Aerial Imagery Registration Using Deep Learning for UAV Geolocalization

The proliferation of drones, also known as unmanned aerial vehicles (UAVs), has shifted from being used for military operations to being available for the public market. The UAV applications mentioned previously and the ones available in the current market are currently being controlled by radio communications through a controller, a software that plans the route, or by using the on-board camera. This chapter aims to demonstrate the capability of using computer vision to geolocalize a UAV without using any inertial sensors or GPS but using only an on-board camera. The abundance of satellite and aerial high-resolution imagery covering most of the globe has facilitated the emergence of new applications such as autonomous vision-only UAV navigation. Having high-resolution aerial imagery from ISPRS of Potsdam inspired the idea of simulating a UAV footage of this area. Deep learning has been extensively used for aerial image analysis.