Visual Localization system for GPS-Blind Environments in Unmanned Aerial Vehicles

Integrating computer vision with unmanned aerial vehicles (UAVs) has significantly reduced costs and increased accessibility across various applications, enabling advanced capabilities such as visual navigation, localization, and obstacle avoidance. However, current limitations in autonomous navigation systems hinder UAV operation in GPS-denied environments, posing challenges for fully autonomous flights. As the adoption of UAVs expands, developing robust backup strategies is crucial for ensuring safe and reliable navigation in the event of Global Navigation Satellite Systems (GNSS) failures or cyber-attacks. Visual localization emerges as a promising alternative, offering accuracy comparable to or exceeding that of GNSS-based systems. In this paper, we present a Visual Place Recognition-Based Navigation Subsystem for UAVs, specifically designed for GPS-denied environments at high altitudes. Our pipeline addresses key localization challenges, enabling precise positioning even without accurate initial coordinates. We validate the effectiveness of our approach through extensive testing on our proposed dataset “AeroVL”, which covers over 80 km<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of diverse and challenging landscapes, including urban areas, farmlands, and forests.