Positioning of rice seedlings is an important step of mechanical intrarow weeding. Aiming to minimize the positioning errors based on canopy of rice, a system that could capture RGB images of rice stem base on the side view was built to position the rice seedlings according to edge fitting in separated region of the stem base. Firstly, relevant factors, including the morphology of rice seedlings, the installation parameters of camera on the sideview and affecting image quality were analyzed, and a method of grabbing image under shading conditions was proposed, contributing to a system with shading device. The original RGB images were distinguished from soil by analyzing the excessive green (2G-R-B) rice seedlings index images. The Otsu algorithm method was employed to transform a gray image to a binary image, and then the binary images were smoothed by morphological operation to remove noise for subsequent analysis. After that the edges of stem base were extracted by the Canny operator, and the fitting region of edges was set up. A method was presented in fitting region based on edge fitting in nonseparated region of the stem base. However, the results showed that this positioning method was not accurate because of effect of unsmooth and unilateral sloping edges of stem base. Therefore, the fitting region was separated into three subintervals where the edges of stem base were fitted by the Hough transform. The center of rice seedlings was gained by the center coordinates of the midline fitted by the mean of midpoint coordinates of fitting edges. The positioning experiments of rice seedlings in terms of stem base and the canopy were carried out respectively. The results of positioning based on stem base showed that the positioning errors were less than 7.0mm, and the positioning errors were 6.9mm, 5.9mm and 6.3mm under the condition of including water and 6.8mm, 5.8mm and 6.5mm under the anhydrous condition on the 10th, 15th and 20th day after transplanting, respectively. The results of positioning based on canopy showed that the positioning errors were above 8.0mm, and the positioning errors were 8.2mm, 11.3mm and 15.4mm under the condition of including water and 9.0mm, 10.7mm and 16.1mm under the anhydrous condition on the 10th, 15th and 20th day after transplanting, respectively. The results suggested that the innovative positioning method that based on edge fitting in separated region of the stem base of rice seedlings met the technical requirement of mechanical weeding in terms of high precision and long applicable period.