3D LIDAR- and Camera-Based Terrain Classification Under Different Lighting Conditions

Terrain classification is a fundamental task in outdoor robot navigation to detect and avoid impassable terrain. Camera-based approaches are well-studied and provide good results. A drawback of these approaches, however, is that the quality of the classification varies with the prevailing lighting conditions. 3D laser scanners, on the other hand, are largely illumination-invariant. In this work we present easy to compute features for 3D point clouds using range and intensity values. We compare the classification results obtained using only the laser-based features with the results of camera-based classification and study the influence of different lighting conditions.