In the study of vehicle safety driving assistance system,one of the most important research is recognition of lanes on complex roads . This paper is based on theories of video image processing technology which divided close-range vision for the region and the regional after preprocessing the video image of the front. The technology use the improved method of Hough detecting close area lane and removing the double edge and pseudo-edge, which get lanes through linear fitting of close range area and curve fitting method to prospect area. It builds virtual lane model based on the prediction algorithm in fuzzy and bad visual conditions. We verified the correctness; effectiveness of this method through the experiment, this method has wide applications in the future.