A simple vision system for lane keeping

We present a system for extracting lane marker information from image data in a form suitable for use in the automated steering of moving vehicles. The algorithm was designed with speed and simplicity in mind. It assumes a flat, dark roadway with light colored lane markers, either solid or dashed, painted on it. The algorithm is based on extracting statistically significant `bright´ regions from the image plane and fitting functions to those identified points which are consistent with the qualitative properties of roadway lanes and ground to image plane geometric relationships. It estimates a virtual lane centerline using left and right lane markers if both are available, otherwise it uses whichever lane marker is visible and a rough estimate of the lane width. The system has been tested in various autonomous vehicle control research projects, culminating in its demonstration on Ohio State University´s automated vehicles at the 1997 AHS Technical Feasibility Demonstration in San Diego