D-WLTS: A LiDAR Point Cloud Water-Land Segmentation Algorithm Based on Delaunay and Multi-Dimensional Topological Constraints

Abstract

Water-land segmentation of LiDAR point clouds is crucial for the environment perception of amphibious robots. Addressing the lack of dedicated techniques for LiDAR point cloud water-land segmentation, this paper proposes a Delaunay topological constraint-based Water-Land Topological Segmentation algorithm (D-WLTS). Leveraging the sparse void characteristic of water surface point clouds, the algorithm constructs a 2.5D Delaunay triangulation and integrates multi-dimensional features, such as edge length and slope, to distinguish water from land. To mitigate complex environmental interference, D-WLTS utilizes a collaborative optimization mechanism: it first purifies the ground base combining density-based clustering, statistical outlier filtering and Cloth Simulation Filter (CSF). Then, to resolve water system discontinuities caused by obstacles like cross-river bridges, an innovative spatial indexing and Union-Find strategy preserves the main water system’s integrity, followed by conditional erosion to eliminate false positive artifacts (FPA). Finally, high-precision histogram inversion calculates the absolute water level, reconstructing the complete water surface via regional flood-filling. Experiments demonstrate that D-WLTS efficiently and accurately processes complex scenarios, significantly improving segmentation adaptability and establishing a unified paradigm for point cloud water-land segmentation.