In this master’s thesis, we examine the segmentation and attribution of a 3D mesh based on building footprints. Most of the work is carried out in ArcGIS Pro. The primary datasets used are a photogrammetric point cloud and 3D mesh of a selected area in Ljubljana, created from oblique aerial imagery. Segmentation is performed using building footprints obtained through manual vectorization, classification of the photogrammetric point cloud, and publicly available sources. We tested and described five sets of building classification results in the point cloud. The quality of the point cloud classification was evaluated in Matlab. The process of obtaining building footprints is described for nine different cases, and their quality was evaluated. Attribute data were also obtained from publicly available sources. Manual vectorization of building footprints proved to be the most effective method for segmenting and attributing the 3D mesh. Good results were also achieved with building footprints derived from manual and automatic point cloud classification, as well as from transferring classification from a georeferenced and classified point cloud obtained through the project of national Cyclical lidar survey of Slovenia. We found that the quality of segmentation and attribution of the 3D mesh depends on the quality of the building footprints. Despite its limitations, the thesis demonstrates a potential approach for integrating 3D meshes into modern GIS environments and provides a basis for further research in the field of automatic building detection.
|
