Semantic enrichment of octree structured point clouds for multi-story 3D pathfinding
Jan 15, 2018
A new paper has been published. It symbolizes the achievement of a successful MSc thesis leaded by the brilliant (ex) student of the MSc Geomatics of TU Delft, Florian Fichtner, in collaboration with CGI.
The outcome of this paper also stands as an important contribution to the NWO/M4S SIMs3D project that is aiming to provide support for advanced navigation in context of emergency in large buildings. Full details are provided below.
PDF DOI BibTeX
@article{Fichtner18Semantic,
author = {Florian W. Fichtner, Abdoulaye A. Diakité, Sisi Zlatanova and Robert Voûte},
title = {Semantic enrichment of octree structured point clouds for multi-story 3D pathfinding},
journal = {Transactions in GIS},
issn = {1467-9671},
url = {http://dx.doi.org/10.1111/tgis.12308},
doi = {10.1111/tgis.12308},
pages = {n/a--n/a},
}
As we realize that we spend most of our time in increasingly complex indoor environments, applications to assist indoor activities (e.g. guidance) have gained a lot of attention in the recent years. The advances in ubiquitous computing made possible the development of several spatial models intending to support context-aware and fine-grained indoor navigation systems. However, the available models often rely on simplified representations (e.g. 2D plans) and ignore the indoor features (e.g. furniture), thereby missing to reflect the complexity of the indoor environment. In this paper, we introduce the Flexible Space Subdivision framework (FSS) that allows to automatically identify the spaces that can be used for indoor navigation purpose. We propose a classification of indoor objects based on their ability to autonomously change location and we define a spatial subdivision of the indoor environment based on the classified objects and their functions. The framework can consider any 3D indoor configuration, the static and dynamic activities it hosts and it enables the possibility to consider all types of locomotion (e.g. walking, flying, etc.). It relies on input 3D models with geometric, semantic and topological information and identifies a set of subspaces with dedicated properties. We assess the framework against criteria defined in previous researches and we provide an example.
