Geo-localisation of aerial images captured with Drones

The Global Position System (GPS) has become an essential sensor for public applications, maritime systems, robotics and aerial vehicles. Traditionally, autonomous flight in outdoor areas is possible thanks to GPS devices that enable Unmanned Aerial Vehicles (UAV) to obtain their position in latitude...

Full description

Bibliographic Details
Author: Aldrich Alfredo Cabrera Ponce
Format: master thesis
Status:Versión aceptada para publicación
Publication Date:2021
Country:México
Institution:Instituto Nacional de Astrofísica, Óptica y Electrónica
Repository:Repositorio Institucional del INAOE
Language:English
OAI Identifier:oai:inaoe.repositorioinstitucional.mx:1009/2136
Online Access:http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/2136
Access Level:Open access
Keyword:info:eu-repo/classification/Inspec/Geo-localisation
info:eu-repo/classification/Inspec/Aerial Images
info:eu-repo/classification/Inspec/UAV Navigation
info:eu-repo/classification/Inspec/GPS
info:eu-repo/classification/Inspec/Deep Learning
info:eu-repo/classification/cti/1
info:eu-repo/classification/cti/12
info:eu-repo/classification/cti/1203
info:eu-repo/classification/cti/120323
Description
Summary:The Global Position System (GPS) has become an essential sensor for public applications, maritime systems, robotics and aerial vehicles. Traditionally, autonomous flight in outdoor areas is possible thanks to GPS devices that enable Unmanned Aerial Vehicles (UAV) to obtain their position in latitude and longitude coordinates. However, GPS may become unreliable when the drone flies in environments where the signal may get occluded. Malicious attacks may also compromise the GPS signal, aiming to block the signal or replace it with spurious data. Motivated by these scenarios, the proposed approach relies on a methodology to estimate the GPS position of a UAV using Convolutional Neural Networks (CNN) and a learningbased strategy. For the latter, we adopted two learning scenarios: 1) offline learning; 2) online learning, where we tackled the re-localisation and geo-localisation problem in scenarios where GPS devices fail. We argue that our approach could be used as a backup system to return the UAV home. Therefore, we performed tests with aerial images and videos captured with the Matrice 100 and Bebop 2 drones in two scenarios with different trajectories to demonstrate our approach using a compact CNN and online learning implementation. The presented experiments with CompactPN got an average error of 2.60 to 6.16 metres and a speed of 107.88 fps. Likewise, online learning implementation based on the AR1* method obtained an accuracy of 0.79 to 0.92 and a speed of 127.33 fps.