The effect of UAV on-board GPS error on the direct georeferencing of three-dimensional structure-from-motion models

Abstract

The use of direct georeferencing workflows for georeferencing UAV surveys and structure-from-motion three-dimensional models is seeing a rapid rise in use. The ability of this technique to obtain georeferenced models of areas that are inaccessible on foot - preventing the use of conventional ground control point georeferencing - and also the time saved by not having to carry out a ground survey are two factors leading to this increase in use. However, little work has been done on the effect of on-board GPS accuracy on the resulting georeferencing accuracy of these models. This study conducted a UAV survey of a winter wheat (Triticum aestivum) field using a UAV equipped with highly-accurate D-RTK GPS. The on-board D-RTK GPS coordinates were subsequently altered to simulate on-board GPS error, and several structure-from-motion models were generated using identical image sets that were attributed with differing levels of simulated GPS error. The study found that higher magnitudes of on-board GPS error resulted in higher model RMSE values, but that the correlation was not as strong as might be assumed. A high amount of variability of errors was present both between and within models, with the variability of errors within models showing a stronger correlation with increasing on-board GPS error. The study also identifies potential sources of error within the direct georeferencing workflow.