Multi-elements Composed Drivable Area Extraction For Unmanned Ground Vehicles In Field Terrain

The obstacles impacting unmanned ground vehicles(UGVs) drivability in structured urban environment are mainly objects above the ground, while multiple environmental elements including ground, positive and negative obstacles, water areas and slope areas need to be considered in field terrain. To achieve safe driving of UGVs in field terrain, the perception system must have the ability to detect all these environmental elements. A 3D laser point cloud data based multi-elements composed drivable area extraction system is proposed for UGVs in this thesis.First, the local environmental 3D point cloud is obtained using the geometric model of Velodyne HDL 32 E laser scanner and the point cloud libray(PCL). A simple calibration method is proposed to calculate the external parameters of the Velodyne HDL 32 E.Second, the local terrain is represented as a polar grid map and the 3D point cloud is projected to certain grid according to the coordinate of the points. Then, multiple environmental elements including ground, positive and negative obstacles, water areas and slope areas are detected. Grid reference ground height is calculated by line fitting the candidate ground points of the grids of the same segment of the polar grid map. Ground and positive obstacles are segmented by calculating the maximum height difference of laser points of the grid and the grid reference ground height. Then, roughness of the ground grids are analysed through the variance of the height of the points of the grid. Negative obstacles are detected by the gap difference of neighbor laser returns in radial direction, while off-road boundary is detected by the height difference of neighbor grids in circumferential direction. Further, an intensity filter is used to detect water areas and slope areas are detected by plane feature extraction through RANSAC method.Last, positive obstacle grids, negative obstacle grids, off-road boundary grids, water area grids and slope area grid are simplified as obstacle grids, and then a connected component labeling based method is adopted to cluster local obstacles. Drivable areas are extracted by expanding the unknown grids between passable grids in the same segment of the polar grid map.The experimental results show that this system which is capable of detecting multiple environmental features, can meet the low speed driving requirements of UGVs in field terrain.