Accuracy evaluation of 3D geometry from low-attitude UAV images: A case study at Zijin Mine

Abstract

This study investigates the usability of low-attitude unmanned aerial vehicle (UAV) acquiring high resolution images for the geometry reconstruction of opencast mine. Image modelling techniques like Structure from Motion (SfM) and Patch-based Multi-view Stereo (PMVS) algorithms are used to generate dense 3D point cloud from UAV collections. Then, precision of 3D point cloud will be first evaluated based on Real-time Kinematic (RTK) ground control points (GCPs) at point level. The experimental result shows that the mean square error of the UAV point cloud is 0.1 1m. Digital surface model (DSM) of the study area is generated from UAV point cloud, and compared with that from the Terrestrial Laser Scanner (TLS) data for further comparison at the surface level. Discrepancy map of 3D distances based on DSMs shows that most deviation is less than ±0.4m and the relative error of the volume is 1.55%.