GNSS-denied geolocalization of UAVs by visual matching of onboard camera images with orthophotos

TL;DR

This work addresses GNSS-denied UAV geolocalization by fusing visual-inertial odometry with orthorectification and map matching in a Monte-Carlo localization framework, avoiding the need for a downward-facing camera. The authors reduce the state to $X_t=(x(t),y(t),\phi(t),s(t))$ and propagate 1000 particles, using a locally planar ground model to generate orthoprojections for image-to-map comparison. They systematically evaluate classical image similarity metrics to select a robust matching score, map this score to a posterior pose probability via KDE-based distributions, and demonstrate convergence for three UAV datasets with modest initialization and 1 m/pixel orthophotos. The approach yields global localization with mean errors competitive with, and often better than, odometry in many scenarios, while illustrating the impact of appearance changes and terrain ambiguity on reliability. The results suggest practical GNSS-denied localization is feasible for small UAVs using a single, arbitrarily oriented camera and pre-existing orthophotos, enabling broader mission capabilities without specialized downward optics.

Abstract

Localization of low-cost Unmanned Aerial Vehicles (UAVs) often relies on Global Navigation Satellite Systems (GNSS). GNSS are susceptible to both natural disruptions to radio signal and intentional jamming and spoofing by an adversary. A typical way to provide georeferenced localization without GNSS for small UAVs is to have a downward-facing camera and match camera images to a map. The downward-facing camera adds cost, size, and weight to the UAV platform and the orientation limits its usability for other purposes. In this work, we propose a Monte-Carlo localization method for georeferenced localization of an UAV requiring no infrastructure using only inertial measurements, a camera facing an arbitrary direction, and an orthoimage map. We perform orthorectification of the UAV image, relying on a local planarity assumption of the environment, relaxing the requirement of downward-pointing camera. We propose a measure of goodness for the matching score of an orthorectified UAV image and a map. We demonstrate that the system is able to localize globally an UAV with modest requirements for initialization and map resolution.

Figures (7)

• Figure 1: Proposed geolocalization framework: Visual-inertial odometry is used for tracking uav motion and visual landmarks, and solving landmark 3D positions with respect to uav pose. uav camera image is projected through planar homography to a virtual view looking straight down. The orthoprojected image is used for map matching and pose tracking in a particle filter framework.
• Figure 2: Block diagram of proposed solution
• Figure 3: Feature point locations (green dots), plane fit to points (blue grid), camera center (red star), and intersection points of rays from camera center through image corner coordinates with plane ($p_{ll}$, $p_{ul}$, $p_{ur}$, $p_{lr}$) for last camera pose in batch, shown in frame $\{B_0\}$ (best viewed in color).
• Figure 4: Original camera image (\ref{['fig:undistorted_original_image']}) and orthoprojection (\ref{['fig:orthoprojection_example']}) of the same image.
• Figure 5: Characterization of one of the matching scores: Moravec