Authors:

Anna Ospina Bedoya

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Ground filtering is an essential preprocessing task in digital forestry and other geospatial applications involving 3D point cloud data. Accurate identification of ground points improves terrain modeling and is critical for downstream tasks such as vegetation analysis and the detection of man- made objects like solar panels. Misclassified ground points can significantly distort these results, highlighting the need for robust filtering strategies. This project presents a modular workflow for ground filtering in LiDAR point clouds, aimed at enhancing terrain classification and enriching publicly available datasets hosted at https://lidar.digitalforestry.org. The main goal is to generate an additional label that distinguishes ground from non-ground points. The method divides large LiDAR scenes into overlapping patches (e.g., 625) using a regular grid. Each patch is processed using local geometric filters based on hyperbolic distance metrics, and a custom classification pipeline assigns binary labels. The patches are then merged to reconstruct the full scene, allowing for visual validation and refinement. Initial results show high accuracy in separating terrain from vegetation and built structures, particularly in forested regions. In Wildland-Urban Interface (WUI) zones, some buildings were misclassified as ground. To address this, we added a correction step using regression and clustering techniques, which improved results in these complex areas. The workflow is designed to be modular, enabling testing of alternative approaches such as hierarchical or sequential clustering, Progressive Morphological Filters, Cloth Simulation Filtering, and Multiscale Curvature Classification. Our approach consistently exceeds baseline methods in challenging environments. Keywords: LiDAR Point Clouds; Ground Filtering; Point Cloud Classification; Hyperbolic Distance

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Anna Ospina Bedoya