Vision based methods using deep learning such as CNN to perform terrain recognition (sandy/rocky/grass/marshy) enhanced with implicit quantities information such as the roughness, slipperiness, an important aspect for high-level environment perception.