融合显著性与运动信息的相关滤波跟踪算法

Translated title of the contribution: Correlation Filter Based Visual Tracking Integrating Saliency and Motion Cues

Wei Jun Zhang*, Sheng Zhong, Wen Hui Xu, Ying Wu

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Rectangle template is a popular target representation adopted by mainstream visual tracking methods. However, by including some background clutter as part of the target representation, the model is likely to drift away from the target gradually and result in tracking failure, especially in challenging situations such as background clutter, target deformation and complex motions. Meanwhile, motion and saliency cues, which play important roles in distinguishing targets from the background and identifying moving objects in the human vision system, have not been modeled into existing tracking methods. To solve these problems, we propose a foreground probabilistic inference formulation that collects pixel-level observations from different sources, and a unified framework integrating the pixel-level model with a widely used correlation filter based method. A saliency-based observation model is proposed by introducing background prior and a distance-based model, which provides reliable evidence to resolve confusion caused by appearance similarity between targets and the background. By taking advantage of continuity and inertia of both target and camera motion, we discover motion patterns in the spatial domain to distinguish targets from the background, and introduce a pixel-level motion-based observation model. Experiments demonstrate that the proposed method outperforms some of the state-of-the-art methods, and shows better robustness in challenging situations such as background clutter, target deformation and in-plane rotation.

Translated title of the contributionCorrelation Filter Based Visual Tracking Integrating Saliency and Motion Cues
Original languageChinese (Traditional)
Pages (from-to)1572-1588
Number of pages17
JournalZidonghua Xuebao/Acta Automatica Sinica
Volume47
Issue number7
DOIs
StatePublished - Jul 2021

Keywords

  • Correlation filter
  • Motion analysis
  • Pixel-level probabilistic model
  • Saliency detection
  • Visual tracking

ASJC Scopus subject areas

  • Software
  • Control and Systems Engineering
  • Information Systems
  • Computer Graphics and Computer-Aided Design

Fingerprint

Dive into the research topics of 'Correlation Filter Based Visual Tracking Integrating Saliency and Motion Cues'. Together they form a unique fingerprint.

Cite this