Abstract: Abstract: To make up the shortages of the existing algorithms for the visual navigation such as the noise interference and the processing speed, a new algorithm for navigation line detection was designed in this article. In the first stage, the image preprocessing was carried out. Firstly, the 2G-R-B method was used to convert the color images into grey scale images in order to distinguish crop and soil better. In general, the green component G is far greater than red R and blue B component for the crops of which main pigment is chlorophyll. The 2G-R-B method was used to graying images in order to emphasize green component and restrain the rest two components. Secondly, the OTSU method was used to transfer the grey scale images into binary images. Because the brightness and color will be different if the image under different light condition, the OTSU method was taken to transform the gray images into binary images. In addition, the weeds, shadow is inevitable because of the complexity of the farmland environment. Therefore, there are a lot of noise points in the image after binarization. In order to get ideal effect, the noise in the image needs to be reduced. In this paper, the corrosion-median filter-expansion was selected to reduce the noise according to the characteristics of the weeds. In the next stage, the designed algorithm named Scan Filtering (SF) navigation line extracting algorithm was used to get the navigation line from the above processed binary image. In SF algorithm, the image was divided into the right sector and the left sector. These two sectors were scanned by a series of triangles with the same area; then, the number of white dots was counted in every triangle and stored in an array. The array was filtered by an IIR filter, and the navigation line was extracted according to the output of the IIR filter. Finally, the parameter of navigation was transformed to the horizontal deviation and the heading deviation by the relative position and orientation algorithm. Three experiments were designed to evaluate the performance of SF algorithm, including compare it with the traditional navigation line detection method such as Hough transformation. The first experiment was designed to test the performance of the image preprocessing methods which included 2G-R-B, OTSU and corrosion-median filter-expansion method. The second experiment was designed to detect the performance of SF algorithm, includes time consuming and accuracy analysis. The experimental results showed that only 76ms were needed to process a 640×320 sized picture, and this algorithm could meet the need of agricultural real-time visual navigation. Through comparing with Hough transform and random navigation line detection algorithm, the results showed that this new method could extract the navigation lines faster and more accurately. The last experiment was design to analyze the stability of SF algorithm. Under different circumstance such as plant lacked and high density weed, winter wheat and corn were selected to test the adaptability of SF algorithm. The experiment results showed that SF algorithm could adapt different environment very well.