Road Curb Detection using Traversable Ground Segmentation: Application to Autonomous Shuttle Vehicle Navigation

This work addresses the problem of road curb detection using a LIDAR 3D sensor in urban environments, which is an important task in autonomous vehicle technology. Existing road curb detection methods often rely on a ground segmentation process and geometric features extraction on ground data. Existing ground segmentation methods, used for road curb detection purposes, have been developed with the flat terrain assumption in mind. However, in real applications, the shape of the terrain is not always flat. Moreover, the surrounding environment (walls, trees, and other obstacles) and the sensor orientation might influence the results of the existing ground segmentation methods. Thus, we propose an efficient ground segmentation algorithm to extract points belonging to the road surface, road curbs, sidewalks, etc. Geometrical features are used to detect road curb like changes in LIDAR 3D points. However, geometrical features cannot distinguish between road curbs, small obstacles and, terrain variations whose shape is similar to road curbs. Then, in addition to geometrical features commonly used in curb detection methods, we use the LIDAR 3D reflectance feature to detect road curbs. Experiments on an autonomous shuttle vehicle demonstrate that the proposed method is capable of detecting road curbs on roads with complicated geometry such as curved roads, roundabouts, and banked turns.