Robust Global Localization for UAVs in GNSS-denied Environments Using a Vision Foundation Model

Robust and accurate localization is an essential technology for the autonomous flight of unmanned aerial vehicles (UAVs). Many existing UAV localization methods rely heavily on GNSS signals, which are often unavailable or unreliable in urban environments due to signal blockage, jamming, or spoofing. To address these limitations, alternative localization approaches that operate in GNSS-denied environments have attracted significant interest. In particular, recent studies have explored methods that estimate global position by matching UAV imagery with geo-referenced databases. Specifically, matching UAV images with satellite images has emerged as a promising solution; however, significant visual differences between these two domains pose major challenges. This paper proposes a localization method designed for GNSS-denied environments, which employs a vision foundation model to match nadir-view UAV images with satellite imagery to estimate the UAV’s global position. To evaluate the performance of the proposed localization algorithm in GNSS-denied environments, we conducted experiments using two real-world aerial datasets.