| Title |
Accuracy Assessment of Ground Extraction and Earthwork Volume Estimation through UAV LiDAR Reflectance Intensity Filtering Based on Unsupervised Learning Algorithm |
| Authors |
강형석(Kang, Hyeongseok) ; 이기림(Lee, Kirim) ; 신현길(Shin, Hyeongil) ; 김정옥(Kim, Jungok) ; 이원희(Lee, Wonhee) |
| DOI |
https://doi.org/10.12652/Ksce.2025.45.2.0265 |
| Keywords |
위성항법시스템; 무인항공기; 라이다; 반사 강도; 비지도학습; 토공량 Global navigation satellite system; Unmanned aerial vehicle; Light detection and ranging; Reflectance intensity; Unsupervised learning; Earthwork volume |
| Abstract |
In this study, we addressed the limitations of conventional earthwork volume estimation methods caused by seasonal and spatial restrictions due to vegetation and the labor-intensive process of artificial vegetation removal. To overcome these challenges, a GNSS(Global Navigation Satellite System) and UAV(Unmanned Aerial Vehicle) equipped with a LiDAR(Light Detection and Ranging) sensor were used to acquire point cloud data. The Reflectance Intensity feature of the data was utilized to generate vegetation and ground-separated datasets. These datasets were then employed to estimate earthwork volume, which was compared and evaluated against the widely used GNSS-based VRS(Virtual Reference Station) surveying method, a standard in civil engineering and research applications This study implemented unsupervised learning algorithms, including K-Means, K-Medoids, and DBSCAN(Density-Based Spatial Clustering of Applications with Noise), to filter reflectance intensity data based on its unique density characteristics and surface textures. The clustering results identified distinct core clusters, and multiple reflections enabled the inclusion of ground points beneath vegetation.
The cluster with the highest point density was classified as ground and extracted for further analysis. The ground data, filtered using these clustering algorithms, were employed to calculate total earthwork volume, combining cut and fill volumes. The results indicated overestimations of 1.4 % and 0.3 % and an underestimation of 0.5 %. A weighted accuracy analysis, considering the proportions of cut and fill volumes, confirmed that the K-Means clustering algorithm achieved the highest accuracy among the methods. This research demonstrates the potential of UAV-based LiDAR data and unsupervised learning algorithms to enhance the accuracy and efficiency of earthwork volume estimation, offering a robust alternative to traditional methods while overcoming seasonal limitations and operational inefficiencies. |