Imagery acquired from unmanned aircraft systems (UAS) and processed with structure from motion (SfM) – multi-view stereo (MVS) algorithms provides transformative new capabilities for surveying and mapping. Together, these new tools are leading to a democratization of airborne surveying and mapping by enabling similar capabilities (including similar or better accuracies, albeit from substantially lower altitudes) at a fraction of the cost and size of conventional aircraft. While SfM-MVS processing is becoming widely used for mapping topography, and more recently bathymetry, empirical accuracy assessments—especially, those aimed at investigating the sensitivity of point cloud accuracy to varying acquisition and processing parameters—can be difficult, expensive, and logistically complicated. Additional challenges in bathymetric mapping from UAS imagery using SfM-MVS software relate to refraction-induced errors and lack of coverage in areas of homogeneous sandy substrate. This dissertation aims to address these challenges through development and testing of new algorithms for SfM-MVS accuracy assessment and bathymetry retrieval.
A new tool for simulating UAS imagery, simUAS, is presented and used to assess SfM-MVS accuracy for topographic mapping (Chapter 2) and bathymetric mapping (Chapter 3). The importance of simUAS is that it can be used to precisely vary one parameter at a time, while perfectly fixing all others, which is possible, because the UAS data are synthetically generated. Hence, the issues of uncontrolled variables, such as changing illumination levels and moving objects in the scene, which occur in empirical experiments using real UAS, are eliminated. Furthermore, simulated experiments using this approach can be performed without the need for costly and time-intensive fieldwork. The results of these studies demonstrate how processing settings and initial camera position accuracy relate to the accuracy of the resultant point cloud. For bathymetric processing, it was found that camera position accuracy is particularly important for generating accurate results.
Even when accurate camera positions are acquired for bathymetric data, SfM-MVS processing is still unable to resolve depths in regions which lack seafloor texture, such as sandy, homogeneous substrate. A new methodology is introduced and tested which uses the results from the SfM-MVS processing to train a radiometric model, which estimates water depth based on the wavelength-dependent attenuation of light in the water column (Chapter 4). The methodology is shown to increase the spatial coverage and improve the accuracy of the bathymetric data at a field site on Buck Island off of St. Croix in the U.S. Virgin Islands. Collectively, this work is anticipated to facilitate greater use of UAS for nearshore bathymetric mapping.