Drivable Area Extraction From Multiple Aliened Frames Of Point Cloud

Environment perception systems are important parts of unmanned ground vehicles(UGVs)and are the first priorities to the safety of UGVs.While drivable area extraction is the foundation to other perception systems.A drivable area extraction framework and its method based on aligned multi frames of point cloud produced by the Velodyne HDL-32E Lidar are proposed in this study.The study include several aspects as listed below.1)The framework for drivable area extraction from multi frames of point clouds is proposed,with some brief introduction of the Lidar sensor Velodyne HDL-32E used in this study.Drivable area is first extracted from every single frame of point cloud in a sequence.Then the sequence of point clouds is aligned by introducing POS data or by utilizing SLAM algorithms.Drivable area are finally extracted from aligned point cloud based on the result from single frame extraction results.2)A method for drivable area extraction from single frame of point cloud is proposed based on the angles formed by consecutive laser points in a vertical section from the single frame of point cloud.3)POS data based and SLAM based point cloud aligning method are studied.A principal component analysis(PCA)based point cloud aligning result evaluation method is proposed for the specific application of Lidar point cloud.4)A method for drivable area extraction from aligned multi frames of point cloud is proposed.Repeated points or points so close to each other are first removed from the aligned single frame detection result.An octree is then build for aligned frames of raw point cloud.Finally,false detections in single frame detection result are removed by comparing the height of adjacent point from raw cloud of an obstacle point.Experiments are carried out with different scenarios of real world datasets.The method proposed in this study is proven reliable in realtime.