The estimation of position and attitude of a camera, addressed as image orientation in photogrammetry, is an important task to obtain information on where a platform is located in the world or relative to objects. Unmanned aerial vehicles (UAV) as an increasingly popular platform led to new applications, some of which involve low flight altitudes and specific requirements such as low weight and low cost of sensors. Image orientation needs additional information to retrieve not only relative measurements but position and attitude in a world coordinate system. Given the requirements on sensors and especially for flights in between obstacles in urban environments classically used information of Global Navigation Satellite Systems (GNSS) and Inertial Measurement Units (IMU) or specially marked ground control points (GCP) are often inaccurate or unavailable. The idea addressed within this thesis is to improve the UAV image orientation based on an existing generalised building model. Such models are increasingly available and provide ground control that is helpful to compensate inaccurate or unavailable camera positions measured by GNSS and drift effects of image orientation. Typically, for UAV applications in street corridors, the geometric accuracy and the level of detail of such models is low compared to the high accuracy and high geometric resolution of the image measurements. Therefore, although the building model differs from the observed scene due to its generalisation, relations of the photogrammetric measurements to the building model are formulated and used in the determination of image orientation. Three approaches to assign tie points to model planes in object space are presented, and a sliding window as well as a global hybrid bundle adjustment are set up for image orientation aided by a generalised building model. The assignments lead to fictitious observations of the distance of tie points to model planes and are iteratively refined by bundle adjustment. Experiments with an image sequence captured flying between buildings show an improvement of image orientation from the metre range with purely GNSS measurements to the decimetre range when using the generalised building model with the simplest assignment method based on point-to-plane distances. No improvement by searching planes in the tie point cloud to indirectly find the relations of tie points to model planes is observed. The results are compared to a building model of higher detail and systematic effects are investigated. In summary, the developed method is found to significantly improve UAV image orientation using a generalised building model successfully.
展开▼
