• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 1
  • 1
  • Tagged with
  • 2
  • 2
  • 2
  • 2
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
1

Reconstruction automatique de maquettes numériques 3D / Automatic reconstruction of 3D digital buildings

Boulch, Alexandre 19 December 2014 (has links)
La maquette numérique de bâtiment est un outil nouveau et en plein essor dans les métiers de la construction. Elle centralise les informations et facilite la communication entre les acteurs : évaluation des coûts, simulations physiques, présentations virtuelles, suivis de travaux, etc. Si une maquette numérique est désormais utilisée pour les grands chantiers de bâtiments nouveaux, il n'en existe pas en revanche pour la plupart des bâtiments déjà construits. Or, avec le vieillissement du parc immobilier et le développement du marché de la rénovation, la maquette numérique serait une aide considérable pour des bâtiments anciens. Des techniques de reconstruction plus ou moins automatique ont été développées ces dernières années, à base de mesures laser ou de photogrammétrie. Les lasers, précis et denses, sont chers mais restent abordables pour les industriels, tandis que la photogrammétrie, souvent moins précise et moins fiable dans les zones uniformes (p.ex. les murs), est beaucoup plus bon marché. Mais la plupart des approches s'arrêtent à la reconstruction de surfaces, sans produire de maquettes numériques. À la géométrie doit cependant s'ajouter des informations sémantiques décrivant les éléments de la scène. L'objectif de cette thèse est de fournir un cadre de reconstruction de maquettes numériques, à la fois en ce qui concerne la géométrie et la sémantique, à partir de nuages de points. Pour cela, plusieurs étapes sont proposées. Elles s'inscrivent dans un processus d'enrichissement des données, depuis les points jusqu'à la maquette numérique finale. Dans un premier temps, un estimateur de normales pour les nuages de points est défini. Basé sur une transformée de Hough robuste, il permet de retrouver correctement les normales, y compris dans les zones anguleuses et s'adapte à l'anisotropie des données. Dans un second temps, des primitives géométriques sont extraites du nuage de points avec leur normales. Afin d'identifier les primitives identiques existantes en cas de sur-segmentation, nous développons un critère statistique robuste et général pour l'identification de formes, ne requérant qu'une fonction distance entre points et formes. Ensuite, une surface planaire par morceaux est reconstruite. Des hypothèses de plans pour les zones visibles et les parties cachées sont générées et insérées dans un arrangement. La surface est extraite avec une nouvelle régularisation sur le nombre de coins et la longueur des arêtes. L'utilisation d'une formulation linéaire permet, après relaxation continue, d'extraire efficacement une surface proche de l'optimum. Enfin, nous proposons une approche basée sur des grammaires attribuées avec contraintes pour l'enrichissement sémantique de modèles 3D. Cette méthode est bottom-up : nous partons des données pour construire des objets de complexité croissante. La possible explosion combinatoire est gérée efficacement via l'introduction d'opérateurs maximaux et d'un ordre pour l'instanciation des variables / The interest for digital models in the building industry is growing rapidly. These centralize all the information concerning the building and facilitate communication between the players of construction : cost evaluation, physical simulations, virtual presentations, building lifecycle management, site supervision, etc. Although building models now tend to be used for large projects of new constructions, there is no such models for existing building. In particular, old buildings do not enjoy digital 3D model and information whereas they would benefit the most from them, e.g., to plan cost-effective renovation that achieves good thermal performance. Such 3D models are reconstructed from the real building. Lately a number of automatic reconstruction methods have been developed either from laser or photogrammetric data. Lasers are precise and produce dense point clouds. Their price have greatly reduced in the past few years, making them affordable for industries. Photogrammetry, often less precise and failing in uniform regions (e.g. bare walls), is a lot cheaper than the lasers. However most approaches only reconstruct a surface from point clouds, not a semantically rich building model. A building information model is the alliance of a geometry and a semantics for the scene elements. The main objective of this thesis is to define a framework for digital model production regarding both geometry and semantic, using point clouds as an entry. The reconstruction process is divided in four parts, gradually enriching information, from the points to the final digital mockup. First, we define a normal estimator for unstructured point clouds based on a robust Hough transform. It allows to estimate accurate normals, even near sharp edges and corners, and deals with the anisotropy inherent to laser scans. Then, primitives such as planes are extracted from the point cloud. To avoid over-segmentation issues, we develop a general and robust statistical criterion for shape merging. It only requires a distance function from points to shapes. A piecewise-planar surface is then reconstructed. Planes hypothesis for visible and hidden parts of the scene are inserted in a 3D plane arrangement. Cells of the arrangement are labelled full or empty using a new regularization on corner count and edge length. A linear formulation allow us to efficiently solve this labelling problem with a continuous relaxation. Finally, we propose an approach based on constrained attribute grammars for 3D model semantization. This method is entirely bottom-up. We prevent the possible combinatorial explosion by introducing maximal operators and an order on variable instantiation
2

Top-Down Bayesian Modeling and Inference for Indoor Scenes

Del Pero, Luca January 2013 (has links)
People can understand the content of an image without effort. We can easily identify the objects in it, and figure out where they are in the 3D world. Automating these abilities is critical for many applications, like robotics, autonomous driving and surveillance. Unfortunately, despite recent advancements, fully automated vision systems for image understanding do not exist. In this work, we present progress restricted to the domain of images of indoor scenes, such as bedrooms and kitchens. These environments typically have the "Manhattan" property that most surfaces are parallel to three principal ones. Further, the 3D geometry of a room and the objects within it can be approximated with simple geometric primitives, such as 3D blocks. Our goal is to reconstruct the 3D geometry of an indoor environment while also understanding its semantic meaning, by identifying the objects in the scene, such as beds and couches. We separately model the 3D geometry, the camera, and an image likelihood, to provide a generative statistical model for image data. Our representation captures the rich structure of an indoor scene, by explicitly modeling the contextual relationships among its elements, such as the typical size of objects and their arrangement in the room, and simple physical constraints, such as 3D objects do not intersect. This ensures that the predicted image interpretation will be globally coherent geometrically and semantically, which allows tackling the ambiguities caused by projecting a 3D scene onto an image, such as occlusions and foreshortening. We fit this model to images using MCMC sampling. Our inference method combines bottom-up evidence from the data and top-down knowledge from the 3D world, in order to explore the vast output space efficiently. Comprehensive evaluation confirms our intuition that global inference of the entire scene is more effective than estimating its individual elements independently. Further, our experiments show that our approach is competitive and often exceeds the results of state-of-the-art methods.

Page generated in 0.0524 seconds