Background subtraction: Model-sharing strategy based on temporal variation analysis

Abstract

This paper presents a new approach for moving detection in complex scenes. Different with previous methods which compare a pixel with its own model and make the model more complex, we take an iterative model-sharing strategy as the process of foreground decision. The current pixel is not only compared with its own model, but may also compared with other pixel’s model which has similar temporal variation. Experiments show that the proposed approach leads to a lower false positive rate and higher precision. It has a better performance when compared with traditional approach.