For the traditional Gaussian mixture model, it is prone to detect the background as the foreground region and has poor effect in complex background, so this paper puts forward an improved algorithm combined with three frame difference method. First, we use the three frame difference method to improve adaptation of the revealed background region. Second, we use the means of background neighborhoods update to improve the resistance to complex background noise. The experiments prove that compared with the traditional method, the improved algorithm reduces large amount of noise and shortens learning cycle, and improves resistance to the weather and camera shake, optimizing the noise caused by background region revealed.