Open AccessOcclusion detection via correlation filters for robust object tracking
Author(s) -
Zhang Wei,
Kang Baosheng
Publication year - 2018
Publication title -
the journal of engineering
Language(s) - English
Resource type - Journals
ISSN - 2051-3305
DOI - 10.1049/joe.2017.0795
Subject(s) - occlusion , tracking (education) , computer vision , computer science , artificial intelligence , video tracking , correlation , object (grammar) , mathematics , medicine , psychology , geometry , cardiology , pedagogy
This study presents a robust object tracking method based on occlusion detection via correlation filters. In the proposed method, multi‐feature kernelised correlation filter is employed to estimate the preliminary location of the tracked target. To predict target occlusion state, the intrinsic relationship between the most reliable tracked target and its context information is exploited via correlation filters, together with a response stability constraint to make the detection more reliable. A long‐term filter is activated to recover the target if the occlusion occurs. Furthermore, the model is updated adaptively based on the changes of occlusion state and target appearance to make the tracking process robust. Extensive experimental results demonstrate that the proposed tracking method with occlusion detection performs favourably against 15 state‐of‐the‐art trackers over 100 challenge sequences on the object tracking benchmark OTB‐2015.