Open AccessCompressive tracking via oversaturated sub‐region classifiers
Author(s) -
Zhu Qiuping,
Yan Jia,
Deng Dexiang
Publication year - 2013
Publication title -
iet computer vision
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.38
H-Index - 37
eISSN - 1751-9640
pISSN - 1751-9632
DOI - 10.1049/iet-cvi.2012.0248
Subject(s) - artificial intelligence , tracking (education) , computer science , computer vision , classifier (uml) , pattern recognition (psychology) , position (finance) , stability (learning theory) , machine learning , psychology , pedagogy , finance , economics
This study proposed a tracking algorithm based on oversaturated sub‐region classifiers. Compared with the compressive tracking (CT), the tracker can reduce the influence of occlusion and improve the stability and accuracy of tracking result. First, the target region is divided into oversaturated sub‐regions randomly, and then some sub‐region classifiers are adaptively selected based on their confidence. Each selected classifier can find a candidate target position. At last, the place with the maximum candidate positions’ distribution density is the final location of the target. Experiments on different videos demonstrate that the proposed algorithm has stronger anti‐occlusion ability than the CT and is more robust and stable than the traditional sub‐region‐based tracking algorithm.