Due to the complex floor circumstances of the outdoors and the greatly changed intensity of light, the visual navigation of a mobile robot is still a problem. A new monocular robot's navigation method based on color image processing was proposed for solving this problem. By use of the image partition of the HSI model and mathematic morphology, the trajectory was picked out and the stop-sign drawn was recognized beforehand from the image. Then the “sampling estimated” method has used to calculate the parameters of navigation. According to the experiment this method is less impacted by light and floor circumstance, less error of aberrancy, better real-time and robustness.