InsAlderKnowledge

Spatial Structure Estimation of Buildings from Outer Building Envelopes

Principal Investigators 

Prof. Andreas Hild,
Professorship of Architectural Design, Rebuilding and Conservation, TUM School of Engineering and Design

Prof. Frank Petzold,
Chair of Architectural Informatics, TUM School of Engineering and Design

Prof. Matthias Niessner,
Chair for Visual Computing & Artificial Intelligence, TUM School of Computation, Information and Technology

Abstract Future structural development must increasingly focus on the existing building stock. Of the 22 million buildings in Germany, 16 million are detached and semi-detached houses, including around 11 million single-family homes (SFHs). Increasing just 10% of SFHs from one to two residential units would statistically create 1.6 million extra units, highlighting their central role in addressing the housing crisis. Rising CO2 prices and energy costs further increase the need for efficient refurbishment. Adding residential units can help finance energy-efficient retrofits, yet planning is hampered by a lack of reliable data. 

Since most SFHs are privately owned, planning documents are often restricted and fragmented, creating a major obstacle for systematic modernization. The InsAIderKnowledge project addresses this gap by developing a BIM ondemand strategy that generates precise digital models from publicly accessible data using AI-driven methods. The approach includes: (a) creating a taxonomy of reference buildings to link external features with floor plans, (b) building a knowledge-based database to structure this taxonomy, (c) applying AI-supported facade recognition to automatically infer floor plans, and (d) generating semantic 3D models derived from facade images and floor plans. This enables data-driven decisions for expansion and renovation, fostering efficient housing solutions and sustainable urban development.