The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLI-B5
https://doi.org/10.5194/isprs-archives-XLI-B5-909-2016
https://doi.org/10.5194/isprs-archives-XLI-B5-909-2016
16 Jun 2016
 | 16 Jun 2016

THE POTENTIAL OF LOW-COST RPAS FOR MULTI-VIEW RECONSTRUCTION OF SUB-VERTICAL ROCK FACES

K. Thoeni, D. E. Guccione, M. Santise, A. Giacomini, R. Roncella, and G. Forlani

Keywords: UAV, Close Range Photogrammetry, Structure from Motion, Action Camera, Calibration, Accuracy, Coded Targets, Natural Features

Abstract. The current work investigates the potential of two low-cost off-the-shelf quadcopters for multi-view reconstruction of sub-vertical rock faces. The two platforms used are a DJI Phantom 1 equipped with a Gopro Hero 3+ Black and a DJI Phantom 3 Professional with integrated camera. The study area is a small sub-vertical rock face. Several flights were performed with both cameras set in time-lapse mode. Hence, images were taken automatically but the flights were performed manually as the investigated rock face is very irregular which required manual adjustment of the yaw and roll for optimal coverage. The digital images were processed with commercial SfM software packages. Several processing settings were investigated in order to find out the one providing the most accurate 3D reconstruction of the rock face. To this aim, all 3D models produced with both platforms are compared to a point cloud obtained with a terrestrial laser scanner. Firstly, the difference between the use of coded ground control targets and the use of natural features was studied. Coded targets generally provide the best accuracy, but they need to be placed on the surface, which is not always possible, as sub-vertical rock faces are not easily accessible. Nevertheless, natural features can provide a good alternative if wisely chosen as shown in this work. Secondly, the influence of using fixed interior orientation parameters or self-calibration was investigated. The results show that, in the case of the used sensors and camera networks, self-calibration provides better results. To support such empirical finding, a numerical investigation using a Monte Carlo simulation was performed.