Robust Autonomous Navigation Method for High-Precision UAV Based on Inertial/Machine Vision Fusion

Aiming at the problems of low accuracy and poor robustness of UAV visual navigation and localization in satellite denial environments, we propose a research of high-precision UAV robust autonomous navigation method based on inertia/machine vision fusion. The inertial information is used to orthorectify the UAV images, the positioning of UAV images in the satellite reference map is achieved based on the SuperPoint&SuperGlue algorithm, which effectively improves the positioning accuracy in different geographic environments, and the inertial/machine vision fusion navigation model is constructed to suppress the divergence of INS errors, remove visual navigation outliers, and maintain the real-time and continuity of navigation. In order to verify the effectiveness of the algorithm, a simulation method based on commercial satellite maps is innovatively proposed to generate UAV on-board datasets, which simulates the output of inertial sensors and images captured by visual sensor through the flight motion parameters and satellite maps to reduce the influence of factors such as sensor measurement and misalignment errors on the evaluation of the algorithm. Tests under three geographic environments, namely, urban, plain and mountain, are designed, and the results show that visual navigation provides a reference position with an error within 10 m in different geographic environments, and the integrated navigation algorithm substantially suppresses inertial error dispersion in all environments and exhibits good robustness, providing a new technological approach for high-precision autonomous navigation under satellite denial environments.