Osaka Metropolitan University researchers proposed a geometry-aware framework for automatically removing human noise from TLS point clouds by projecting 2D instance segmentation masks into 3D space and validating candidates through multi-stage geometric filtering.
