Robustness of Structure from Motion Accuracy/Precision Against the Non-Optimality in Analysis Settings: Case Study in Constant-Pitch Flight Design

Abstract

Unmanned aerial vehicle (UAV)-based photogrammetry that employs structure from motion (SfM) and multi-view stereo (MVS) has been widely used in many disciplines, particularly in topographic surveying. However, several factors can affect the accuracy and precision of these techniques, including the analysis settings of the SfM process. In this study, we evaluated the robustness of SfM accuracy and precision against the non-optimal analysis settings by employing 750 analysis settings of SfM for 15 sets of images taken at five different pitch angles and three distinct ground sample distances. Flights were performed over a 100×100 m2 flat surface using the constant-pitch flight design. The results demonstrated the robustness of 20° and 30° pitch angles against non-optimality in SfM settings, producing relatively small root mean square errors for validation points (no larger than 0.056 m). This indicates that using these pitch angles for the flight design helps avoid concern over the SfM settings. Conversely, constant-pitch shooting with a 10° pitch angle was found to be insufficient for accurate estimation of camera intrinsic parameters (focal length f), and shooting with a 40° pitch angle showed a high risk of pose estimation failure, depending on the analysis settings. These findings can be useful for practitioners and researchers to improve their future applications of UAV-based photogrammetry.