Application of a direct georeferencing method of drone images for smart farming

In smart farming, information collected by autonomous devices needs to be related to its exact field location, that is, georeferenced. In this work, we study the applicability and accuracy of a simplified direct georeferencing method of drone images for typical smart farming applications, such as weed detection. Based solely on an affine homography, the method results in accuracies < 0.82 m measured at 30 m above ground level even in the presence of pronounced steepness when applied to a real-time kinematic (RTK) enterprise drone (DJI Matrice 300 RTK). Our method uses only single images and does not rely on feature matching; therefore, it is inexpensive to compute. Depending on the targeted use case, the proposed georeferencing method yields errors in the order of the object of interest’s dimensions, which we demonstrate for our envisioned use case of dock plant (‘Rumex’) detection on meadows. The method may be seen as an upper bound for georeferencing errors that can be applied easily to other drone systems.