Robust Context-Aware Tracking with Temporal Regularization

Tianhao Li, Tingfa Xu*, Yu Bai, Axin Fan, Ruoling Yang

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Discriminative Correlation Filters demonstrate superior capabilities, while still suffering from background clutter. The proposed context-aware correlation filter (CACF) framework effectively avoids the interference of background noise with the explicit incorporation of global context information. However, there is still sequential context information that is not considered. This work proposes a robust context-aware tracking based on hand-crafted features by adding a temporal regularization. The temporal regularization term provides temporal information for learning filter, which limits the mutation of the filter. Experiments on OTB-100 show that our tracker demonstrates excellent accuracy and significantly improves the robustness of CF trackers and those trackers in the CACF framework.

Original languageEnglish
Title of host publicationCommunications, Signal Processing, and Systems - Proceedings of the 8th International Conference on Communications, Signal Processing, and Systems, CSPS 2019
EditorsQilian Liang, Wei Wang, Xin Liu, Zhenyu Na, Min Jia, Baoju Zhang
PublisherSpringer
Pages858-865
Number of pages8
ISBN (Print)9789811394089
DOIs
Publication statusPublished - 2020
Event8th International Conference on Communications, Signal Processing, and Systems, CSPS 2019 - Urumqi, China
Duration: 20 Jul 201922 Jul 2019

Publication series

NameLecture Notes in Electrical Engineering
Volume571 LNEE
ISSN (Print)1876-1100
ISSN (Electronic)1876-1119

Conference

Conference8th International Conference on Communications, Signal Processing, and Systems, CSPS 2019
Country/TerritoryChina
CityUrumqi
Period20/07/1922/07/19

Keywords

  • Context-aware tracking
  • Correlation filter
  • Temporal regularization

Fingerprint

Dive into the research topics of 'Robust Context-Aware Tracking with Temporal Regularization'. Together they form a unique fingerprint.

Cite this