A study on comparison of CSF and ATIN LiDAR ground filtering algorithms in digital terrain model production

Three-dimensional numerical models are important sources of information for many disciplines in determining the details of the earth. From the past to the present, these models are produced by many classical methods such as photogrammetry and geodetic measurement. One of the methods used to produce these models is the LiDAR scanning system. Especially in the 20th century this technology, which became widespread towards the end, was used in many different activities in western countries, but it is not yet widespread in our country. In the production of digital models belonging to the earth, LiDAR technology stands out with its great advantages in terms of labor, time and accuracy compared to traditional methods. Processing LiDAR data consisting of an irregular point cloud is an important process for the numerical models we will create. In this process, filtering of raw LiDAR data is an important step. There are many different filtering algorithms that determine ground points by filtering LiDAR data. When the studies are examined, the strengths and weaknesses of each algorithm are determined for different terrain surfaces. Each algorithm has different parameters in its content. In this study, the performance of CSF and ATIN algorithms in filtering the ground points from the Air LiDAR point data was tested for a difficult land surface consisting of dense woodland areas and steep slopes. In the application process, the ground and object points are separated by filtering with different values in both algorithm parameters. As a result of the study, a Digital Terrain Model was created from the ground points filtered for each value. The accuracy of the produced land models were compared with the Digital Terrain Model we use as a reference. As a result, the advantages and disadvantages of algorithms operating with different filtering techniques in filtering the ground points and in the production of Digital Terrain Model were evaluated.