Camera Selection and Flight Planning for Post Processing 3D Reconstruction Automatization

Currently available 3D reconstruction and aerial mapping solutions based on the Structure-from-Motion (SfM) and bundle adjustment, require careful equipment selection/configuration (camera, optics and imaging parameters’ settings), accurate flight planning (number of images and their overlap percentage) and often user intervention during post processing. This paper studies the various factors affecting the quality of the final 3D reconstruction, their dependencies and trade-offs trying to push the whole 3D reconstruction and aerial mapping solutions to a more automated scheme with minimum user intervention. We address the design trade-offs between the camera and the flight plan parameters and their effect on the images’ quality and the number and quality of matched features consequently. Then we consider the camera calibration as a fundamental stage in the SfM framework that affects afterwards the depth estimation and point cloud densification steps. An automated photogrammetry calibration solution - based on the Caltech Camera Calibration Matlab Toolbox and Andreas Geiger corner detector - is implemented and tested, allowing for a user-free precise camera calibration process. Finally, a professional photogrammetry application software based on the SfM (PIX4Dmappper) is used to assess the use of additional data sets - like the commonly used GPS data – on the overall resulting 3D reconstruction quality in terms of model richness and accuracy. Two geo-localization data sets (standard GPS used by the UAV autopilot and an L1/L2 GPS corrected with PPK) were used with various stated uncertainty bounds processed with the same images’ data set. The comparison of the resulting 3D reconstructed models highlights the decisive importance of identifying the actual uncertainty bounds - combining the equipment nominal uncertainty with the system introduces factors like the camera capture/GPS synchronization- instead of using relaxed (larger) uncertainty bounds or tighter bounds.