Uncertainty modelling of static laser scanning using deep learning

Terrestrial laser scanners (TLS) are one of the standard methods in 3D point cloud acquisition due to their high data rate and resolution. In some application areas, such as deformation analysis, it is essential to be able to model the uncertainties of the 3D point clouds as accurate as possible. In order to investigate the uncertainties of the Z+F Imager 5016, a reference point cloud was recorded in the HiTec laboratory of the geodetic institute. Subsequently, 50 static scans were measured with the Z+F Imager 5016 at different positions inside the laboratory. By comparing the reference point cloud with the laser scan point cloud, the uncertainties of the laser scanner can be investigated and modelled. In a previous thesis, uncertainty modelling was carried out using various machine learning approaches. Conventional approaches such as multiple linear and non-linear regression as well as the XGboost regressor (based on decision trees) achieved promising results. The aim of this work is to investigate the suitability of deep learning methods for modelling the uncertainties of static laser scanning.