Urban DEM generation from raw lidar data: A labeling algorithm and its performance

This paper addresses the separation of ground points from raw lidar data for bald ground digital elevation model (DEM) generation in urban areas. This task is considered to be a labeling process through which a lidar point is labeled as either a ground point or a non-ground point. Mathematical formulation is presented to define the ground. A new approach is proposed that conducts one-dimensional labeling in two opposite directions followed by a linear regression, both along the lidar scan line profile. The study shows that the one-dimensional characteristic makes the calculation efficient, and the reliability is assured by the bidirectional labeling process. Lidar data over suburban and downtown Baltimore (Maryland), Osaka (Japan), and Toronto (Canada) are used for the study. Quality assessment is designed and conducted to investigate the performance of the labeling approach by using manually-selected ground truth. It is shown that 2.7 percent ground points are wrongly labeled as building points, and 2.6 percent building points are mistakenly labeled as ground points over the four study areas. Detailed graphic and numerical results are provided to illustrate the proposed labeling approach and its performance for complex urban areas.