Comparison of various drone positioning systems when creating surface models

Abstract: There are many applications where 3D models of landscapes can be used, suchas determining volume of objects, inspecting buildings and planning of infrastructure.One common way of creating 3D models of a geographical area is totake overlapping geotagged photos with a drone and then perform an aerotriangulation(AT). The aerotriangulating software find common key points in theimages and create a 3D surface model of the area. This process requires the use ofground control points (GCPs), which are used to map the 3D model onto a globalcoordinate system. These GCPs consume a lot of time to place at the location,measure accurately with total station and manually pinpoint in several photos.The purpose of this study is to compare models created by images taken with differentGNSS-based drone positioning systems and investigate for example howmany GCPs are needed, how the GCPs should be placed, and how the accuracyof models vary between the drone positioning systems considering both relativeand absolute accuracy.Two data acquisition sessions were done where images from two differentlocations were collected. In the first session the use of regular GPS and the use ofa local reference station are used as positioning system for the drone, and in thesecond session network real time kinematics (RTK) is also used as a third kind ofpositioning system.From the produced 3D models there is no significant difference between modelsusing a local reference station and network RTK, but when only using GPS thevertical accuracy drops significantly which means that more GCPs are required inorder for the model to be accurate. The standard deviation for points in createdmodels is calculated in easting, northing and vertical for the coordinate differencesfor the three positioning methods. When no GCPs are used the absoluteaccuracy is drastically lowered to meter level accuracy. In conclusion, by usingnetwork RTK or a local reference station the same accuracy for the 3D model canbe acquired with much fewer GCPs than if stand-alone GPS is used. None ofthe positioning systems can fully replace GCPs when a high absolute accuracy isneeded. With a relative accuracy requirement of 10 cm or more, both networkRTK and use of a local reference station has the potential to provide such qualitythat a 3D model would not need GCPs.