (CS) Patches In The Local Domain
And The Histogram Of Gradients (Hogs) In Local And Semi-Local Domains. To Effectively Remove The
Boundary Effect Of The CF, The Histogram Of Gradients (Hogs) Is Extracted For Positive And Negative
Samples Which In Turn Referred The Semi-Local Domains. Finally, The Linear Combination Of The Two
Domain Models Is Used To Determine The Object Location. The Proposed Visual Tracking Method (OTCS) Apparently Reduces The Distractors And Effectively Removes The Boundary Using The Combination
Of Object Tracking And Circularly Shifted (OT-CS) Approach. The Proposed OT-CS Model Has Been
Evaluated With Various Tracking Benchmark Datasets And Results Are Compared. The Results Show
That, OT-CS Has High Contribution To Effectively Resist The Distractors With Low Computational Cost
For Efficient Visual Tracking.