The accuracy of aerial triangulation products automatically generated from hyper-spatial resolution digital aerial photography

In recent years, unmanned airborne systems (UAS) have emerged as an important platform for collecting hyper-spatial resolution airborne remote sensing data. Using this hyper-spatial resolution imagery as input, modern aerial triangulation (AT) techniques, also known as structure-from-motion or SfM, can rapidly produce orthophotos and digital surface models (DSMs) at fine scales. Such data hold great promise for a number of applications, including routine and post-disaster assessment of transportation infrastructure, which provided the impetus for this research. Using hyper-spatial resolution (0.002 m) natural colour digital aerial photography acquired from a low-altitude UAS as input images, this research systematically investigated the horizontal and vertical accuracy of the AT generated orthophotos and DSMs, respectively. Hyper-spatial resolution aerial data were collected for a total of 28 study sites and, for each study site, coordinate information of 16 ground control points (GCPs) was collected using a survey grade real-time kinematic (RTK) Global Navigation Satellite System. Among the 16 GCPs for each site, 10 were used to calibrate the AT process while the remaining six GCPs were reserved to evaluate the horizontal and vertical accuracy of the orthophotos and DSMs. An average horizontal root mean squared error (RMSE) of 0.004 m and a vertical RMSE of 0.007 m across all sites indicate great promise for AT processed hyper-spatial resolution airborne remote sensing data to play a significant role in transportation infrastructure monitoring, particularly when considering the horizontal and vertical accuracy of the surveyed GCPs (0.004 and 0.006 m, respectively).