Stereovision-based side lane and guardrail detection

This paper presents a fast and robust side lane and guardrail detection technique, based on the 3D information obtained from stereovision and on the already detected geometry of the current lane. The 3D points that belong to the road surface are selected using the road surface parameters inferred from the current lane detection, and they are used for detection of the side lane. The 3D points above the road are used in the detection of guardrails and fences, with the same technique. In both cases, the point sets are smoothed in an attempt to compensate for the variable density, and then analyzed through lateral distance histograms. The peaks of the histograms locate the position of the side lane delimiters and fences, while the shape of the histogram is used for the validation of the side lane or fence's existence. The results are then tracked for increased stability