Retrieval of 3-Dimensional Rigid and Non-Rigid Objects

This dissertation focuses on the problem of 3D object retrieval from large datasets in a near realtime manner. In order to address this task we focus on three major subproblems of the field: (i) pose normalization of rigid 3D models with applications to 3D object retrieval, (ii) non-rigid 3D object description and (iii) search over rigid 3D object datasets based on 2D image queries. Regarding the first of the three subproblems, 3D model pose normalization, two main novel pose normalization methods are presented, based on: (i) 3D Reflective Object Symmetry (ROSy) and (ii) 2D Reflective Object Symmetry computed on Panoramic Views (SymPan/SymPan+). Considering the second subproblem, a non-rigid 3D object retrieval methodology, based on the properties of conformal geometry and graph-based topological information (ConTopo++) has been developed. Furthermore, a string matching strategy for the comparison of graphs that describe 3D objects, is proposed. Regarding the third subproblem a 3D object retrieval method, based on 2D range image queries that represent partial views of real 3D objects, is presented. The complete 3D objects of the database are described by a set of panoramic views and a Bag-of-Visual-Words model is built using SIFT features extracted from them. The methodologies developed and described in this dissertation are evaluated in terms of retrieval accuracy and demonstrated using both quantitative and qualitative measures via an extensive consistent evaluation against state-of-the-art methods on standard datasets.