Abstract
Abstract. A small unmanned aerial vehicle (UAV) with survey-grade GNSS positioning is used to produce a point cloud for topographic mapping and 3D reconstruction. The objective of this study is to assess the accuracy of a UAV imagery-derived point cloud by comparing a point cloud generated by terrestrial laser scanning (TLS). Imagery was collected over a 320 m by 320 m area with undulating terrain, containing 80 ground control points. A SenseFly eBee Plus fixed-wing platform with PPK positioning with a 10.6 mm focal length and a 20 MP digital camera was used to fly the area. Pix4Dmapper, a computer vision based commercial software, was used to process a photogrammetric block, constrained by 5 GCPs while obtaining cm-level RMSE based on the remaining 75 checkpoints. Based on results of automatic aerial triangulation, a point cloud and digital surface model (DSM) (2.5 cm/pixel) are generated and their accuracy assessed. A bias less than 1 pixel was observed in elevations from the UAV DSM at the checkpoints. 31 registered TLS scans made up a point cloud of the same area with an observed horizontal root mean square error (RMSE) of 0.006m, and negligible vertical RMSE. Comparisons were made between fitted planes of extracted roof features of 2 buildings and centreline profile comparison of a road in both UAV and TLS point clouds. Comparisons showed an average +8 cm bias with UAV point cloud computing too high in two features. No bias was observed in the roof features of the southernmost building.
Highlights
Generating dense point clouds from photography from small unmanned aerial vehicles (UAV) has been an attractive method for mapping and 3D reconstruction due to its efficiency and largely automated workflow
This study focuses on the analysis of a dense point cloud derived from UAV imagery based on bundle block adjustment (BBA) results using five ground control points by comparison with a point cloud generated from terrestrial laser scanning (TLS)
UAV imagery processed at full image scale and adjusted in Pix4Dmapper using 5 GCP produced a root mean square error (RMSE) value of 1.70 cm horizontally based on 75 checkpoints
Summary
Generating dense point clouds from photography from small unmanned aerial vehicles (UAV) has been an attractive method for mapping and 3D reconstruction due to its efficiency and largely automated workflow. Reshetyuk and Mårtensson performed a study showing that a computer vision-based software resolved UAV imagery-derived elevations with higher certainty in flat terrain portions of a quarry site than a noncomputer vision-based software. They showed that the opposite was the case over undulating terrain (Reshetyuk and Mårtensson, 2016). A comparison was performed by a point-topoint difference analysis between clouds (Eling et al, 2016) Another comparison was made between DSM surfaces generated by data collected in sub-optimal positioning conditions where high-elevation terrain obstructions affected the establishment of ground control quality (Jaud et al, 2016). The study focused on interpolation and point filtration methods finding the UAV imagery-derived point cloud to be more accurate in all cases noting the more consistent point density over TLS method
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
