Background updating in illumination-variant scenes

In this paper, after a brief overview of existing methods, we present a new pixelwise background subtraction algorithm based on color image. This algorithm can compensate for fast and low variation of illumination by its smoothness over space and time, and therefore the detection of moving objects is more robust. We first describe the background model by analyzing the relationship among background, moving objects, illumination and noise produced by camera. Then Gray World Assumption is used to remove the global changing of illumination. Third, a new background updating algorithm is presented, which combines the variation of neighboring area with the difference between the current and previous values in order to predict the new values of the pixels on background efficiently. Experimental results on real scenes have substantiated the effectiveness of the proposed method.