Modeling 3D buildings of LOD2 from airborne point cloud using unsupervised classification

Abstract

A variety of applications utilize 2D data in some form or the other to complete their tasks. But we are living in a 3D world and in most cases 2D information is not sufficient. Today the need for 3D Geoinformation has increased rapidly mainly because there is a significant improvement in maintaining, processing and visualizing these data. A variety of applications have been introduced in relation to visualization like a 3D city model. A 3D city model includes buildings, vegetation, street furniture and other city objects. 3D city models can be generated from various sources of data like aerial images, CAD, satellite imagery, LiDAR and terrestrial laser scan. But LiDAR and terrestrial laser scan holds as the best source of data in terms of accuracy. With LiDAR accurate 3D models can be generated when compared to other conventional method like the photogrammetric technique. The data is collected as set of points called as point cloud. MicroStation with extension TerraScan was used to process these 3D point clouds from which the 3D models and the 3D surface model (DTM) were generated. This study aims to generate 3D city model from the airborne LiDAR and incorporate them in CityServer3D where the 3D geodatabase is created. All the generated models are based on the standard CityGML format. Each building is given an external code based on the CityGML format defined by the Open Geospatial Consortium (OGC). The models inside the CityServer3D can be visualized as well as queried. The Level of Detail of the 3D models is restricted to 2 without façade textures. This 3D city model will be of good use to the local authorities of Miri during times of flood because the study area is located relatively close to a river meeting the sea. This 3D city model can be improved by adding textures, increasing the level of detail which will be more virtual and realistic.