A robust and stable road model

Abstract

In this paper a robust and stable model for road detection and road following in traffic scenes is proposed. Scene analysis consists of an initial phase, in which objects are recognized for the first time, and a phase of object tracking with simultaneous object recognition. Road surface analysis is done by means of an explicit model for the road shape that is adjusted stepwise to image regions. In the tracking phase the adjusted model is used for robust model-based segmentation of the road surface. For a reliable analysis of scenes containing obstacles, the system is equipped with a look-ahead mechanism.