This paper performs a vision-based lane tracking system for a variable-speed autonomous mobile robot. A camera is set on the mobile robot as the sensor to get real-time road images. The lane marks in images are extracted by real-time image processing algorithm. We design fuzzy controller according to the information of the lane marks and steering angle. Then the autonomous mobile robot moves following the lane marks. We apply the four-point rolling grey modeling GM(1, 1) to prediction of the lane position and confirm whether the autonomous mobile robot is in the sharp curve of a road. Afterward we slow down car speed in the curve road. Finally experimental results show the effectiveness of the proposed lane tracking and different car speed system.