A vision based road detection algorithm

Road detection is one of the basic tasks for automatic guidance. In this paper, a new approach to detect the road is proposed, which firstly segments the images into three classes of regions (road, non-road, uncertain regions) by some features and rules described in the paper, then uses hypothesis and verification strategy to amalgamate the uncertain regions to road or non-road correctly. The algorithm has been applied in an ATRV (all-terrain vehicle) made by Irobot Company, and experiments show that it is fast and robust for road detection at a park on a university campus.