Scalable building reconstruction and window detection for urban building energy modelling applications
Amber Jiayu Su, Ann Ren, Kewei Curtis Xu, Timur Kent Dogan · Journal of Building Performance Simulation · 2025
Creating analytical building simulation models of neighbourhoods and cities is becoming increasingly important as cities and utilities seek to use simulation to inform electrification, decarbonization and quality-of-life improvement strategies of the building stock. However, essential parameters needed for such models like detailed building, contextual geometry and key parameters such as windows or window-to-wall-ratios (WWR) are often not widely available at the municipal scale. To facilitate urban analysis such as urban building energy modelling (UBEM), we introduce a scalable and automated method for façade reconstruction and window detection using textured mesh models that can be produced from imagery, and purchased from commercial data providers for most metropolitan areas. This procedure framework leverages fine-tuned image segmentation models and post-optimization through pattern recognition and large vision model. This workflow is designed to be robust against noise and blurriness typical of large-scale photorealistic data, providing a reliable and scalable solution for building geometry reconstruction.