Metoder för 3D-modellering i olika LoD : Automatisering för framställning av byggnader skapade med LIDAR-data i ArcGIS Pro

Sammanfattning: Building models in 3D have evolved from simple physical models to detailed and descriptive digital models (Pepe m.fl. 2021) that can be used in various applications such as urban planning, virtual reality, and navigation (Li m.fl. 2022). However, automated methods for creating 3D buildings remain a challenge due to the complexity of their structure and geometric shape, as well as the limited quality of input data. Automated workflows enable the processing of larger amounts of datain a shorter period (Borisov m.fl. 2022). The aim of the thesis was to investigate how ModelBuilder in the ArcGIS Pro software can be used to streamline the creation of buildings in LoD0, LoD1 and LoD2. It also examined how building volumes (calculated in the FME software) vary and whether there are differences in shape and position in LoD0 compared to building footprints in “fastighetskartan”. The study area covers 49,872 m2 and is in the eastern part of the Säffle urban area. Tools used in ArcGIS Pro included Create Footprints from Raster, Segment Roof Parts, and Create Buildings. In total, 46 buildings were created from LiDAR data. The results showed a completeness of 51.79% and 67.44% when considering only buildings over 20 m2. Out of the created 3D buildings, 80% had the same roof type as in reality. Several smaller buildings with pitched roofs were instead created with flat roofs. RMSE (Root Mean Square Error) was checked against a DSM (Digital Surface Model), and the result was 1.61 m. RMSE in the planar dimension was calculated for five buildings, and the result was 0.564 m. In summary, the process in ModelBuilder provides a useful result for creating and visualizing buildings at different LoDs. The studied area was relatively complex, with various roof types and significant vegetation surrounding most buildings. In future studies, it would be interesting to use different input data, as the results of this study indicated that the quality of the input data affects the outcome.