"Structure from Motion as an option for the replication of hydraulic structures in physical modelling "The creation of physical models is a necessary procedure for complex hydromorphological simulations. This thesis investigates the applicability of the Structure from Motion (SfM) approach to create a high-resolution 3D model of a detention basin with the software PhotoScan from Agisoft. The close range images are acquired by an Unmanned Aerial Vehicles (UAV). It is investigated which quantity and quality of the images is necessary and how they should be acquired and spatially distributed. This is done by comparing the Reprojection Error the Alignment Time and Point Number in the resulting Sparse Cloud for different combinations of image quantity and quality. Also the influence of mixed image acquisition, e.g. images of airborne and ground based acquisition with different cameras, is investigated.
Furthermore, the parameters Accuracy, Tie Point Limit and Key Point Limit of the image alignment in Agisoft PhotoScan are analyzed and evaluated for this special use case. Based on this investigations configurations of the Sparse Cloud generation are recommended for different quality requirements. The quality of the model is investigated by comparing the Sparse Cloud of the recommended settings with the Sparse Cloud of the default settings. In order to do that the Roughness of the point distribution, the Point Density and the overall surface roughness of the Cloud is evaluated with the software CloudCompare. Furthermore, the point cloud of the recommended setting is compared to the CAD data by a point distance to mesh calculation. It is shown that the SfM approach can produce a 3D point cloud of high quality with several difficulties in the further processing. The main problems which could be stated in this thesis are the insufficient under water detectability of SfM and, which is mandatory for 3D printing, turning the point cloud into a mesh and a solid CAD object.