Background extraction is a fundamental problem in video surveillance and motion tracking. Considering actual conditions in traffic video surveillance where image brightness varies over time and crossroad vehicle occupancy often keeps high

a new background extract method is put forward based on textural and statistical features. First

textural description of the image is established with grayscale-primitive cooccurence matrix. Then the textural feature is used to roughly evaluate the presence of moving targets in a region. In the next step

background pixels are determined according to the mixed Gaussian distribution model

by which the background of video sequence is extracted. Finally

the multi-resolution method is introduced to improve the computational efficiency. Experimental results show that the proposed method allows various changes of lighting conditions

and extracts exact background scenes even in vehicle-packed crossroad with only 1/4 computation period of the original method. Therefore

the method can better satisfy the requirements of background extraction algorithm

which needs to be robust and automatically adapt to complex environment.