Fast and efficient geo-referencing for drone-based wildlife tracking

The study of animal behaviour and movement has been greatly enhanced by drones, but we have yet to harness their full potential in tracking animal groups. Georeferencing provides an exciting opportunity to expand the animal tracking toolbox. Here, we demonstrate a simple mathematical translation of feature pixel positions in drone videos into latitude and longitude positions in real-world space using information on the drone position, height and camera perspective. Validation using ground control points suggested a median error of 1.5 m (95% CI  0.5-2.8) with a downward facing camera. Field tests tracking a GPS-equipped volunteer showed a realised error of 2.38 m (95% CI 0.47-5.63), which is comparable to current state-of-the-art GPS loggers. A strength of our mathematically derived approach is its robustness to varying camera angles and drone movement. The georeferencing error increases by only 0.1 cm per meter increase in drone height and 0.6 cm per degree increase in the camera angle. Integrating this procedure with deep learning enables automated tracking of free-ranging animals, making it a powerful tool for studying fine-scale behaviour in diverse ecological settings. We discuss the application of our method to free-ranging animals, provide practical guidelines for its implementation, and facilitate its implementation through a simple Graphical User Interface our Python package FastGeoRef.