Self-supervised road detection from a single image

In this paper, we study the problem of detecting road from a single image without priori knowledge of the general structure or the appearance of road surface. In contrast with the road, the background is more cluttered and heterogeneous, therefore different statistical models can be built for the road and the background based on a roughly selected initial road mask, respectively. Then the self-learned statistical models will re-label each pixel in the input image base on a likelihood ratio classifier. This classification based algorithm can detect difficult urban roads without tracking results from image to image. The performance of the proposed algorithm is evaluated on two benchmark databases. Compared with some state-of-the-art vision-based road detection approaches, the proposed algorithm is simple and efficacious.