Automatic levee surface extraction from mobile LiDAR data using directional equalization and projection clustering

Abstract

A levee is an engineering structure that physically suppresses flooding by directly touching the river. Enough levee elevation prevents water from overflowing the levee. Therefore, obtaining the physical shape of the river levee is the starting point of levee management based on the digital twin. Total station, global navigation satellite system (GNSS), or small unmanned aircraft systems (sUASs) have been used to investigate the physical shape of the levee. However, there are economical and temporal limitations in applying these methods to the entire levee alignment installed in more than ten thousand kilometers along the water system. Mobile laser scanner (MLS) mounted on a vehicle can resolve the problem because it moves at high speeds for a long time. Therefore, many researchers are conducting studies acquiring geospatial data with vehicle-type MLS. MLS collects data in point cloud form with locational information. Ground filtering algorithms for point clouds have been studied in order to obtain the physical shape of the levee from point clouds. However, since existing ground filtering algorithms are mostly designed for airborne laser scanner (ALS)- or terrestrial laser scanner (TLS)-based data, they have limitations in applying to point cloud data acquired from MLS. Therefore, we developed an algorithm to extract levee point cloud from MLS-acquired point cloud data. The algorithm contains several methodologies named directional equalization and point cloud projection clustering, which are designed to reflect the properties of MLS-acquired point cloud. The algorithm first divides the data into small units according to the direction of the river. Second, the levee points are extracted by applying directional equalization and point cloud projection clustering. Finally, the levee point extraction was completed through a process of increasing accuracy through a test based on the levee design value. As a result of its application to the 1 km Anyang-cheon stream area, an urban river located in Seoul, Korea, the classification accuracy of 95.9 % was obtained. Also, the levee elevation was automatically calculated in 0.194 m and 0.13 m of mean absolute error (MAE) compared to manual and total station measurements, respectively.

Keywords

• Mobile laser scanner
• Point cloud filtering
• Levee management
• Digital twin