A Terrain-Feature-Aware Multiscale PTD for Airborne LiDAR Ground Filtering

Airborne light detection and ranging (LiDAR) has become an indispensable tool for earth observation, supporting applications such as digital terrain modeling, environmental monitoring, and geohazard assessment. A fundamental challenge in LiDAR data processing is the accurate separation of ground and nonground points, which critically affects the quality of downstream analyses. The progressive TIN densification (PTD) algorithm has been the dominant solution for airborne LiDAR ground filtering and is widely implemented in software such as TerraSolid. However, in complex terrains, conventional PTD often introduces excessive smoothing and suppresses fine-scale terrain features. To address this limitation, we propose a terrain feature aware multiscale PTD algorithm. The approach incorporates three strategies: 1) multiscale seed point selection guided by TIN geometry for robust initialization, 2) adaptive boundary expansion to maintain edge integrity during progressive iterations, and 3) terrain feature compensation models designed for steep and discontinuous terrains. Experiments on both the ISPRS filtertest dataset and the large-scale OpenGF dataset demonstrate substantial improvements. On the ISPRS dataset, the average total error is 2.96% and the Kappa coefficient is 89.36%. On the OpenGF dataset, the average total error is 3.38% and the Kappa coefficient reaches 90.9%. The results confirm that the proposed method enhances ground filtering accuracy and provides more reliable topographic mapping in complex environments.