Visual Localization for UAVs using Satellite Images

The primary objective of this thesis is to develop a method to generate offline a map containing only Google Earth (GE) rendered images against which an Unmanned Aerial Vehicle (UAV) is able to localize using vision online. This eliminates the need for a manual mapping flight and would allow the UAV to fly in new unseen areas autonomously. The first part of this thesis presents a novel method for localizing live UAV images against a collection of rendered and encoded GE images that is fast and has low storage requirements. The second part implements this new measurement into the existing Visual Teach \ Repeat (VT\) system. This includes writing a script to generate a compatible map, computing the measurement in the VT\ pipeline, and using it in the localization optimization. We demonstrate the performance on a dataset of real UAV collected images at six different times of day.M.A.S