A fast forest road segmentation for real-time robot self-navigation

Road-following by mobile robots under varying outdoor illumination demands special care to be taken on development of an algorithm, which should be flexible to color changes in sunny or shadow parts. This paper addresses the technical feasibility of an automatic color evaluation method for fast segmentation of roads with unknown geometrical shapes. Instead of using multiple image features, we describe a selective color-based road extraction method, where appropriate color parameters are projected to a one dimensional feature. We also report on mission experiments of a small 4-wheel mobile robot equipped with a color CCD camera sensor through a real forest field.