Road detection based on shape prior, color representation and discrimination analysis

Vision based road detection is a key technique in Autonomous driving and Advance Drive Assistance System. We propose an approach for vision-based road detection which exploits road shape prior, color consistence representation and discrimination analysis. Firstly, road shape prior probability map is constructed from available road images which have been annotated by matching and classifying of road shape feature. Secondly, based on the road shape prior probability map, for one thing, road samples are trained to obtain the best combining coefficient of multiple color channels, so as to obtain consistent representation probability map based on minimal variance criteria, for another, road and off-road samples are extracted for discrimination analysis to obtain discrimination probability map. Finally, three maps are combined to obtain road detection result. Experiments on both public and our own database demonstrate the effectiveness of the proposed method under complex illumination conditions.