In order to improve the effect of a background model in traffic video surveillance which was not good for the complex environment with too many foreground objects or light varying, a background modeling of multi-level block was proposed. The model was based on the frame differential method with the N frames interval and multilevel block, combining with center-symmetric local binary pattern and codebook algorithm. Using the model, the background obtained is clear and unbroken, and is the base for the foreground object extraction. To test the validity of the method, the designed experiment was compared with local binary pattern, center-symmetric local binary pattern, codebook algorithm and mixture of Gaussian. The proposed model got the more complete foreground objects, more clearly boundary of the objects, and no significant block figure. We scored the methods and the proposed method got higher scores. In the foreground object extraction methods, the method we proposed had the better results.