Sensitivity Analysis of a Visual Inertial Odometry-based Navigation System for UAV

This paper presents a detailed sensitivity analysis of a Visual-Inertial Odometry (VIO) framework designed for unmanned aerial vehicle (UAV) navigation, utilizing sensor fusion from cameras, LiDAR, altimeters, and inertial measurement units (IMUs). The analysis quantifies the impact of various sources of uncertainty, including environmental and lighting conditions, intrinsic sensor parameters, and algorithmic factors associated with keypoint detection and matching. Through Monte Carlo simulations, several keypoint detection techniques, specifically ORB and semantic segmentation, are evaluated across diverse operational scenarios. The results demonstrate that the analysed framework achieves reliable trajectory estimation, with both ORB and semantic segmentation providing accurate and consistent positioning. Notably, semantic segmentation achieves the lowest uncertainty under varying conditions. This study validates the proposed uncertainty model and identifies critical factors influencing the reliability of VIO-based UAV navigation, offering insights for enhancing system performance in real-world applications.