In order to meet the requirements of autonomous navigation and walking in field management such as intertillage and topdressing in maize seedling stage, the real-time detection technology of multiple crop row lines based on machine vision was studied. First of all, based on green component enhancement method, improved Otsu algorithm of segmentation threshold optimization, variable threshold denoising method and morphological denoising method, pre-processing such as grayscale, binarization and denoising was carried out. The pre-processing was not affected by natural light changes, shadows, precipitation/ponding and planting pattern, which had a better cleaning effect on the inter-row space of crops under the condition of plant canopy overlap or the interference of finely fragmentary weeds, and a better removal effect on the noise of small size, round leaf weed with scattered distribution between crop rows and weed in the form of transverse growth or aggregation. And then, the binary image was divided into 20 horizontal bars along the ordinate direction. The characteristic parameters of horizontal spacing and horizontal span were established for target areas inside each horizontal bar, and the characteristic parameters of vertical spacing, trend angle and coverage width were established between target areas across horizontal bars. Based on the distribution difference of the above parameters between target areas within crop rows and ones across crop rows, the localization and segmentation of target areas belonging to different crop rows in each horizontal bar and the clustering of target areas belonging to the same crop row in different horizontal bars were completed, and good segmentation and clustering results was obtained. Finally, the centerlines of each crop row in current frame were obtained by linear fitting based on the least square method after removing outlier feature points. The test results showed that the overall detection accuracy was no less than 91.2%, and the single-frame image processing time was no more than 368ms, which can quickly realize the synchronous detection of multiple crop row lines under the interference of different environmental factors.