The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Baru-baru ini, penjejak visual berdasarkan rangka kerja penapis korelasi kernel (KCF) mencapai hasil keteguhan dan ketepatan. Penjejak ini perlu mempelajari maklumat tentang objek daripada setiap bingkai, oleh itu perubahan keadaan objek mempengaruhi prestasi penjejakan. Untuk menangani perubahan keadaan, kami mencadangkan penjejak KCF novel menggunakan peta tindak balas penapis, iaitu peta keyakinan dan model penyesuaian. Kaedah ini mula-mula mengambil kaedah kolam skala langkau yang menggunakan saiz tetingkap berubah pada setiap dua bingkai. Kedua, lokasi objek dianggarkan menggunakan gabungan tindak balas penapis dan persamaan histogram pencahayaan pada berbilang titik dalam peta keyakinan. Selain itu, kami menggunakan pengesanan semula berbilang puncak peta keyakinan untuk mengelakkan hanyut sasaran dan mengurangkan pengaruh pencahayaan. Ketiga, kadar pembelajaran untuk mendapatkan model objek diselaraskan, menggunakan tindak balas penapis dan persamaan histogram pencahayaan, dengan mengambil kira keadaan objek. Secara eksperimen, penjejak yang dicadangkan (CFCA) mencapai prestasi cemerlang untuk jujukan penanda aras yang mencabar (OTB2013 dan OTB2015).
Zhaoqian TANG
Meiji University
Kaoru ARAKAWA
Meiji University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Salinan
Zhaoqian TANG, Kaoru ARAKAWA, "Correlation Filter-Based Visual Tracking Using Confidence Map and Adaptive Model" in IEICE TRANSACTIONS on Fundamentals,
vol. E103-A, no. 12, pp. 1512-1519, December 2020, doi: 10.1587/transfun.2020SMP0007.
Abstract: Recently, visual trackers based on the framework of kernelized correlation filter (KCF) achieve the robustness and accuracy results. These trackers need to learn information on the object from each frame, thus the state change of the object affects the tracking performances. In order to deal with the state change, we propose a novel KCF tracker using the filter response map, namely a confidence map, and adaptive model. This method firstly takes a skipped scale pool method which utilizes variable window size at every two frames. Secondly, the location of the object is estimated using the combination of the filter response and the similarity of the luminance histogram at multiple points in the confidence map. Moreover, we use the re-detection of the multiple peaks of the confidence map to prevent the target drift and reduce the influence of illumination. Thirdly, the learning rate to obtain the model of the object is adjusted, using the filter response and the similarity of the luminance histogram, considering the state of the object. Experimentally, the proposed tracker (CFCA) achieves outstanding performance for the challenging benchmark sequence (OTB2013 and OTB2015).
URL: https://global.ieice.org/en_transactions/fundamentals/10.1587/transfun.2020SMP0007/_p
Salinan
@ARTICLE{e103-a_12_1512,
author={Zhaoqian TANG, Kaoru ARAKAWA, },
journal={IEICE TRANSACTIONS on Fundamentals},
title={Correlation Filter-Based Visual Tracking Using Confidence Map and Adaptive Model},
year={2020},
volume={E103-A},
number={12},
pages={1512-1519},
abstract={Recently, visual trackers based on the framework of kernelized correlation filter (KCF) achieve the robustness and accuracy results. These trackers need to learn information on the object from each frame, thus the state change of the object affects the tracking performances. In order to deal with the state change, we propose a novel KCF tracker using the filter response map, namely a confidence map, and adaptive model. This method firstly takes a skipped scale pool method which utilizes variable window size at every two frames. Secondly, the location of the object is estimated using the combination of the filter response and the similarity of the luminance histogram at multiple points in the confidence map. Moreover, we use the re-detection of the multiple peaks of the confidence map to prevent the target drift and reduce the influence of illumination. Thirdly, the learning rate to obtain the model of the object is adjusted, using the filter response and the similarity of the luminance histogram, considering the state of the object. Experimentally, the proposed tracker (CFCA) achieves outstanding performance for the challenging benchmark sequence (OTB2013 and OTB2015).},
keywords={},
doi={10.1587/transfun.2020SMP0007},
ISSN={1745-1337},
month={December},}
Salinan
TY - JOUR
TI - Correlation Filter-Based Visual Tracking Using Confidence Map and Adaptive Model
T2 - IEICE TRANSACTIONS on Fundamentals
SP - 1512
EP - 1519
AU - Zhaoqian TANG
AU - Kaoru ARAKAWA
PY - 2020
DO - 10.1587/transfun.2020SMP0007
JO - IEICE TRANSACTIONS on Fundamentals
SN - 1745-1337
VL - E103-A
IS - 12
JA - IEICE TRANSACTIONS on Fundamentals
Y1 - December 2020
AB - Recently, visual trackers based on the framework of kernelized correlation filter (KCF) achieve the robustness and accuracy results. These trackers need to learn information on the object from each frame, thus the state change of the object affects the tracking performances. In order to deal with the state change, we propose a novel KCF tracker using the filter response map, namely a confidence map, and adaptive model. This method firstly takes a skipped scale pool method which utilizes variable window size at every two frames. Secondly, the location of the object is estimated using the combination of the filter response and the similarity of the luminance histogram at multiple points in the confidence map. Moreover, we use the re-detection of the multiple peaks of the confidence map to prevent the target drift and reduce the influence of illumination. Thirdly, the learning rate to obtain the model of the object is adjusted, using the filter response and the similarity of the luminance histogram, considering the state of the object. Experimentally, the proposed tracker (CFCA) achieves outstanding performance for the challenging benchmark sequence (OTB2013 and OTB2015).
ER -