HDID: a method for removing dynamic objects in the 3D point cloud map

3D point cloud maps play a critical role in the autonomous localization and navigation of robots. However, when constructing point cloud maps using scanning sensors such as LiDAR and cameras, the presence of dynamic objects in the environment can lead to map distortions and "ghost artifacts," resulting in map ping errors that severely impact the robot's localization and navigation tasks. To address this issue, we propose a novel method, to our knowledge, based on height-density occupancy descriptors and density ratios, which effectively removes dynamic objects from 3D point cloud maps. First, we introduce the height-density occupancy descriptor, which utilizes the quantity of point clouds in the upper and lower layers to represent the height information of the current map. Subsequently, we propose strategies including the density ratio test, inner loop detection, and density ratio deletion, overcoming the limitations of existing methods. Through extensive evaluation on open-source datasets, our method demonstrates superior performance in accurately removing dynamic objects and correctly preserving static objects compared to state-of-the-art approaches.