Terrain Classification from Aerial Data to Support Ground Vehicle Navigation

Sensory perception for unmanned ground vehicle navigation has received great attention from the robotics community. However, sensors mounted on the vehicle are regularly viewpoint impaired. A vehicle navigating at high speeds in off- road environments may be unable to react to negative obstacles such as large holes and cliffs. One approach to address this problem is to complement the sensing capabilities of an unmanned ground vehicle with overhead data gathered from an aerial source. This paper presents techniques to achieve accurate terrain classification by utilizing high-density, colorized, three- dimensional laser data. We describe methods to extract relevant features from this sensor data in such a way that a learning algorithm can successfully train on a small set of labeled data in order to classify a much larger map and show experimental results. Additionally, we introduce a technique to significantly reduce classification errors through the use of context. Finally, we show how this algorithm can be customized for the intended vehicle’s capabilities in order to create more accurate a priori maps that can then be used for path planning.