A Phase-Domain Postfilter Approach for LiDAR DSM in Urban Areas

Abstract

Light detection and ranging (LiDAR)-derived digital surface models (DSMs) provide high-resolution representations of the Earth’s surface and serve as important input data for 3-D urban modeling. However, these datasets are inevitably noisy due to sensor ranging errors and the inherent roughness of surface materials. In dense urban areas, traditional spatial-domain filters rely on a local smoothness assumption and therefore struggle to handle abrupt height variations, such as building edges, resulting in signal-to-noise aliasing. To address this issue, we propose a novel phase-domain postfilter approach (PFA) that transfers the denoising process from the spatial domain to the phase domain by introducing a virtual period parameter. LiDAR elevation data are first mapped into wrapped phase fringes, after which a frequency-domain Goldstein filter is applied to effectively separate noise from structural information according to their spectral characteristics. Finally, the denoised elevation model is reconstructed through constrained phase unwrapping. Experimental results on dense urban LiDAR datasets from New York City (NYC) and Dublin demonstrate that, compared with conventional filters, the proposed method better preserves the vertical edges of buildings while effectively smoothing flat regions.