A Traversability Detection Method Dealing with Shaded Terrains

Traversability detection is one of the basic methods of vision navigation for autonomous robot. Shaded terrains are ever-present in real road and urban environment, however, the current research work related to the solutions of shaded terrains is very limited. A traversability detection method dealing with shaded terrains is proposed. Firstly, utilizing SLIC superpixel algorithm, the scene image is divided into irregular image patches (superpixels) in which the pixels are homogenous and at the same illumination level. Secondly, a novel positioning method for feature extraction window is proposed to decide the location of feature extraction window which contains as much pixels which are homogenous and at the same illumination level as possible. Then, the local binary pattern (LBP) extraction method is exploited to describe the patches. Finally, One-Class SVM (one-class support vector machine) algorithm is used to run training and classification. Results show that the proposed method performs well not only in those areas where the illumination level of its neighborhood pixels are almost at the same illumination level, but also in the intersection of unshaded area and shaded area. The proposed algorithm can even achieve good results in those regions of mottled shades, and eliminates the disconnectedness of shade boundaries.