Open AccessA Camera Tracking System Based on Closed-loop Kernelized Correlation Filters
Author(s) -
Muzi Li,
Bo Cheng,
Shuai Zhao,
Junliang Chen
Publication year - 2019
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/533/1/012037
Subject(s) - artificial intelligence , computer vision , tracking (education) , computer science , kernel (algebra) , tracking system , video tracking , benchmark (surveying) , object (grammar) , filter (signal processing) , object detection , pattern recognition (psychology) , mathematics , psychology , pedagogy , geodesy , combinatorics , geography
Camera tracking is an important application in the computer vision. With the progress of the tracking-by-detection algorithm, industrial cameras can track targets autonomously. However, during the long-time tracking, it’s prone to miss target owing to heavy object occlusion, environment changing and objects appearance variations. Our camera tracking system is based on the kernelized correlation filter to track object, and we add a self-verification module to judge if the current tracking results are reliable. This can be useful in solving target missing when object meets occlusion or variations, and avoid model drift during long-time detection. Last but not the least, we keep the targets in the corner of the image by controlling our camera, avoiding object’s moving out of view. The kernel correlation filter bounded with self-verification and re-detection modules boost the performance. Extensive experiments on the OTB-2013 benchmark show that it performs better than state-of-the-art methods.