Abstract
Structure-from-Motion (SfM) photogrammetry is increasingly employed in geomorphological applications for change detection, but repeatability and reproducibility of this methodology are still insufficiently documented. This work aims to evaluate the influence of different survey acquisition and processing conditions, including the camera used for image collection, the number of Ground Control Points (GCPs) employed during Bundle Adjustment, GCP coordinate precision and Unmanned Aerial Vehicle flight mode. The investigation was carried out over three fluvial study areas characterized by distinct morphology, performing multiple flights consecutively and assessing possible differences among the resulting 3D models. We evaluated both residuals on check points and discrepancies between dense point clouds. Analyzing these metrics, we noticed high repeatability (Root Mean Square of signed cloud-to-cloud distances less than 2.1 cm) for surveys carried out under the same conditions. By varying the camera used, instead, contrasting results were obtained that appear to depend on the study site characteristics. In particular, lower reproducibility was highlighted for the surveys involving an area characterized by flat topography and homogeneous texturing. Moreover, this study confirms the importance of the number of GCPs entering in the processing workflow, with different impact depending on the camera used for the survey.
Highlights
Every measurement process, with any technology, is always affected by errors, which, if not properly considered, lead to inevitable biased or wrong evaluations of the phenomenon under study
Several papers in the literature compared the SfM results with other very high resolution (VHR) acquisition techniques, such as terrestrial or airborne laser scanning [3,15], or evaluated SfM accuracy on a discrete set of points, whose coordinates are measured through Global Navigation Satellite System (GNSS) [16,17]
We reported the mean Ground Sampling Distance (GSD) computed for each image set after the SfM process
Summary
With any technology, is always affected by errors, which, if not properly considered, lead to inevitable biased or wrong evaluations of the phenomenon under study. To reliably detect and quantify temporal surface changes avoiding false positives [14], a deep knowledge of the data processing steps and of the uncertainties affecting the model is mandatory To this end, several papers in the literature compared the SfM results with other VHR acquisition techniques, such as terrestrial or airborne laser scanning [3,15], or evaluated SfM accuracy on a discrete set of points, whose coordinates are measured through Global Navigation Satellite System (GNSS) [16,17]. Comparing SfM results with regard to other (possibly more accurate) surveying techniques can provide an estimate of measurement accuracy and potential systematic errors, but it is not able to capture precision and repeatability [20] The latter are fundamental aspects to quantify digital elevation model (DEM) uncertainties, which significantly affect surface change detection
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
