Abstract: To address the challenge of UAV navigation in GPS-free and nighttime environments, this paper proposes an airborne autonomous localization and navigation method based on heterogenous image matching and inertial guidance fusion. The proposed method enables the alignment of infrared or visible aerial images with geo-referenced satellite images, and subsequently realizes autonomous localization estimation and heading correction based on the alignment results. The heterogenous image matching method in this paper consists of a fast structural feature-based image matching module and an end-to-end learning-based fine registration network. The former provides position corrections at approximately 20 Hz, while the latter enables high-precision position and heading refinement. A pose fusion filtering strategy is employed to integrate the position and heading information obtained from image matching with inertial navigation data, effectively suppressing the drift in inertial systems and ultimately improving navigation accuracy. The flight test shows that the proposed method can effectively improve the positioning accuracy and reliability compared with existing methods, and can reach real-time processing on the airborne embedded computing platform.

Key words: image matching, MEMS INS, computer vision, sensor integration, autonomous navigation