Reliable real-time video background subtraction: mixture of Gaussians approach

Abstract

Background subtraction is a commonly used class of techniques for segmenting out objects of interest in the context of static camera for applications such as surveillance. Trivial background subtraction algorithms (e.g. frame differencing, median filter) can perform quite fast but they are not robust enough to be used in various computer vision problems. The main purpose of this paper is to present common background subtraction flow, describe Mixture of Gaussians model and discuss experimental results.