The evolution of 3D visual content calls for innovative methods for modelling shapes based on their intended usage, function and role in a complex scenario. Even if different attempts have been done in this direction, shape modelling still mainly focuses on geometry. However, 3D models have a structure, given by the arrangement of salient parts, and shape and structure are deeply related to semantics and functionality. Changing geometry without semantic clues may invalidate such functionalities or the meaning of objects or their parts. We approach the problem by considering semantics as the formalised knowledge related to a category of objects; the geometry can vary provided that the semantics is preserved. We represent the semantics and the variable geometry of a class of shapes through the parametric template: an annotated 3D model whose geometry can be deformed provided that some semantic constraints remain satisfied. In this work, we design and develop a framework for the semantics-aware modelling of shapes, offering the user a single application environment where the whole workflow of defining the parametric template and applying semantics-aware deformations can take place. In particular, the system provides tools for the selection and annotation of geometry based on a formalised contextual knowledge; shape analysis methods to derive new knowledge implicitly encoded in the geometry, and possibly enrich the given semantics; a set of constraints that the user can apply to salient parts and a deformation operation that takes into account the semantic constraints and provides an optimal solution. The framework is modular so that new tools can be continuously added. While producing some innovative results in specific areas, the goal of this work is the development of a comprehensive framework combining state of the art techniques and new algorithms, thus enabling the user to conceptualise her/his knowledge and model geometric shapes. The original contributions regard the formalisation of the concept of annotation, with attached properties, and of the relations between significant parts of objects; a new technique for guaranteeing the persistence of annotations after significant changes in shape's resolution; the exploitation of shape descriptors for the extraction of quantitative information and the assessment of shape variability within a class; and the extension of the popular cage-based deformation techniques to include constraints on the allowed displacement of vertices. In this thesis, we report the design and development of the framework as well as results in two application scenarios, namely product design and archaeological reconstruction.
|Titolo della tesi:||A Framework for the Semantics-aware Modelling of Objects|
|Data di discussione:||29-mar-2021|
|Appare nelle tipologie:||Tesi di dottorato|