Abstract: In the field of Airborne LiDAR point clouds processing, building points extraction is always an active area. The core of this task is to separate building points from vegetation or other objects, but it is very difficult in different urban scenes. Therefore, this paper proposes a hierarchical method to precisely detect building points, aiming to improve the ability of separating buildings and other objects in various complex urban scenes. The method firstly separates non-ground points from ground points based on the filtering process, and to extract building candidate regions from non-ground points according to some simple geometrical features of a building. For each building candidate region, the morphological reconstruction and the point segmentation are fused to generate multi-scale space, and to construct topological relationship graphs between adjacent scales. Then, the proposed method extracts building regions from all object regions by five features based on topological relationship graphs. Finally, the proposed method removes non-building points from building regions for obtaining the final building points. In order to verify the validity and reliability of the proposed method, ISPRS (International Society for Photogrammetry and Remote Sensing) benchmark datasets from Vaihingen and Toronto are selected to perform experiments, and the results are evaluated by ISPRS. Results show that Completeness, Correctness, and Quality of object-based or area-based are larger than 87.8%, 94.7% and 87.3%, respectively. And compared with other methods, the method is the most robust in object-based or area-based evaluation. It demonstrates that the method could robustly detect building points in different urban scenes with a high correctness.