Robust visual tracking using discriminative stable regions and K-means clustering

  • Authors:
  • Can-Long Zhang;Zhong-Liang Jing;Han Pan;Bo Jin;Zhi-Xin Li

  • Affiliations:
  • School of Aeronautics & Astronautics, Shanghai Jiao Tong University, Shanghai 200240, China and College of Computer Science & Information Technology, Guangxi Normal University, Guilin 541004, Chin ...;School of Aeronautics & Astronautics, Shanghai Jiao Tong University, Shanghai 200240, China;School of Aeronautics & Astronautics, Shanghai Jiao Tong University, Shanghai 200240, China;School of Aeronautics & Astronautics, Shanghai Jiao Tong University, Shanghai 200240, China;College of Computer Science & Information Technology, Guangxi Normal University, Guilin 541004, China

  • Venue:
  • Neurocomputing
  • Year:
  • 2013

Quantified Score

Hi-index 0.01

Visualization

Abstract

This paper presents a method of extracting discriminative stable regions (DSRs) from image, and applies them for object tracking. These DSRs obtained by using the criterion of maximal entropy and spatial discrimination present high appearance stability and strong spatial discriminative power, which enables them to tolerate more appearance variations and to effectively resist spatial distracters. Meanwhile, the adaptive fusion tracking incorporated k-means clustering can handle severe occlusion as well as disturbance of motion noise during target localization. In addition, an effective local update scheme is designed to adapt to the object change for ensuring the tracking robustness. Experiments are carried out on several challenging sequences and results show that our method performs well in terms of object tracking, even in the presence of occlusion, deformation, illumination change, moving camera and spatial distracter.