Real-Time System for Video Background Subtraction

Accurate detection of moving objects is an important precursor to stable tracking or recognition. The algorithm used here aims to segment the foreground objects in live video when background scene textures change over time. To accommodate interactions among neighboring pixels, a global online learning algorithm is then derived that explicitly distinguishes objects versus background. The average deviation method is used in order to subtract the object from the live video. Formulate background subtraction as minimizing a penalized instantaneous risk functional yielding a local online discriminative algorithm that can quickly adapt to temporal changes. In the context of background subtraction, we achieve better, more accurate segmentation than competing methods using a model whose complexity grows with the underlying complexity of the scene, rather than the amount of time required to observe all aspect of textures. By proposed method, develop an implementation that can run efficiently on the highly parallel graphics processing unit, thus yielding good processing speed.