The YUV color space model is used to grayscale rice seedling images with complete table searching method, which removes the influence of illumination intensity and improves the real–time performance of the system. The structural elements of the morphological operation of the images are generated by the guidance of Fourier transform. A method of expansion and corrosion combined with Fourier transform is proposed to extract the contour of each seedling line. The vertical overlooking projection of the seedling sequence image is carried out by the modified inverse projection transform to obtain the actual seedling row position in the field. Then the machine vision navigation system is used to track the train. The results of image processing in different angles of the image show that the accuracy of guiding datum line recognition and extraction of seedling centerline is 95.2% when the recognition error is less than 50 pixels and the angle deviation is less than 6°. It is found that the background segmentation of seedling image and the extraction of centerline of seedling train are well realized under field natural environment.