A method describes the architecture of a vision-supported navigation system in a terrain using available topographical maps, which is based on already existing methods for detecting prominent terrain features. An offline process used in this case creates a feature database from existing topographical maps, which is subsequently used in an online process for recognizing features that have been seen. In this case, virtual reference images of the expected situation are created using computer graphical methods, which are analyzed using methods for feature extraction, and a feature catalog made up of vision-based feature vectors and the associated 3D coordinates of the localized features is derived from the same. These data are stored in a flight system in the form of a model database, for example, and enable navigation close to a reference trajectory and for a planned lighting situation. High localization characteristics and a correspondingly high navigation accuracy are achieved by means of the possibility of using any desired, not necessarily lighting-invariant, feature extractors.