Model-based detection and localization of circular landmarks in aerial images

The photogrammetric exploitation of aerial images essentially requires the accurate reconstruction of the imaging geometry. This especially includes the determination of the orientation of the camera. Usually, the orientation parameters are determined by spatial resection, knowing the exact coordinates of control points on the ground and in the image. The reliability and accuracy of this registration task strongly depend on the selection of suitable landmarks as well as on the precision obtained for landmark localization. In this contribution, we consider the problem of automatic landmark extraction for the purpose of aerial image registration. We suggest to use manhole covers as a specific type of circular landmarks which frequently occur in urban environments and we introduce a model-based approach for localizing these features with high subpixel precision. Our approach is based on a parametric intensity model. Localization of the landmarks is done by directly fitting this model to the observed image intensities. Since we have an explicit description of the landmark it is possible to verify the result by exploiting the estimated parameters. We also address the problem of landmark detection which can greatly be supported by template matching. The template used is a prototype model which is generated from representative examples during a training phase. The training scheme also provides initial values for the fitting procedure as well as thresholds for the final verification step. The full approach has been tested on synthetic as well as on real image data.