Time-consistent foreground segmentation of dynamic content from color and depth video

  • Authors:
  • Anatol Frick;Markus Franke;Reinhard Koch

  • Affiliations:
  • Christian-Albrechts-University Kiel, Computer Science Department, Kiel, Germany;Christian-Albrechts-University Kiel, Computer Science Department, Kiel, Germany;Christian-Albrechts-University Kiel, Computer Science Department, Kiel, Germany

  • Venue:
  • DAGM'11 Proceedings of the 33rd international conference on Pattern recognition
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper introduces an approach for automatic foreground extraction from videos utilizing depth information from time of flight(ToF) cameras.We give a clear definition of background and foreground based on 3D scene geometry and provide means of foreground extraction based on one-dimensional histograms in 3D space. Further a refinement step based on hierarchical grab-cut segmentation in a video volume with incorporated time constraints is proposed. Our approach is able to extract timeconsistent foreground objects even for a moving camera and for dynamic scene content, but is limited to indoor scenarios.