Video-based recognition of unmarked lanes via texture interpretation and N-level-set-fitting

In modern driver assistance systems the environment perception plays a decisive role in order to evaluate the current traffic scene. The reliable recognition of the drivable area provides essential information for lane departure warning systems which in turn contribute to active road safety. Most systems on lane recognition do reliable work on well marked roads and under good weather and lighting conditions but they fail in more difficult scenarios especially in the absence of lane markings. In this paper we present a two-stage algorithm for the recognition of unmarked lanes using only a single monochrome camera. In the first step our approach transforms texture differences into gray value information in order to intensify the separation of on-road and off-road parts. Then a N-level-set-fitting is applied on this new image which uses a clothoid road model as underlying set model. Minimizing the error function yields to the parameter estimation of the road model. The method is independent of the availability of lane markings or any other landmarks. The parameter propagation to the next frame is done via a standard kalman filter. The given results show the performance of our approach in robust lane recognition under various weather and lighting conditions.