Détail de l'auteur
Auteur F. Quint
Documents disponibles écrits par cet auteur
Ajouter le résultat dans votre panier Affiner la recherche Interroger des sources externes
Kartengestützte Interpretation monokularer Luftbilder / F. Quint (1997)
Titre : Kartengestützte Interpretation monokularer Luftbilder Titre original : [Interprétation de photographies aériennes monoculaires fondée sur les cartes] Type de document : Thèse/HDR Auteurs : F. Quint, Auteur Editeur : Munich : Bayerische Akademie der Wissenschaften Année de publication : 1997 Collection : DGK - C Sous-collection : Dissertationen num. 477 Importance : 105 p. Format : 21 x 30 cm ISBN/ISSN/EAN : 978-3-7696-9517-5 Note générale : Bibliographie Langues : Allemand (ger) Descripteur : [Vedettes matières IGN] Photogrammétrie numérique
[Termes descripteurs IGN] carte topographique
[Termes descripteurs IGN] extraction automatique
[Termes descripteurs IGN] grande échelle
[Termes descripteurs IGN] image aérienne
[Termes descripteurs IGN] image en couleur
[Termes descripteurs IGN] modèle orienté objet
[Termes descripteurs IGN] segmentation d'image
[Termes descripteurs IGN] système expert
[Termes descripteurs IGN] théorie de Dempster-Shafer
Résumé : (Auteur) Information gained from aerial and satellite images is widely used in industry and public life. Thus, the inter-national community supports big efforts to automate the image analysis process and to increase the reliability of the results. In this context we aim to develop a knowledge based system for the interpretation of monocular aerial images. By using knowledge gained from topographical maps, the system shall be able to automatically build a description of complex urban scenes. As input data large scale colour aerial images, digitised to a pixel size of 30 cm x 30 cm on the ground, and topographical maps with scale 1:5000 are used.
A central component of the system are the object models. In order to adequately represent the available knowledge and to assure the flexibility of the system we build several models. They are arranged in a model hierarchy and we represent them in semantic networks. The most general model is the generic scene model. From it we derive the map-domain model and the image-domain model. They describe the projection of a scene in a topographic map and in an aerial image, respectively.
The models are used in the four steps of our analysis process. In the first step the map-domain model is used to analyse the map. As result, a structural description of the scene, as far as it is represented in the map, is built. As soon as digital maps already provide a structural scene description, this step becomes obsolete. One can expect that such maps will be available in the future. In the second step the scene description gained after the first step is automatically combined with the image-domain model. As result one obtains a model which is specific for the current scene to be analysed. In the third step line segments and regions are calculated from the aerial image. The line segments are calculated with a gradient based procedure. The regions are gained by segmenting J;he image with a region growing procedure. Starting points for the region growing are extracted from the specific model. In the segmentation process we use a new homogeneity predicate based on the a-posteriori probability of the pixels to fulfil a region model. The proposed region models cover most of the practical relevant cases and allow an efficient computation of the homogeneity predicate.
The specific model is used again in the fourth step. This step is performed in three phases. In the first phase the objects represented in the map are verified in the image data. The knowledge gained in this phase is used as context in the following phases. In the second phase the description of the verified objects is improved by extracting additional attributes from the image data. Finally, in the third phase those areas of the scene are addressed, for which the map does not provide any information or for which in the first phase the verification has failed.
The analysis is performed in all three phases by using the same problem independent control algorithm. It implements the analysis as a search process. We use the E-A*-algorithm to conduct this search. The merit functions for the search algorithm have been embedded in the Dempster-Shafer-theory of evidence and to calculate them we have developed a new method to propagate evidence in an hierarchical environment. We discuss the performance of our system and show its applicability not only for the recognition of single object classes or single objects, but for the recognition and reconstruction of buildings of different size and shape, of parking areas, cars and vegetation areas.
Numéro de notice : 28013 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Permalink :
RéservationRéserver ce document
Code-barres Cote Support Localisation Section Disponibilité 28013-01 33.30 Livre Centre de documentation Photogrammétrie - Lasergrammétrie Disponible 28013-02 33.30 Livre Centre de documentation Photogrammétrie - Lasergrammétrie Disponible