Static region detection using background subtraction and a pixel level analysis in single static camera environments

Abstract

In this paper, we propose a fast static region detection method, which is based on background subtraction in a single static camera. A background image is constructed by employing a Gaussian mixture model, which is then updated by applying an online Bayesian method. By observing the pixel status at different times, two time-different foregrounds are generated. Connected component labeling is processed on two foregrounds, with each labeled blob being analyzed to determine if it is a static region or not. The detected static regions are then classified as either an abandoned or removed status by edge analysis. The proposed method uses only one background model, which provides the beneficial effect of the dual background method with a minimum difference in performance. In addition, the proposed method can reduce considerable computational overhead costs compared with a previous method that used a dual background approach.