The recent emergence of Structure-from-Motion Photogrammetry (SfM) has created a cost-effective alternative to conventional laser scanning for the production of high-resolution topographic datasets. There has been an explosion of applications of SfM within the geomorphological community in recent years, however, the focus of these has largely been small-scale (102 - 103 m2), building on innovations in low altitude Unmanned Aircraft Systems (UAS). This thesis examines the potential to extend the scope of SfM photogrammetry in order to quantify of landscape scale processes. This is examined through repeat surveys of a ~35 km2 reach of the Dart River, New Zealand. An initial SfM survey of this reach was conducted in April 2014, following a large landslide at the Slipstream debris fan. Validation of the resulting digital elevation models using Independent Control Point's (ICPs) suggested encouraging results, however benchmarking the survey against a long-range laser scanned surface indicated the presence of significant systematic errors associated with inaccurate estimation of the SfM bundle adjustment. Using a combination of scaled laboratory field experiments, this research aimed to develop and test photogrammetric data collection and modelling strategies to enhance modelling of 3D scene structure using limited constraints. A repeat survey in 2015 provided an opportunity to evaluate a new survey strategy, incorporating a convergent camera network and a priori measurement of camera pose. This resulted in halving of mean checkpoint residuals and a reduction in systematic error. The models produced for both 2014 and 2015 were compared using a DEM differencing (DoD) methodology to assess the applicability of wide-area SfM models for the analysis of geomorphic change detection. The systematic errors within the 2014 model confound reliable change detection, although strategies to correlate the two surveys and measure the residual change show promise. The future use of SfM over broad landscape scales has significant potential, however, this will require robust data collection and modelling strategies and improved error modelling to increase user confidence.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:766127 |
| Date | January 2018 |
| Creators | James, Joe Steven |
| Publisher | Queen Mary, University of London |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://qmro.qmul.ac.uk/xmlui/handle/123456789/36224 |
Page generated in 0.0026 seconds