3D Modeling of Ultra-High-Resolution UAV Imagery using Low-Cost UAVs and Structure from Motion

Keywords: UAS, drone, accuracy, low-cost, structure-from-motion
Session Type: Virtual Poster
Day: Saturday
Session Start / End Time: 4/10/2021 03:05 PM (Pacific Time (US & Canada)) - 4/10/2021 04:20 PM (Pacific Time (US & Canada))
Room: Virtual 52

Abstract

The availability of advanced low-cost unmanned aerial systems (UASs), aftermarket applications, and a competitive market for processing software have provided researchers new opportunities for employing high resolution remote sensing in research. The UAS allows for the capture of close range aerial imagery that can then be used to generate dense point clouds using structure from motion (SfM). A variety of digital products can be created form these dense point clouds such as three-dimensional models, digital elevation models (DEMs), digital surface models (DSMs) and orthomosaics. This research looks at methods and accuracies associated with the creation of digital mapping products from dense point clouds generated from imagery captured by two low-cost UASs.
The UASs were used to capture imagery over a 2-hectare vineyard in the Uwharrie mountains of North Carolina. Aspects of imagery collection, such as altitude, ground control, camera types, flight paths, and target styles, were investigated for their impacts on accuracy. Thirty-one ground control points were created in the vineyard using a survey grade GNSS receiver and total station for use in georeferencing. The number of ground control points used for georeferencing were reduced until a significant difference in accuracy was found using t-tests.
Five ground control points were shown to be the least amount of ground control needed before accuracy began to change significantly. Four flight altitudes were tested with 80-meters generating the least error. Orthomosaics created from structure from motion and imagery collected using a 20-megapixel global shutter camera had total RMS errors between 2-4 cm.