LAND SURVEYING PERFORMANCE OF COMMERCIALLY-AVAILABLE INEXPENSIVE SMALL UAV

ABSTRACT: We preliminary evaluated 3-dimensional accuracy of Digital Surface Model (DSM) derived from a set of digital imagery, which was obtained using commercially-available inexpensive small UAV-based systems, and Structure-from-Motion (SfM) and Multi-View-Stereo (MVS) techniques. Two types of electronic-battery-powered quadcopter, DJI Phantom and DJI Phantom 2 Vision+, were used as a small UAV (sUAV) platform. The main difference of them is that the latter has an autonomous flight capability allowing the sUAV to follow a series of predefined waypoints, whereas the former has to be controlled manually by an operator. The digital cameras mounted on the sUAV were GoPro HERO3 for Phantom and Ricoh GR for Phantom 2 Vision +. Thirteen ground control points (GCPs) were deployed over the entire study site and their 3-dimensional coordinates were measured with static GNSS surveying method. The DSM generated from a set of imagery observed by Ricoh GR mounted on Phantom 2 Vision + was georeferenced with 5 GCPs and, its accuracy was evaluated based on the remaining 8 GCPs. A set of imagery observed by GoPro HERO3 was not used for accuracy assessment because of two reasons: (1) a set of imagery failed to cover the entire study site due to difficulty of manually controlling the sUAV; and (2) low identifiability of the GCP targets due to blurring of imagery. The results showed that the DSM had centimeter accuracy (horizontal RMSE = 0.077m, vertical RMSE = 0.063m), which was accurate enough for educational purposes but fell short of our expectations. To achieve higher accuracy, we need further considerations on the shutter speed and F-number of digita