To make up the shortages of the existing vision navigation algorithms such as the noise interference and the processing speed, a new approach for navigation line detection is presented in this article. This new method was based on image boundary detection. The color images were transformed into binary ones after pre-processing. Then, three stages in this method were presented. At first, the image was divided into several sectors and the boundary of the crops was detected. Then a number of points indicating the centers of the rows were determined. At last, the navigation line was estimated by an improved linear regression algorithm. Compared to other navigation line detection algorithms, this new method could find the navigation lines more accurately. It need only about 89ms to process a picture with 640×320. So it can meet the need of agricultural real-time visual navigation when the speed is 1~2m/s.