关键词: 机载激光扫描, 离散度, 高程, 区域增长, 后向加密

Abstract: To tackle the problems on the generation of Digital Elevation Model（DEM） in urban areas, this paper proposes lidar-driven multi-scale filtering algorithm. By analyzing and calculating spatial geometric relationships of each point in a certain neighbourhood it firstly classify point clouds into three basic classes（planar, edge, and discrete points）. Thanks to urban areas featured by relatively flat piecewise planes, the planar points are further processed to obtain building, on-ground and uncertain points using region growing and global height thresholds. To improve the DEM accuracy, a Delaunay triangular network of the filtered on-ground points is built to backward search possible on-ground points in edge and uncertain points. The experiments of ISPRS datasets validate the effectiveness and robustness of the proposed coarse-to fine multi-scale filtering algorithm in urban areas.

Key words: lidar-driven, eigen-values, height, region growing, backward densification