Centre de documentation
scientifique

Oberflächenrekonstruktion mit Hilfe einer Mehrbild-Shape-from-Shading-Methode / C. Piechullek (2000)

Public Titre : Oberflächenrekonstruktion mit Hilfe einer Mehrbild-Shape-from-Shading-Methode Titre original : [Reconstruction de surfaces au moyen de la méthode multi-image Shape-From-Shading] Type de document : Thèse/HDR Auteurs : C. Piechullek, Auteur Editeur : Munich : Bayerische Akademie der Wissenschaften Année de publication : 2000 Collection : DGK - C Sous-collection : Dissertationen num. 518 Importance : 98 p. Format : 21 x 30 cm ISBN/ISSN/EAN : 978-3-7696-9557-1 Note générale : Bibliographie Langues : Allemand (ger) Descripteur : [Vedettes matières IGN] Traitement d'image [Termes IGN] filtre passe-bas [Termes IGN] modèle numérique de sursol [Termes IGN] orthoimage [Termes IGN] reconstruction 2D du bâti [Termes IGN] réflectance Index. décimale : 35.20 Traitement d'image Résumé : (Auteur) Shape from Shading is a method for surface reconstruction from digital images which exploits the fact that surface patches, having different orientations relative to a light source, are imaged with different brightness in the images. The resulting two-dimensional brightness distribution is denoted as shading and is used for the subsequent reconstruction of the surface inclinations by suitable solution methods. In this thesis a novel multi image Shape from Shading method for the reconstruction of surfaces from multiple digital images is presented and investigated. Starting from a model for the formation of a digital image, the image grey values are related to the geometric and radiometric parameters of a surface model defined in object space. The direct computation of absolute surface heights of a digital terrain model (DTM) is accomplished by moving the object reconstruction process into object space, together with the incorporation of the perspective imaging geometry. The surface model ensures the correspondence between multiple images with low texture. The unknown DTM heights and the surface reflectance coefficient are estimated in a least squares adjustment. In comparison to existing Shape from Shading methods, the simultaneous evaluation of multiple images on the basis of a surface model has the advantage of circumventing the inherent indeterminability of the surface inclinations. The mesh heights of a DTM are reconstructed, rather than both inclination components of all surface elements. All surface elements which lie within adjacent DTM meshes contribute to the determination of a particular DTM height. The assumption of completely diffuse reflectance of the surface is generalised by incorporating a surface reflectance model which also takes into account the observation directions of each image. The potential of the method for the reconstruction of surfaces is investigated by theoretical considerations and accompanying simulation studies. This is accomplished by varying the spatial arrangement between surface normal, illumination direction and the observation directions, as well as by varying the number of available images, and the surface reflectance properties. It is shown that the surface reconstruction is possible, if the grey value variations within the images are greater than errors which occur during the acquisition and processing of digital imagery. Thus, planar surfaces, having no or comparatively small grey value variations cannot be reconstructed by the presented method. The indeterminability which occurs when evaluating single images with a lack of grey value gradients due to vertial observation direction or parallel illumination direction and surface normal, respectively, is overcome by the use of at least two images with different exterior orientation. In practical simulation studies using synthetic images which have been created under the imaging and illumination conditions of the real test example, the convergence behaviour of the method is investigated. The radius of convergence is found to be comparably high (about 20 pixels in image space), but is seriously reduced by image noise. The convergence behaviour can be improved considerably. however, by low-pass filtering the images. In investigations with radiometrically distorted synthetic images it is shown that systematic errors within the reconstructed surface shape may be reduced by the simultaneous evaluation of multiple images, since there is additional, independent grey value information available. In practical tests of the method, a set of three digitised aerial images (Mb ~ 1 : 11000) of a low-textured sand dune area are evaluated. By referring to the results of the simulation studies, the images are radiometrically corrected prior to the surface reconstruction. The radius of convergence is found to be about 16 pixels for the real imagery; thus, the independence of the obtained results from the available initial information can be assured to a large extent. The results obtained from the simultaneous evaluation of the radiometrically preprocessed images clearly demonstrate the applicability of the method to real imagery. Remaining systematic differences between the reconstruction results and the reference surface, however, indicate that grey values which have been obtained by the digitalisation of analogue images are not fully described by the underlying model for the formation of a digital image. A concluding assessment of the correctness of all assumptions concerning the reflectance properties of the imaged surface is only possible by the evaluation of images which have been acquired digitally by a radiometrically calibrated sensor. Numéro de notice : 62964 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=61153

Exemplaires(1)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
62964-01	35.20	Livre	Centre de documentation	Télédétection	Disponible