Real Time Foreground Segmentation for Video Sequences with Dynamic Background

Modeling background and segmentation of foreground objects are techniques required for various video surveillance and related applications. We propose an intensity and colour based background modeling technique for pixel level foreground segmentation of video sequences with dynamic background. It models the background using initial few frames called training frames and calculates a threshold using statistical analysis on these frames. As the new frames appear in real time, the value of threshold is updated and the background model adapts to the changing background with each incoming frame. Quantitative results suggest that the proposed method can handle the sequences with highly dynamic background.