Geo-trax
Extract georeferenced vehicle trajectories from drone videos - by Robert Fonod
Urban traffic monitoring increasingly relies on detailed vehicle trajectory data, yet traditional ground-based sensors are often costly, inflexible, and limited in coverage. Drone-based monitoring offers a scalable alternative, but extracting high-quality data from aerial footage is challenging due to drone motion, perspective distortions, and the need for accurate georeferencing.
Our mission with Geo-trax (GEO-referenced TRAjectory eXtraction) is to transform raw, high-altitude drone videos into precise, real-world vehicle trajectories. The pipeline integrates state-of-the-art computer vision and deep learning methods for vehicle detection and tracking, a novel stabilization method that eliminates drone motion artifacts, and a georeferencing step that aligns trajectories with a high-resolution orthophoto.
The result is a framework that delivers spatially and temporally consistent data for traffic analysis, simulation, and digital twin applications.
Geo-trax is open source and designed to be accessible to both researchers and practitioners. The repository includes ready-to-use scripts, configuration options, and visualization tools. Users can process drone videos into structured datasets enriched with vehicle-level metadata, including speed, acceleration, and dimension estimates. The framework leverages the dedicated stabilization library Stabilo and its benchmarking companion Stabilo-Optimize to ensure robust and well-tuned motion compensation.
The framework has been validated in a large-scale experiment in Songdo, South Korea, where 10 drones captured 12TB of 4K video, producing the Songdo Traffic dataset (∼700,000 trajectories) and the Songdo Vision dataset (∼300,000 annotated vehicles). By releasing the full source code and datasets, Geo-trax sets new benchmarks for reproducibility and scalability in drone-based traffic research, enabling intelligent transportation systems and next-generation mobility studies.
A divide-and-conquer approach to plan under constraints - by Panagiotis Grontas