Road detection via superpixels and interactive image segmentation

Although vision based road detection has been extensively studied in the past decades, road detection in adverse conditions still remains challenging. In this paper, we propose a road detection approach via superpixels and an automated version of interactive image segmentation. We first segment the input road image into superpixles, and we design a novel seed selection method based on multiple novel cues extracted from a single frame to correctly select road and non-road seeds. Then maximum similarity based interactive image segmentation is applied to detect road regions with the selected seeds. Our method is free of models and no temporal information is used. Experimental evaluations with state-of-the-art algorithms on public road datasets demonstrate the merits of the proposed algorithm.