Projektdetails
Beschreibung
Recently, the optimization of existing buildings, especially in Germany (up to 80 % of all projects), has increased sharply, with the focus on the energy-efficent refurbishment of post-war residential buildings. In the coming years, residential buildings built up to 1918 (15 %) and from 1920 to 1940 (13 %) offer great potential for saving energy.
Challenges arise due to unavailable or outdated plans and data; therefore complex building surveys are needed. Unlike new buildings, existing ones evolve throughout their life cycle, making investigation a "reverse design and planning process." This involves measuring "visible" surfaces/points and using additional documentations and partial finds, and crucial domain knowledge is forgotten today.
Replicating existing buildings precisely is impossible, resulting in imperfect models. NERF2BIM addresses these challenges by leveraging AI advancements to introduce a detailing-on-demand strategy, enhancing the existing BIM methodology with variable information integration, as follows:
I. Enhance point cloud-based surveys by addressing acquisition, georeferencing, and uncertainty quantification, focusing on sustainable solutions. Improve scanning performance and reliability through strategies for evaluating and documenting uncertainty.
II. Develop Semantic 3D Understanding for automatic labelling of indoor and outdoor acquisitions, building on the NeRF model to integrate 2D and 3D data. This enables accurate 3D-consistent semantic understanding of BIM parts in raw 3D scans, alongside simple BIM Component Reconstruction using geometric representations.
III. Apply Component Detailing to "simple" BIM Components, leveraging a knowledge base with architectural insights. Employ novel AI methods to introduce a user feedback system, refining the system continuously.
Challenges arise due to unavailable or outdated plans and data; therefore complex building surveys are needed. Unlike new buildings, existing ones evolve throughout their life cycle, making investigation a "reverse design and planning process." This involves measuring "visible" surfaces/points and using additional documentations and partial finds, and crucial domain knowledge is forgotten today.
Replicating existing buildings precisely is impossible, resulting in imperfect models. NERF2BIM addresses these challenges by leveraging AI advancements to introduce a detailing-on-demand strategy, enhancing the existing BIM methodology with variable information integration, as follows:
I. Enhance point cloud-based surveys by addressing acquisition, georeferencing, and uncertainty quantification, focusing on sustainable solutions. Improve scanning performance and reliability through strategies for evaluating and documenting uncertainty.
II. Develop Semantic 3D Understanding for automatic labelling of indoor and outdoor acquisitions, building on the NeRF model to integrate 2D and 3D data. This enables accurate 3D-consistent semantic understanding of BIM parts in raw 3D scans, alongside simple BIM Component Reconstruction using geometric representations.
III. Apply Component Detailing to "simple" BIM Components, leveraging a knowledge base with architectural insights. Employ novel AI methods to introduce a user feedback system, refining the system continuously.
| Kurztitel | NERF2BIM |
|---|---|
| Status | Laufend |
| Tatsächlicher Beginn/ -es Ende | 1/11/24 → 31/10/28 |
Projektbeteiligte
Fingerprint
Erkunden Sie die Forschungsthemen, die von diesem Projekt angesprochen werden. Diese Bezeichnungen werden den ihnen zugrunde liegenden Bewilligungen/Fördermitteln entsprechend generiert. Zusammen bilden sie einen einzigartigen Fingerprint.