|Timeline:||July 2015 - 30 June 2019|
In recent years there has been an impressive progress in indoor data collection technologies namely mobile laser scanners (MLS), Microsoft Kinect and Google Tango. Indoor 3D models have various applications in disaster management, real states, facility management and indoor routing. The main user of our product in this research is Nederland Fire Brigade for disaster management in large buildings such as airports and hotels.
The process of indoor 3D reconstruction is a tedious work and for most of the buildings, 2D plans are not updated. This problem and indoor 3D model applications motivate us of using other sources such as point clouds for indoor 3D reconstruction.
The state-of-the-art for indoor 3D reconstruction is in early stage and most of the research has been carried out on a simple indoor structure without clutter (e.g. furniture). In this research, Shayan Nikoohemat proposes a grammar-based approach for indoor 3D reconstruction from point clouds. Grammars can be used to describe repetitive structures and regularity (i.e. high-level knowledge) which mostly are found in man-made structures. Grammar approaches are widely applied for city modelling, while there are few works for indoor 3D reconstruction using grammar. This research focuses on indoor 3D reconstruction and has four main objectives: i. geometry reconstruction, ii. designing the grammar, iii. semantic labelling, iv. consistency and accuracy control, whereas in all objectives source of the data does not only point clouds but also images.
Our approach initializes with high-level point cloud processing for geometry reconstruction. Through a data-driven process, grammar rules will be learned from the data and during an iterative active learning with user interaction grammar components improve. By adding semantics and improving the level of details the indoor 3D model evolves from a coarse model to a fine model. Finally, consistency and accuracy control will be performed to confirm the correctness of the model against IndoorGML and IFC standards. The resulting models are meant to be used in disaster management in large buildings for emergency responses. Case studies are public buildings such as museums, hospitals, concert halls and airports in which The Netherlands BHV (Bedrijfhulpverleners) and fire brigade are dealing with the safety management. The researchers in TU Delft are responsible for designing evacuation plans based on our indoor 3D model. The generated smart 3D model can ensure the intelligent evacuation of buildings in case of an emergency.