The vision-guided AGV acquires images of road through machine vision and the navigation parameters are gotten by processing these images. In order to recognize the road robustly and quickly, some image-processing methods such as filtering, image segmenting were presented. Additionally, some measures were taken to keep the recognition in real time. A PID controller has been designed to control the motors in order to achieve the line-tracking according to the parameters. The simulation and experimental results showed that the road could be recognized even under certain disturbances and the AGV could track the road accurately and reliably.