Point cloud filtering method for slope LiDAR based on attention mechanism and model scaling of PointNet++

Abstract

Accurate topographic filtering of LiDAR point clouds is critical for terrain change detection and digital elevation model (DEM) generation. In complex terrains with dense vegetation, airborne LiDAR data often suffer from a low ground-point ratio and uneven point density, limiting the performance of conventional filtering methods and standard deep learning models. To address these issues, this study proposes an enhanced PointNet++ algorithm integrating an attention mechanism and a model scaling strategy. The OpenGF dataset is used for training and validation, and a self-attention module is introduced to strengthen terrain feature perception, while a model scaling module is adopted to adapt to uneven point density. Through optimized network architecture, data augmentation, and training strategies, the proposed model achieves accurate ground/non-ground point classification. Experimental results show that the method achieves 95.48% accuracy and 91.25% mIoU on Test Set 1, and 96.26% accuracy and 88.30% mIoU on Test Set 2, outperforming PointNet, PointNet++, and other mainstream models. DEM construction for the Three Sisters Peak scene indicates that the proposed method provides higher accuracy than the traditional cloth simulation filter (CSF). This improved filtering approach offers reliable technical support for DEM generation and terrain change detection, with significant scientific and practical value.

Keywords

Topographic filtering / LiDAR point cloud / Attention mechanism / Model scaling / Digital elevation models (DEM)

References

[1] Aryan A, Bosché F, Tang P. Planning for terrestrial laser scanning in construction: A review. Automation in Construction, 2021, 125. 103551

[2] Chen C, Xu L, Hao J, Li Y, Chen D. Efficient interpolation-based and terrain-adaptive hierarchical filter for ultra-large-scale point cloud over complex landscapes. Optics and Laser TechnoloGy, 2025, 187. 112860

[3] Fox D, Burgard W, Thrun S. The dynamic window approach to collision avoidance. IEEE Robotics & Automation Magazine, 1997, 412333.

[4] Guo Q, Su Y, Hu T, Guan H, Jin S, Zhang J, Zhao X, Xu K, Wei D, Kelly M, Coops NC. Lidar boosts 3D ecological observations and modelings: A review and perspective. IEEE Geoscience and Remote Sensing Magazine, 2021, 91232257.

[5] Hu Q, Yang B, Xie L, Rosa S, Guo Y, Wang Z, Markham A. Randla-net: Efficient semantic segmentation of large-scale point clouds. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.

[6] Kim JY, Cho JW, Choi CH, Lee SY. UAV-based quantitative assessment of road embankment smoothness and compaction using curvature analysis and intelligent monitoring. Remote Sensing, 2025, 17111867

[7] Li Y, Luo Y, Lian D, Bu C, Wang H. Outdoor LiDAR point cloud classification algorithm based on multilevel point cluster feature fusion. Heliyon, 2025, 114e42623

[8] Lin X, Zhang J. Segmentation-based filtering of airborne LiDAR point clouds by progressive densification of terrain segments. Remote Sensing, 2014, 6212941326.

[9] Niemeyer J, Rottensteiner F, Soergel U. Contextual classification of LiDAR data and building object detection in urban areas. ISPRS Journal of Photogrammetry and Remote Sensing, 2014, 87: 152-165.

[10] Qi CR, Su H, Mo K, Guibas LJ. Pointnet: Deep learning on point sets for 3d classification and segmentation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017.

[11] Qi CR, Yi L, Su H, Guibas LJ. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. Advances in Neural Information Processing Systems., 2017, 30: 5099-5108.

[12] Qian G, Hammoud H, Li G, Thabet A, Ghanem B. ASSANet: An anisotropic separable set abstraction for efficient point cloud representation learning. Advances in Neural Information Processing Systems., 2021, 34: 28119-28130.

[13] Qian G, Li Y, Peng H, Mai J, Hammoud H, Elhoseiny M, Ghanem B. PointNext: Revisiting PointNet++ with improved training and scaling strategies. Advances in Neural Information Processing Systems., 2022, 35: 23192-23204.

[14] Qin N, Tan W, Ma L, Zhang D, Li J. OpenGF: An ultra-large-scale ground filtering dataset built upon open ALS point clouds around the world. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021.

[15] Santo A, Heredia E, Viegas C, Valiente D, Gil A. Ground segmentation for lidar point clouds in structured and unstructured environments using a hybrid neural–geometric approach. Technologies, 2025, 134162

[16] Sinha D, El-Sharkawy M. Thin mobilenet: an enhanced mobilenet architecture. 2019 IEEE 10th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON)., 2019.

[17] Suchocki C, Nowak R, Rutkowski R, Błaszczak-Bąk W. Using handheld 3D laser scanner and high-resolution handheld digital microscope for hybrid building condition measurements. Measurement, 2025, 253. 117751

[18] Suleymanoglu B, Soycan M. A novel framework for road information extraction from low-cost MMS point clouds. IEEE Access, 2024, 12: 195450-195463.

[19] Varney N, Asari VK, Graehling Q. DALES: A large-scale aerial LiDAR data set for semantic segmentation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2020.

[20] Zhang W, Qi J, Wan P, Wang H, Xie D, Wang X, Yan G. An easy-to-use airborne LiDAR data filtering method based on cloth simulation. Remote Sensing, 2016, 86501.

[21] Zhao H, Jiang L, Jia J, Torr PH, Koltun V. Point transformer. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021.

[22] Zheng Q, Zhang C, Sun J. Pointmt: Efficient point cloud analysis with hybrid MLP-transformer architecture. IEEE Transactions on Multimedia, 2025.

[23] Zhou T, Cheng X, Lin P, Wu Z, Liu E. A general point-based method for self-calibration of terrestrial laser scanners considering stochastic information. Remote Sensing, 2020, 12182923

[24] Zolanvari SM, Ruano S, Rana A, Cummins A, Da Silva RE, Rahbar M, Smolic A. DublinCity: annotated LiDAR point cloud and its applications. arXiv Preprint, 2019.

Rights & permissions

The Author(s)