In order to quickly and accurately identify the guidance line in complex environment, a line recognition algorithm based on prior knowledge was proposed. Images were divided into several regions by field range and line characteristic. Both means of full image and region were used to make adaptive threshold of each region. Scanning mode was decided by information of previous image, then noises were filtered and intermediate points were acquired. An improved Hough transformation with search angle limitation and nonuniform sampling was applied to localize guidance line. Finally cross ratio theorem was used to calibrate the parameters. The experimental result showed the method was characterized by robustness and real-time, the average computing time was about 90ms for each image.