An assessment of UAV-generated digital elevation model using ground surveying techniques

Abstract

The use of traditional surveying methods and instruments is common for the geodetic survey of the terrain surface. New technologies are currently being developed and utilised, for example, Unmanned Aerial Vehicles (UAVs) and their combination with a digital camera, also bring new opportunities to document earth’s surface. This combination of technologies allows for low-cost digital photogrammetry to document the earth’s surface in relation to already tested and adopted technologies. This thesis aims to evaluate the accuracy of the Digital Elevation Model (DEM) generated from aerial imagery obtained using low-cost UAVs by comparing it with the results for ground surveying techniques. The open field in the District 6 area in Cape Town was chosen as the test area. This open space field with its morphologically dissected surface and undulating terrain is ideal for testing the use of UAV photogrammetry to capture intricate surface details. The drone-based elevations were derived using UAV drone computer vision techniques. Using the UAV drone, the re-constructed camera positions and terrain features were used to derive ultra-high-resolution point clouds, ortho-photos and digital surface models from the multi-view UAV camera photos. A total number of 159 common points were compared, and a Root Mean Square Error (RMSE) of 0.338m (at 100m flight height), 0.078m (at 120m flight height) and 0.265m (at 140m flight height) for the vertical differences between the co-ordinates measured with the total station and the co-ordinates measured with a UAV drone were obtained. These residuals varied based on the elevations of the selected test points. It was found that the same points gave a RMSE of 0.050m (at 100m flight height), 0.046m (at 120m flight height) and 0.051 (at 140m flight height) on the X positions and a RMSE of 0.040m (at 100m flight height), 0.038m (at 120m flight height) and 0.052m (at 140m flight height) on the Y position for horizontal differences. The research conducted demonstrated that the accuracy of the model created from UAV photogrammetric data is similar to that of traditional ground survey techniques. From the selection of the test consisting of 159 points, nine points across all the 3 flights (100, 120 and 140 m) failed the conditions for the accuracy of the detailed points. These results indicate that combining a low-cost UAV and a digital camera may be a viable alternative for collecting data to document surface structures and form three-dimensional (3D) models. It is evident that several factors, including the terrain under investigation, flight configuration, flying height, sensor specifications, ground control distribution have a role to play in the final survey accuracy.