A model of weeding robot, machine vision-based weed detection and navigation software in VC++ were developed. This software system could lead the weeding robot walk along crop rows and recognize weeds between rows simultaneously. A new image segmentation algorithm was proposed subsequently. Crops could be segmented from background directly in RGB space, and then the center lines of crop rows were identified by Hough transform. The robot poses were obtained according to the camera pose and perspective principle. Experiments showed that a colorful image (resolution: 1 536×1 152) could be segmented within 450 ms, and the new algorithm could also adapt the lightness variation. Optimized Hough transform algorithm reduced the segmentation time by 1/2. The average error of navigation distance and navigation angle was -0.6 mm and 0° respectively. The results showed that the navigation method was feasible. 