Centre de documentation
scientifique

Attributierte Grammatiken zur Rekonstruktion und Interpretation von Fassaden / J. Schmidtwilken (2012) 

Public Titre : Attributierte Grammatiken zur Rekonstruktion und Interpretation von Fassaden Titre original : [Grammaires attribuées pour la reconstruction et l'interprétation des façades] Type de document : Thèse/HDR Auteurs : J. Schmidtwilken, Auteur Editeur : Munich : Bayerische Akademie der Wissenschaften Année de publication : 2012 Collection : DGK - C Sous-collection : Dissertationen num. 686 Importance : 169 p. Format : 21 x 30 cm ISBN/ISSN/EAN : 978-3-7696-5098-3 Note générale : Bibliographie Langues : Allemand (ger) Descripteur : [Vedettes matières IGN] Lasergrammétrie [Termes IGN] données lidar [Termes IGN] données localisées 3D [Termes IGN] façade [Termes IGN] objet géographique complexe [Termes IGN] reconstruction 3D du bâti [Termes IGN] semis de points Résumé : (Auteur) The needs and benefits of highly detailed 3D city and building models is opposed to their expensive acquisition. Though the automatic derivation of simplified roof structures from aerial images or laser altimetry is mostly solved, only a few approaches for the automatic reconstruction of facade structures from terrestrial data were presented yet. Most of the existing work is on the interpretation of large, planar facades that are reduced to a regular arrangement of windows. Therefore, it lacks in both, the reconstruction of additional facade elements such as stairs or doors as well as methods for the interpretation of heterogeneous a,nd complex facades with protrusions, oriels and only a small number of windows. This thesis deals with the interpretation of 3D point clouds that have been captured by terrestrial laser scanning of urban areas. The focus is on 3D point clouds of heterogeneous facades, that are characterized by their narrow width and by protrusions and oriels. The main contribution is the reconstruction and interpretation of facades by a parser that combines attribute grammars and robust estimators. Due to the generic attribute grammar the estimation of models with a previously unknown number of parameters is possible and the heterogeneity of the objects is thus considered. The concept is distinguished by the heavily integration of prior knowledge. The geometric and the semantic model of facades are defined and transferred to the attributed grammar. Furthermore, the probability densities of the shape and position parameters of the considered facade parts are estimated and Incorporated in the attributed grammar. The corresponding a priori probabilities are essential for the presented robust estimator MOSAP and the parsing algorithm parseSd. The relations between single parts of a facade are mapped to the attribute grammar. Therefore, the grammar represents the complex structures of facades and constitutes the foundation of the parsing algorithm. Beginning with the most common symbol parse3d iteratively applies production rules and, thus, generates hypothesis represented by derivation trees. Finally the most likely derivation tree is selected. Both, the selection of the symbol to be derived and the selection of the production rule to be applied base on the a priori probabilities and the densities that are estimated from the context information like the parameters of already reconstructed objects or subsets of the 3D point cloud. The interface between the grammar and the 3D point clouds is realized by the robust estimator MOSAP and guards. The latter determine the likelihood of a production rule and, therefore, provide a selection criterion. The symbols of the grammar t.iat are geometrically defined by a small set of parameters axe reconstructed by a robust estimator. MOSAP (Model Based Sampling and Prediction) bases on the hypothesize-and-verify paradigm and is characterized by the model-based sampling strategy that takes advantage of the probability densities of the shape and position parameters of the objects to be reconstructed. Furthermore, MOSAP evaluates the hypothesis by the newly introduced goodness of prediction criterion. Numéro de notice : 14499 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=62665

Documents numériques

en open access 14499_dgk-c-686.pdf Adobe Acrobat PDF