Based on the analysis of some feature videos of traffic scene

the values of background pixels are obtained according to variety of pixel colors in video flows. A modeling method of estimating background pixels using color grade of different frames is proposed. The image is divided into some sub-regions in computing

and every sub-region background is gotten according to variety of color grade

then the background image is rebuilt according to the computing results of every sub-region. Experimental results show that the process of background modeling can be completed after computing average initial 100 frames

which is less than that of other methods

and the average computing time of every frame is 15.566 ms. The speed of modeling and the number of initial frames can meet the requirements of the video surveillance system in traffic scene using this method.