Road Boundary Detection and Tracking using monochrome camera images

In this paper a method for Road Boundary Detection and Tracking on monochrome camera images is presented. Using this approach, it is possible to recognize road boundaries in areas without lane markings. To describe the shape of the boundary a clothoid road model is used. The parameters of this model are estimated and tracked by a Particle Filter. The implemented Particle Filter is based on the well known Condensation Algorithm. Only texture features are used for road recognition. For evaluation of these features a patch-based classifier is used. Based on a set of sequences the probability distribution of the classifier output is empirically determined. So the single patches can be assigned to “road” or “non-road” class with a certain probability. The presented algorithm shows good performance with reliable results on a variety of recorded video sequences.