150ms。To improve real-time performance of agriculture vehicle navigation, an algorithm based on least square method for detection of crop rows, especially of crops in the early stage of growth was proposed. Crops were segmented from background by the index of G-R. Parts of the noises of weeds in the image were eliminated according to their length. Crop line numbers were detected dynamically by vertical projection method. Center points of crop rows were extracted as feature points and were classified into different clusters. Least square method was used twice for fitting the center line of crop row to the feature points. The number of the feature points was counted to judge the reliability of the detection result of crop row with plant deficiency. The experimental result showed that the algorithm could overcome the effect of weed noise and plant deficiency. The average time of image processing was less than 150ms.