An improved crop rows detection presented based-Hough transform. In the paper, the two argument r, q of peaks in Hough space were mapped into the Cartesian coordinate system to extract straight line segments fitting the crop rows, in which the template matching was firstly designed to filter the noises from isolated, or scattering points, then the weighted averaging were employed to eliminate line segments family due to the peaks and pseudo-peaks in Hough space. Experiment results showed that the novel crop rows recognition method could extract perfect crop row line segments from real image with strong noises or disturbance, and it was real-time, robust against great area grass, loss of crop row, or lack of crop row information in real image, with high accuracy and availability.