Illuminant-invariant model-based road segmentation

Abstract

Road segmentation is an essential functionality for supporting advanced driver assistance systems (ADAS) such as road following or vehicle detection and tracking. Significant efforts have been made in order to solve this task using vision-based techniques. One of the major challenges of these techniques is dealing with lighting variations, especially shadows. Many of the approaches presented within this field use ad-hoc mechanisms applied after an initial segmentation to recover shadowed road patches. In this paper, we present an innovative method to obtain a road segmentation algorithm robust to extreme shadow conditions. The novelty of the proposal is the use of a shadowless feature space in combination with a model-based region growing algorithm. The former projects the color images such that the shadow effect is greatly attenuated. The latter uses histogram models to label the pixels as belonging to the road or to the background. These models are constructed on a frame by frame basis independently of the road shape to avoid limitations when addressing unstructured roads. The results presented show the validity of our approach.