Unstructured road detection via combining the model‐based and feature‐based methods

Abstract

To achieve robust and accurate unstructured road detection, a model-based and feature-based combinatory method is proposed. This method obtains the road saliency map via the graph-based manifold ranking approach and accurately segments the road region by fitting the road model. First, a road graph is constructed with the superpixel nodes and the road distribution model. Then, the seed query nodes are robustly selected. Second, the manifold rankings with the foreground and background queries are conducted successively. To improve the robustness to false queries, the authors propose using two different ranking approaches to obtain the foreground and background saliency maps, and then they are combined into the road saliency map. In the end, they propose using a binomial model to fit the road with which the accuracy of the detected road region is improved. The performance of the proposed method is illustrated using two datasets in terms of the subjective visual quality and the objective quality measurements. The experimental results demonstrate that the proposed method achieves accurate and robust unstructured road detection even under the influence of water stains, kerbstones and hard light.