Today, measuring of buildings and rooms is still undertaken with tachymeters and partly also with measuring tapes. However, in recent years the share of 3D laser scan-ners has rapidly increased, since they are more accurate, less prone to human error and faster in capturing a scene. When scanning a scene, some parts of the scene are missing due to occlusions by objects. An example of this is a cabinet standing in front of a wall. The 3D scanner then only captures a point cloud with the surface of the cabinet and the surface information of the wall behind it is missing. In order to predict these occluded points in a point cloud, this master thesis analyzes the existing methods of scene completion and proposes a novel approach to convert a point cloud into a truncated signed distance field. This truncated signed distance field is then used on an existing volumetric scene completion network that is fully self-su-pervised. Additionally, the suitability of the LiDAR sensor of iPhones and iPads for generating a dataset is examined.     «

Today, measuring of buildings and rooms is still undertaken with tachymeters and partly also with measuring tapes. However, in recent years the share of 3D laser scan-ners has rapidly increased, since they are more accurate, less prone to human error and faster in capturing a scene. When scanning a scene, some parts of the scene are missing due to occlusions by objects. An example of this is a cabinet standing in front of a wall. The 3D scanner then only captures a point cloud with the surface o...     »