A method for generating an
augmented reality image (ARB), wherein: - a
real image (RB) of a real object is captured using a real camera; - an edge image of the real object is generated from the
real image (RB) of the real object, wherein the position of a
virtual image component (VIRT) relative to the
real image (RB) captured by the real camera is determined using the edge image; and - the
virtual image component (VIRT) is combined with at least a part of the real image (RB) to form an
augmented reality image (ARB), wherein, for markerless tracking of the edge image, - an initialization phase is provided for the initialization of the markerless tracking, in which an initialization edge image (KB) of the real object is generated depending on the
pose of the real object from the perspective of the real camera, by - providing a
database in which images of the object in different poses are stored.where each image is assigned a
pose, - a real initialization image (RB) of the real object is captured using the real camera, whereby the image from the
database that is most similar to the initialization image is automatically selected, and - the
pose assigned to the selected image from the
database is selected to generate the initialization edge image (KB) by transforming the edge image of the real object into the selected pose.