Efficient interpolation-based and terrain-adaptive hierarchical filter for ultra-large-scale point cloud over complex landscapes

Abstract

Complex landscapes, typically characterized by ​outliers, objects with diverse structures, vegetation on steep slopes, and terrain discontinuities, pose significant challenges ​to traditional filtering methods, especially when processing ​ultra-large-scale point clouds. To address these challenges, this paper proposes an efficient ​interpolation-based and terrain-adaptive hierarchical filtering method. Specifically, a hybrid algorithm ​combining a moving-window detector with robust surface fitting is developed to ​optimize the selection of initial ground seeds. Subsequently, a ​weighted finite-difference-based Thin Plate Spline (TPS) method is introduced to generate reference ground surfaces, ​thereby improving computational efficiency and ​mitigating the impact of misclassified object points. Finally, a ​terrain-adaptive filtering threshold incorporating different orders of terrain roughness is designed to accurately extract ground points near terrain discontinuities. To evaluate the proposed method, extensive experiments were conducted on ​the ISPRS benchmark samples and ​the ultra-large-scale OpenGF dataset. Results demonstrate that our method achieves an average Kappa coefficient of 92.3% across all 15 ISPRS samples, ​outperforming 24 state-of-the-art filtering methods published since 2010. On the OpenGF dataset, the proposed method ​surpasses five classical filters, reducing the average total error by 34.1%–77.4% and improving the average Kappa coefficient by 2.8%–21.7%. Overall, this framework provides a robust solution for filtering ultra-large-scale point clouds in complex landscapes.