UAV position estimation through image matching between drone imagery and maps

Abstract

Accurate position estimation is a fundamental requirement for the autonomous navigation of unmanned aerial vehicles (UAVs). Most modern drones rely on Global Navigation Satellite Systems (GNSS), such as GPS, to determine their geographical position. However, satellite navigation can become unreliable in environments with signal obstruction, interference, or poor satellite visibility. An alternative approach is visual localization, which estimates the position of a drone by comparing images captured by its onboard camera with reference maps or georeferenced satellite imagery. This paper presents the principles and methods of UAV position estimation through image matching between drone imagery and maps. The main stages of the process include image acquisition, preprocessing, feature extraction, image matching, geometric transformation, and geographical coordinate estimation. Particular attention is given to traditional computer vision techniques, such as Scale-Invariant Feature Transform (SIFT), Oriented FAST and Rotated BRIEF (ORB), and feature matching using geometric models. The paper also discusses deep learning-based approaches, image registration, and the challenges associated with differences in viewing angle, image resolution, illumination, and environmental conditions. Finally, evaluation metrics and possible applications are examined. Visual localization can provide an additional source of positioning information and may complement conventional navigation systems. Its effectiveness, however, depends on the quality of reference maps, the reliability of image correspondences, and the accuracy of the geometric transformation.