A Lightweight Georeferencing Workflow for Dynamic UAV footage using Feature- matching and Minimal Drone Metadata

Abstract

The emergence of unmanned aerial vehicles (UAVs), commonly known as drones, has transformed aerial imaging and photogrammetry, offering a cost-effective and flexible alternative to traditional methods. While commercially available drones are useful and affordable, the metadata provided in flight logs often falls short for robust photogrammetric applications. To address this limitation, we propose a novel method for the automated georeferencing of UAV footage that combines a feature-matching algorithm, Scale Invariant Feature Transform (SIFT), with telemetry data. Our system begins by initializing the homography by matching an input frame with a reference orthomosaic with a known spatial projection. Subsequently, the homography of the following frames is adjusted using the translation component estimated from the drone's telemetry. For the drone’s rotation, the Oriented FAST and Rotated BRIEF (ORB) algorithm was utilized to detect changes between consecutive frames, allowing for reinitialization of the homography when needed. To quantify uncertainty and assess temporal dependence in frame-wise accuracy, a moving-block bootstrap (MBB) approach was employed for estimating confidence intervals. The proposed workflow is designed to be modular, meaning that the algorithms can be swapped out based on the data and conditions. Experimental results indicate that the method achieves sub-meter accuracy, with mean RMSE ranging from 54.9 to 95.9 centimeters