Particle Filter-Based Localization Using Visual Feature Synchronization in GNSS-Denied Navigation

Autonomous navigation in Global Navigation Satellite System (GNSS)-denied environments remains a critical challenge, particularly for unmanned aerial vehicles (UAVs) operating in complex or contested areas. In this work, we propose a vision-aided localization framework that integrates a CycleGANbased domain adaptation technique with a Marginalized Particle Filter (MPF) for robust state estimation. The proposed approach addresses the domain discrepancy between UAV onboard camera images and reference satellite imagery, improving feature matching reliability. The MPF is employed to refine state estimates by leveraging visual feature correspondences while mitigating the limitations of inertial navigation system (INS) drift. The proposed method has been validated through real-world flight tests, during which the UAV trajectory estimation performance was analyzed under various visual domain conditions. These tests demonstrate the method's robustness, feature alignment, and computational feasibility for UAV applications.