LiDAR has become a key technology for acquiring high-precision surface information.However

due to the diversity of terrain conditions and the limitations in adaptability of existing algorithms

accurately separating ground points from non-ground points in complex point cloud data—known as point cloud ground filtering—remains challenging.This paper comparatively studies three ground filtering methods:the rule-based PTD and CSF

as well as the deep neural network-based TransGF.The experiment analyzes point cloud data from a study area in Zhenjiang City

Jiangsu Province

which features complex terrain characteristics.The performance of each method is evaluated using metrics such as Type I error

Type II error

Total error

and the Kappa coefficient.The results show that the TransGF algorithm performs optimally across all performance metrics.Particularly in handling terrain features such as steep cliffs

TransGF demonstrates superior precision in retaining ground information while generating smoother digital surface models.