Content area
Full text
ABSTRACT
Spatial 3D-reconstruction applications today use mostly fully automated methods based on Structure from Motion (SfM) and Multi-View Stereo (MVS) algorithms. In these cases it is emphasized that the camera calibration parameters can be approximate, i.e. downloaded from EXIF data or even avoided, to facilitate the use of uncalibrated cameras in everyday measuring tasks. However, the accuracy of the spatial reconstruction of the model, recorded with an uncalibrated camera can be significantly reduced. Especially if the number of control points is limited, serving only to determine the correct model scale. Sometimes, the orientation points are entirely omitted (which is also allowed in the reconstruction procedure) making the determination of the proper model scale almost impossible. In this paper, the impact of the accuracy of the predetermined camera calibration parameters on the accuracy of 3D models will be explored and presented. The impact of different shooting arrangements as well as the influence of control points and control lengths will be analyzed to minimize the effects of insufficiently known initial camera calibration elements.
Keywords: SfM, MVS, camera calibration, 3D-reconstruction
INTRODUCTION
Over the last decade, photogrammetric measurement methods find their application in almost all professions, where you need to perform a reliable spatial survey. Besides this, photogrammetric measurements are increasingly becoming an alternative to traditional geodetic measurements. This is because they are faster and easier to perform, and the results are received almost immediately [1]. Especially in the era of digital photogrammetry, the cheap and affordable data acquisition equipment supported by a high level of automation in photogrammetric software widespread the photogrammetric measuring methods among all professions outside geodesy. Unmanned Aerial Vehicle (UAV) has succeeded in attaining a position in the aerial image survey as a cheaper and more effective for 3-dimension-modeling compared to traditional photogrammetry and Light Detection and Ranging (LIDAR) [2]. Although there is a lot of articles about the use of photogrammetric methods i.e. in geology [3] and forestry [4] where the accuracy of measurement is explored and described, the impact of camera calibration on the accuracy of the final result is not discussed. However, the most widely used software for photogrammetric reconstruction, i.e. Agisoft Metashape or Pix4Dmapper by default perform camera calibration...