Advanced Semantic Deep Search for 3D Scenes

In this paper, we present iRep3D 2.0, a semantic-based repository for efficient deep search of 3D scenes in X3D, XML3D and COLLADA. It supports real time hybrid semantic retrieval of3D scenes in terms of their syntactic, conceptual, functional and geometric features. This is enabled via off-line indexing over annotated 3D (sub-) scenes with respect to different aspects of semantics and parallel sub-query processing. In addition, an efficient incremental (re-)indexing strategy is derived for iRep3D 2.0, which offers the quick adaption of system to the updates of 3D scene annotations. Further, iRep3D 2.0 in particular scales to advanced semantic deep search and reuse of relevant sub-scenes down to the individual object level of a scene. For this, sub-scene extraction method for XML-based 3D scenes is performed in case that a new scene is being indexed. Finally, the repository comes with a web-based graphic user interface together with a 3D scene annotation toolkit for both non-experts and expert users.