An Adaptive Model–Based Road Detection and Retrivel Using Query by Shadowless Features

Road detection is an essential functionality for autonomous driving. The key of vision–based road detection algorithms is the ability of classifying image pixels as belonging or not to the road surface. In this paper, the propose an adaptive color– based road detection algorithm which combines a physics–based illuminant–invariant color space with a model–based classifier in a frame by frame framework using a monocular camera. The novelty of our approach resides in using Query by shadowless features to characterize road pixels. Besides, the road model is built on–line and dynamically updated based on feedback from the current detection and predictions of a Markov model. Experiments are conducted on different road sequences including different scenarios, different weather conditions, extreme shadows and the presence of other vehicles. Qualitative results validate the proposal for reliable road detection.