The paper addresses the long-range navigation challenge at high altitudes under GNSS-denied conditions, proposing an inertial-visual autonomous localization and navigation algorithm based on scene matching. This algorithm leverages SuperPoint and SuperGlue networks for extracting and matching feature points between real-time and reference images, resulting in the computation of the aircraft’s current absolute position based on the matching outcomes. Subsequently, Kalman filtering is employed to fuse visual positioning information with inertial navigation data, culminating in the derivation of the final navigation result. Validation using flight data demonstrates that the algorithm achieved an average positioning error of 112.8 m during 2-hour flights conducted at altitudes ranging from 1000 m to 4000 m, showcasing its capability for accurate positioning and navigation solely reliant on onboard visual sensors in GNSS-denied conditions.