Abstract—The background estimation is often required in video surveillance applications. This paper presents some new techniques for background estimation in very high traffic conditions, where the background is visible for a very small fraction of the time. The algorithm exploits the fact that regions containing the background (assumed stationary) in the video sequence would always show a stable content, as opposed to the moving foreground which is highly transient, and presents an approach to eliminate the frames containing the foreground. Moreover, our proposed formulation of block wise stability provides a better estimate than a pixel wise approach. Results show that the obtained background comes very close to the actual background.

Index Terms—background estimation, requantization, spatial consistency, temporal consistency

Cite: Bhumi Sabarwal, Priya Singh, and K S Venkatesh, "A Robust Technique for Background Estimation in Heavily Intruded Videos," Journal of Image and Graphics, Vol. 2, No. 1, pp. 46-53, June 2014. doi: 10.12720/joig.2.1.46-53