Efficient nonparametric background modeling of dynamic scenes

Background modeling is one of the most important parts of visual surveillance systems. In the video sequences only small parts of scene show non-stationary characteristics but these parts account for significant portion of false alarms. So they need to be modeled more accurately. We use this fact to propose an improved blockbased algorithm for background modeling. Each block is modeled by a group of weighted histograms. The number of histograms for each block changes adaptively with complexity of that block. False alarm reduction is a direct result of our proposed method. The experimental results demonstrate the effectiveness of the algorithm when applied in different outdoor and indoor environments.