Flame foreground extraction is an important step in video-based fire detection algorithms, and it is the basis of following flame feature recognition algorithms. Given the fact that existing flame foreground extraction algorithms cannot extract flame foreground accurately under the interference of intense light or when the background color is similar to the flame, we propose a new algorithm to extract fire foreground. Firstly, the algorithm determines the primary suspected region by calculating the momentary motion region and the fire colored region. Secondly, we obtain the moving foreground region by different background subtraction methods according to the primary suspected region. Finally, we get the fire foreground region according to the moving foreground region and the fire colored region. Experimental results show that compared with four existing flame foreground extraction methods, the proposed algorithm has a much higher accuracy rate (100%). The proposed method can adapt to different complicated backgrounds and meet real-time requirement.