Abstract: We have developed an efficient method for localizing a mobile robot based on a color image. The color image is transformed from RGB space into HSI space. The color becomes the unique feature in the image. It simplifies feature extraction in images and improves greatly the image processing speed and the robustness of extracting features. In structure, the adoption of global static vision system avoids controlling the camera. Furthermore, a single colored cue in an environment is enough for localizing the mobile robot. During the motion of the mobile robot, the processing data is further reduced by using dynamic window. Extended Kalman filter is employed to fuse the visual data and the dead-reckoning data considering the time delay caused by image processing. Experiment results exhibit the validity and the effectiveness of the proposed method.