Stereo-based Terrain Traversability Estimation using Surface Normals

Abstract:
A stereo-based terrain classification for traversability estimation of all terrains in offroad mobile robots is presented. The proposed method defines the roughness of the surrounding terrain for every single pixel in the image or point in a point cloud using surface normals and also explains how this is applied to all terrain by knowing the kinematics capability, mechanical constraints and size of the UGV. Roughness estimation using surface normals helps us model the whole terrain with a virtual function which increases the accuracy of the resulting terrain travesability map considerably.