Evaluation of high-accuracy terrain model generation performance in archaeological excavations using LiDAR point cloud

Archaeological excavations are of great importance for preserving cultural heritage and reconstructing human history. In recent years, the development of remote sensing and photogrammetric methods has accelerated the analysis of archaeological sites. Archaeological excavation sites are often covered by dense vegetation, and obtaining a true terrain model in these areas has been an important research topic. In this thesis, in order to perform ground filtering using LiDAR point cloud data, Cloth Simulation Filter (CSF), Progressive Triangulated Irregular Network (PTIN) and Support Vector Machine (SVM) methods in the literature are compared with different parameter combinations. Çıngırt Castle, located in Yapraklı Neighborhood of Fatsa District, Ordu Province, and registered as a Grade 1 Archaeological Site, was chosen as the test site. In the comparison with 1159 reference data obtained with ground-based LiDAR data in the field, the ground extraction performances of CSF, PTIN and SVM methods resulted in F-score of 97.66%, 96.80% and 73.40% respectively. The combination of PTIN and CSF methods resulted in an F-score of 96.89%. The average height difference with respect to the reference data was approximately 10 cm. The results show that PTIN is more successful in vegetation and ground filtering, while CSF is more successful in rocky and canvas areas. Precise terrain model generation with the integrated method (PTIN+CSF) resulted in more efficient ground extraction.