Unmanned Aerial Vehicles (UAVs) primarily rely on GPS-assisted localization and navigation due to the accessibility and ubiquity of such systems. However, this presents a potentially catastrophic single point of failure that may prevent autonomous UAVs from becoming truly reliable, as GPS is prone to dropout, spoofing, and inaccuracy. Thus, there is a need for onboard sensor-assisted localization in order to ensure autonomy in all settings. Visual sensing is well-suited for use with UAVs because cameras are low-weight, low-power, and low-cost. However, traditional visual localization methods are brittle to changes in lighting, season, and weather due to their reliance on local features which can vary drastically in outdoor environments. Visual Place Recognition (VPR) methods, on the other hand, have been shown to perform well under stark visual changes as they work with pooled global features which are better for capturing the high level structure of an image.

In this thesis, we present a VPR-based technique for accurate and robust large-scale UAV localization. Our technique utilizes a VPR framework that allows for quick and accurate database queries. Additionally, our method is generalizable and robust, such that it is able to localize across domains and in unseen and unvisited environments. This method also utilizes an algorithm for minimizing false positive associations during test time in order to greatly decrease convergence time and increase localization accuracy. We demonstrate this method on large-scale real-world trajectories, and show how it performs in global re-localization and online localization tasks.