Structure-from-Motion on shallow reefs and beaches: potential and limitations of consumer-grade drones to reconstruct topography and bathymetry

Reconstructing the topography of shallow underwater environments using Structure-from-Motion – Multi View Stereo (SfM-MVS) techniques applied to aerial imagery from Unmanned Aerial Vehicles (UAVs) is a challenging problem, as it involves non-linear distortions caused by water refraction. This study presents an experiment with aerial photographs collected with a consumer-grade UAV on the shallow-water reef of Fuvahmulah, the Maldives. Under conditions of rising tide, we surveyed the same portion of the reef in ten successive flights. For each flight, we used SfM-MVS to reconstruct the Digital Elevation Model (DEM) of the reef, and used the fli...  more

Reconstructing the topography of shallow underwater environments using Structure-from-Motion – Multi View Stereo (SfM-MVS) techniques applied to aerial imagery from Unmanned Aerial Vehicles (UAVs) is a challenging problem, as it involves non-linear distortions caused by water refraction. This study presents an experiment with aerial photographs collected with a consumer-grade UAV on the shallow-water reef of Fuvahmulah, the Maldives. Under conditions of rising tide, we surveyed the same portion of the reef in ten successive flights. For each flight, we used SfM-MVS to reconstruct the Digital Elevation Model (DEM) of the reef, and used the flight at low tide (where the reef is almost entirely dry) to compare the performance of DEM reconstruction under higher water levels. Our results show that differences with the reference DEM increase with increasing depth, but are substantially larger if no underwater ground control points are taken into account in the processing. Correcting our imagery with algorithms that account for refraction did not improve the overall accuracy of reconstruction. We conclude that reconstructing shallow-water reefs (less than 1 m depth) with consumer-grade SfM-MVSs and SfM-MVS is possible, but its precision is limited and strongly correlated with water depth. The best results are achieved when ground control points were placed both above and underwater and no refraction correction is used in our processing.