A point cloud filtering approach for complex valley terrains integrating terrain segmentation and surface compensation

Extracting ground points from point clouds in complex valley terrains remains challenging due to steep slopes, dense vegetation, and diverse surface features. This study proposes a novel extraction method that combines terrain segmentation with surface compensation. The workflow segments the point cloud into valley floor, road, vertical, and slope surfaces using RANSAC, region growing, and elevation–normal constraints. A Thin Plate Spline (TPS) is then applied to fit the slope trend surface, reducing slope steepness and achieving surface flattening through surface compensation. Ground points are then extracted using Cloth Simulation Filtering (CSF), and the original terrain is reconstructed by restoring slope elevations. Experiments on test area S1 show that the proposed method achieves a Type I error of 7.30%, a Type II error of 4.69%, and a total error of 6.02%, with a Kappa coefficient of 0.880. Compared with conventional algorithms, the method demonstrates superior accuracy and robustness, making it well suited for point cloud filtering in complex, vegetation-rich valley environments.