File Download
There are no files associated with this item.
Links for fulltext
(May Require Subscription)
- Publisher Website: 10.1109/TMM.2020.2990064
- Scopus: eid_2-s2.0-85100248597
- WOS: WOS:000613560200020
- Find via
Supplementary
- Citations:
- Appears in Collections:
Article: Intermittent contextual learning for keyfilter-aware uav object tracking using deep convolutional feature
Title | Intermittent contextual learning for keyfilter-aware uav object tracking using deep convolutional feature |
---|---|
Authors | |
Keywords | Unmanned aerial vehicles Correlation Visualization Training Object tracking |
Issue Date | 2021 |
Publisher | IEEE. The Journal's web site is located at http://www.ieee.org/organizations/tab/tmm.html |
Citation | IEEE Transactions on Multimedia, 2021, v. 23, p. 810-822 How to Cite? |
Abstract | Visual tracking, one of the most favorable multimedia applications, has been widely used in unmanned aerial vehicle (UAV) for civil infrastructure monitoring, aerial cinematography, autonomous navigation, etc. Most existing trackers utilize deep convolutional feature to enhance tracking robustness in scenarios of various appearance variation. However, they commonly neglect speed which is crucial for UAV with restricted calculation resources. In this work, a novel correlation filter-based keyfilteraware tracker with a new intermittent context learning strategy is proposed to efficiently and effectively alleviate the problems of background clutter, deficient description, occlusion, illumination change, etc. Specifically, context information is utilized to empower the filter higher discriminating ability through response repression of the omnidirectional context patches. Furthermore, keyfilter is produced from the periodically selected keyframe. The latest produced keyfilter is used to restrain the current filter's corrupted changes. Most importantly, context learning of correlation filter is implemented intermittently to fully increase the tracking efficiency. This intermittent learning strategy can ensure every filter maintain context awareness owing to the restriction of keyfilter, periodically enhancing the context awareness. Additionally, hand crafted and deep features are fused to establish a comprehensive appearance model of the tracked object. Substantial experiments on three challenging UAV benchmarks totally with 213 image sequences have shown that our tracker surpasses the state-of-the-art results, and exhibits a remarkable generality in short-term and long-term UAV tracking tasks as well as a variety of challenging attributes. |
Persistent Identifier | http://hdl.handle.net/10722/284899 |
ISSN | 2021 Impact Factor: 8.182 2020 SCImago Journal Rankings: 1.218 |
ISI Accession Number ID |
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Li, Y | - |
dc.contributor.author | Fu, C | - |
dc.contributor.author | Huang, Z | - |
dc.contributor.author | Zhang, Y | - |
dc.contributor.author | Pan, J | - |
dc.date.accessioned | 2020-08-07T09:04:05Z | - |
dc.date.available | 2020-08-07T09:04:05Z | - |
dc.date.issued | 2021 | - |
dc.identifier.citation | IEEE Transactions on Multimedia, 2021, v. 23, p. 810-822 | - |
dc.identifier.issn | 1520-9210 | - |
dc.identifier.uri | http://hdl.handle.net/10722/284899 | - |
dc.description.abstract | Visual tracking, one of the most favorable multimedia applications, has been widely used in unmanned aerial vehicle (UAV) for civil infrastructure monitoring, aerial cinematography, autonomous navigation, etc. Most existing trackers utilize deep convolutional feature to enhance tracking robustness in scenarios of various appearance variation. However, they commonly neglect speed which is crucial for UAV with restricted calculation resources. In this work, a novel correlation filter-based keyfilteraware tracker with a new intermittent context learning strategy is proposed to efficiently and effectively alleviate the problems of background clutter, deficient description, occlusion, illumination change, etc. Specifically, context information is utilized to empower the filter higher discriminating ability through response repression of the omnidirectional context patches. Furthermore, keyfilter is produced from the periodically selected keyframe. The latest produced keyfilter is used to restrain the current filter's corrupted changes. Most importantly, context learning of correlation filter is implemented intermittently to fully increase the tracking efficiency. This intermittent learning strategy can ensure every filter maintain context awareness owing to the restriction of keyfilter, periodically enhancing the context awareness. Additionally, hand crafted and deep features are fused to establish a comprehensive appearance model of the tracked object. Substantial experiments on three challenging UAV benchmarks totally with 213 image sequences have shown that our tracker surpasses the state-of-the-art results, and exhibits a remarkable generality in short-term and long-term UAV tracking tasks as well as a variety of challenging attributes. | - |
dc.language | eng | - |
dc.publisher | IEEE. The Journal's web site is located at http://www.ieee.org/organizations/tab/tmm.html | - |
dc.relation.ispartof | IEEE Transactions on Multimedia | - |
dc.subject | Unmanned aerial vehicles | - |
dc.subject | Correlation | - |
dc.subject | Visualization | - |
dc.subject | Training | - |
dc.subject | Object tracking | - |
dc.title | Intermittent contextual learning for keyfilter-aware uav object tracking using deep convolutional feature | - |
dc.type | Article | - |
dc.identifier.email | Pan, J: jpan@cs.hku.hk | - |
dc.identifier.authority | Pan, J=rp01984 | - |
dc.description.nature | link_to_subscribed_fulltext | - |
dc.identifier.doi | 10.1109/TMM.2020.2990064 | - |
dc.identifier.scopus | eid_2-s2.0-85100248597 | - |
dc.identifier.hkuros | 312100 | - |
dc.identifier.volume | 23 | - |
dc.identifier.spage | 810 | - |
dc.identifier.epage | 822 | - |
dc.identifier.isi | WOS:000613560200020 | - |
dc.publisher.place | United States | - |
dc.identifier.issnl | 1520-9210 | - |