Precise Georeferencing of RGB and Multispectral Drone Images without Ground Control Points

Abstract. Drones equipped with multispectral cameras are increasingly used in environmental studies. These surveys often take place in challenging locations such as mudflats, where marking and surveying ground control points (GCPs) is difficult, making the use of RTK-enabled drones essential. This paper analyses the achievable accuracy of orthomosaics (RGB and multi-spectral) and digital elevation models (DEMs) obtained with a drone, in order to carry regular surveys in estuarine areas in northern Portugal and Galicia, Spain. Accurate geolocation, particularly in the horizontal component, is crucial for combining data from different epochs, often under significantly changing conditions. Vertical accuracy is also important to assess tidal effects. In this study, a DJI Mavic 3M drone was used. Images were captured with a nadir orientation and supplemented with a final strip taken with the camera tilted, to optimize focal length self-calibration. After 21 survey flights, the stability of the interior orientation parameters was evaluated. The focal length varied within a range of 9 pixels, with a standard deviation of 2.2 pixels (0.06% of the focal length). Given the open question of whether there exists a single set of optimal calibration parameters for all cases, or whether, for example, due to environmental factors, the parameters, particularly the focal length, may vary, tests were conducted in a built-up area of approximately 30 hectares, where many GCPs could be surveyed. It was observed that each flight should have its calibration recalculated. The results showed that calibration based solely on the inclusion of oblique images and without the use of control points produced good results, with horizontal errors at the level of the image resolution and vertical error twice as that. The photogrammetric processing of the multispectral bands was carried out separately, and it was possible to conclude that georeferencing with positional accuracy comparable to the ground resolution can also be achieved, when rolling shutter compensation is applied. Results proved that the methodology followed for estuary surveys is appropriate to achieve results with accurate geolocation.