In order to extract rice seedling rows accurately, a detection method of centerlines of rice seedling row based on sub-regional feature points clustering was proposed. 2G-R-B characteristic factor and Otsu method were used to separate seedling and background from RGB rice seedling image. By sub-regional analyzing the distribution of seedling pixels, candidate feature points of seedling row were extracted. Then feature points were clustered with the nearest neighbor relationship between feature points, and the number of seedling rows and starting points of each seedling row were determined. According to the characteristics of row planting of seedlings, trend line was introduced to refine feature points. The real feature points indicating seedling rows were obtained by comparing the shortest distance of candidate point with its corresponding trend line with a distance threshold value. Afterwards, the centerlines were detected by fitting a straight line with the least square method. The experimental results showed that the proposed method achieved good anti-noise performance. The accuracy of centerlines detection was 95.6%, but the traditional Hough method and the randomized Hough method can only reach 84.1% and 89.9%, respectively. The average processing time of a 320 pixels×237 pixels color image was less than that of the two other algorithms. It can be seen that the proposed algorithm had the advantages of high real time and high accuracy, which can accurately extract seedling row centerlines, and the research result provided navigation parameters for an automatic rice transplanter walking along the seedling row in paddy fields.