Combining discriminative and descriptive models for tracking

  • Authors:
  • Jing Zhang;Duowen Chen;Ming Tang

  • Affiliations:
  • National Lab of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China;National Lab of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China;National Lab of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China

  • Venue:
  • ACCV'09 Proceedings of the 9th Asian conference on Computer Vision - Volume Part I
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, visual tracking is treated as an object/back-ground classification problem. Multi-scale image patches are sampled to represent object and local background. A pair of binary and one-class support vector classifiers (SVC) are trained in every scale to model the object and background discriminatively and descriptively. Then a cascade structure is designed to combine SVCs in all scales. Incremental and decremental learning schemes for updating SVCs are used to adapt the environment variation, as well as to keep away from the classic problem of model drift. Two criteria are originally proposed to quantitatively evaluate the performance of tracking algorithms against model drift. Experimental results show superior accuracy and stability of our method to several state-of-the-art approaches.