Abstract
Structure from motion photogrammetry presents a low-cost and time saving way of generating digital surface models and orthoimages using image matching algorithms. The use of structure from motion poses great potential where the implementation of aerial imagery should not be underrated, for example in regards to landslide mapping. Previous work has mostly focused on rather close range image acquisition techniques, but little research has been done to implement aerial imagery for larger geophysical processes, and previous research has left a gap for investigation of large landslides using structure from motion from aerial imagery. Firstly in this thesis, a sensitivity study was carried out of Årdal in western Norway (61.2684° N, 8.0385° E) to investigate the structure from motion processing of aerial imagery and possible processing issues. This was done by measuring absolute accuracy and random-access memory usage as well as carrying out a topography assessment. Secondly, an application of the technique was made to Fulånebba in western Norway (62.7233° N, 8.5801° E) to investigate a potential landslide site by making digital surface model difference maps and displacement maps. The main results showed firstly that topography, ground control point accuracy and the degree of well calibrated intrinsic camera parameters were governing factors for the absolute accuracy. Secondly, structure from motion was found to possibly yield digital surface models with similar accuracies without intrinsic camera parameters as with pre-calibrated cameras. Thirdly, random-access memory capacity was found to limit the number of used aerial images, level of automation or quality of the result. Further, this study suggests that structure from motion from aerial imagery might be used to imply general terrain changes. Lastly, the displacement error between a pair of orthoimages was found to be relative to the pixel size and the accuracy of the orthoimages after co-registration.