Using low-cost UAVs in post-mining exploration - a case study

Abstract

The use of low-cost, unmanned aerial vehicles (UAVs) has been growing in many sectors. Due to the sufficient accuracy of the products acquired from UAVs, this technology has also been applied in geodesy and remote sensing. It results from many factors: the low prices of UAVs and the availability of different sensors and software applications, which allows for simple data processing. Due to the required high accuracy, the inventory of a mine is usually performed with the use of conventional surveying techniques, such as tacheometry. This paper discusses the possibilities of applying low-cost UAVs to inventory open-cut mining. Using Phantom 3 Professional equipped with a factory-made camera, RGB photographs were acquired, which were then processed using three commercial software applications: Pix4D, 3D Survey and Agisoft Metashape. Different algorithms for image orientation (Structure-from-Motion, SfM) and dense point generation (Multi-View Stereo, MVS) were implemented for each of those applications, which influenced the accuracy of the final products. The results of the experiments proved that the highest accuracy in terms of photograph processing was achieved using the Pix4D software. The mean difference between the DTM (Digital Terrain Model) generated from surveys, and the DTM generated from photographs using Pix4D was equal to 0.106 m. This paper compared the DTMs and the DSMs (Digital Surface Models) generated by the selected software applications. The models generated with the use of Pix4D were assumed as a reference. According to the analysis of the DTMs and the DSMs, the smallest differences were obtained for the models generated by Pix4D and Agisoft Metashape. They equalled 0.080 m for the DTM and 0.246 m for the DSM. The differences between the DSMs generated by Pix4D and 3D Survey were two times bigger; the differences between the DTMs generated by those software applications were six times bigger. The differences between the models may result from the presence of vegetation and escarpments at the edges of the test site and different algorithms for generating dense point clouds applied in particular applications.