Spatio-Temporal Detection and Filtering of Dynamic Objects in Mobile LiDAR Point Clouds

Abstract. Mobile LiDAR systems are increasingly utilized for high-precision mapping in dynamic environments, yet the presence of moving objects introduces significant noise and distortions in the resulting point clouds. Addressing this challenge, this study proposes a novel and efficient method for detecting and removing moving objects from mobile LiDAR point clouds. The approach involves an initial separation of ground and non-ground points using the Cloth Simulation Filtering (CSF) algorithm, followed by density-based clustering (DBSCAN) of non-ground points. By analyzing the temporal distribution of LiDAR points (gpstime) within each cluster relative to ground points, clusters are classified as either static or dynamic. Dynamic clusters, corresponding to moving objects, are then excluded from the dataset, yielding a refined point cloud that better represents the static environment. The method is implemented in R using various open-source libraries and validated on high-traffic urban datasets acquired with the Riegl VMX-450 mobile LiDAR system. Experimental results demonstrate that the proposed pipeline effectively detects and removes dynamic objects, thereby improving the accuracy and reliability of LiDAR-based mapping in complex, real-world scenarios.