INVESTIGATIONS ON THE GEOMETRIC QUALITY OF CAMERAS FOR UAV APPLICATIONS USING THE HIGH PRECISION UAV TEST FIELD ZOLLERN COLLIERY

Abstract. The geodetic-photogrammetric test field at the industrial monument Zollern colliery in Dortmund offers a scenario for carrying out geometric and radiometric tests of UAV systems. The foundation for this builds a geodetic precision network (position and height accuracy approx. 2 mm) with a total of 45 ground control points, distributed over an area of approx. 7 hectares. Within the scope of a campaign carried out in autumn 2017, various UAV sensor systems were tested under comparable conditions. Within this paper geometric investigations of two current DJI cameras, Zenmuse X4S (20 Mpix) and X5S (20.8 Mpix), as well as a Phase One IXU 1000 (100 Mpix) are presented. While the Zenmuse cameras reflect the current state of development of the manufacturer DJI, the medium format camera system from Phase One is primarily settled in the classic aerial segment. However, the desire for increased measurement accuracy (e. g. for engineering applications) also makes such a high-performance sensor interesting for UAV applications.
In addition to the configuration of the test field, the system comparison requires identical parameters for flight planning, in particular image overlapping, a complete cross flight configuration at different flight altitudes and the definition of a uniform ground resolution (GSD = 14 mm).

The investigations show clear differences in the achievable quality of the cameras. Though the high-priced Phase One system shows the best results, the most cost-effective system, the Zenmuse X4S, delivers only slightly worse results. In contrast, the Zenmuse X5S performs significantly worse than the other systems, mainly resulting from the mechanically unstable camera concept with interchangeable lenses. Finally, the comparison of the software products Pix4D Pix4Dmapper, Inpho UASMaster by Trimble and Agisoft PhotoScan partly shows significant differences in the results of image orientation. In particular in settings with sparse GCP usage the results vary considerably, indicating different strategies on how the residuals are distributed and the datum is defined, mostly Pix4Dmapper outperforms the others. In better GCP configurations there is no significant difference between Pix4mapper and Agisoft PhotoScan, while UASMaster does never deliver the best results.