Wessel, Raoul Henrik Joseph Frédéric: Shape Retrieval Methods for Architectural 3D Models. - Bonn, 2014. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5n-34928
@phdthesis{handle:20.500.11811/6023,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5n-34928,
author = {{Raoul Henrik Joseph Frédéric Wessel}},
title = {Shape Retrieval Methods for Architectural 3D Models},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2014,
month = feb,

note = {This thesis introduces new methods for content-based retrieval of architecture-related 3D models. We thereby consider two different overall types of architectural 3D models. The first type consists of context objects that are used for detailed design and decoration of 3D building model drafts. This includes e.g. furnishing for interior design or barriers and fences for forming the exterior environment. The second type consists of actual building models. To enable efficient content-based retrieval for both model types that is tailored to the user requirements of the architectural domain, type-specific algorithms must be developed. On the one hand, context objects like furnishing that provide similar functions (e.g. seating furniture) often share a similar shape. Nevertheless they might be considered to belong to different object classes from an architectural point of view (e.g. armchair, elbow chair, swivel chair). The differentiation is due to small geometric details and is sometimes only obvious to an expert from the domain. Building models on the other hand are often distinguished according to the underlying floor- and room plans. Topological floor plan properties for example serve as a starting point for telling apart residential and commercial buildings.
The first contribution of this thesis is a new meta descriptor for 3D retrieval that combines different types of local shape descriptors using a supervised learning approach. The approach enables the differentiation of object classes according to small geometric details and at the same time integrates expert knowledge from the field of architecture. We evaluate our approach using a database containing arbitrary 3D models as well as on one that only consists of models from the architectural domain. We then further extend our approach by adding a sophisticated shape descriptor localization strategy. Additionally, we exploit knowledge about the spatial relationship of object components to further enhance the retrieval performance. In the second part of the thesis we introduce attributed room connectivity graphs (RCGs) as a means to characterize a 3D building model according to the structure of its underlying floor plans. We first describe how RCGs are inferred from a given building model and discuss how substructures of this graph can be queried efficiently. We then introduce a new descriptor denoted as Bag-of-Attributed-Subgraphs that transforms attributed graphs into a vector-based representation using subgraph embeddings. We finally evaluate the retrieval performance of this new method on a database consisting of building models with different floor plan types.
All methods presented in this thesis are aimed at an as automated as possible workflow for indexing and retrieval such that only minimum human interaction is required. Accordingly, only polygon soups are required as inputs which do not need to be manually repaired or structured. Human effort is only needed for offline groundtruth generation to enable supervised learning and for providing information about the orientation of building models and the unit of measurement used for modeling.},

url = {https://hdl.handle.net/20.500.11811/6023}
}

Die folgenden Nutzungsbestimmungen sind mit dieser Ressource verbunden:

InCopyright