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

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 time-consistent foreground objects even for a moving camera and for dynamic scene content, but is limited to indoor scenarios. © 2011 Springer-Verlag.