.png){kind=icon}
Return
Structural local sparse and low-rank tracker using deep features
.png){kind=icon}
PRE
AI
DOI：10.1007/s00530-023-01064-3.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Visual trackers based on the sparse representation and low-rank constraints imposed on candidate targets have achieved commendable results. However, combining low-rank constraint imposed on local patches with local sparsity has not been explored for designing visual tracker yet. In this work, we propose a novel deep features-based structural local sparse low-rank tracker, which not only exploits sparsity and low-rank constraint about local patches, but also takes the spatial structure of the target regions into account. Specifically, according to the magnitude of change of targets' appearance, all local patches are divided into stable and unstable ones since their contribution to tracking may be sharply different. Then, elegant pre-locating and pruning schemes are developed to maintain the proposed tracker's performance even in the very challenging scenarios and meanwhile decrease the heavy computational burden caused by large number of candidate targets. Moreover, an upgraded effective template update scheme is designed by utilizing long-term and short-term memory of different templates to adapt to changes of the target's appearance. Comprehensive experiments on OTB100, TC128, VOT2016, VOT2018 and GOT-10k benchmark datasets demonstrate that the proposed method outperforms several popular handcrafted features-based trackers and is comparable to state-of-the-art deep trackers in various tracking environments.
Keywords:
Visual tracking
Sparse representation
Low-rank representation
Deep features
Journal
IF:
3.1
Papers:
2.8K
Citations:
2.7K
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveShareSave
View more