Facade Scanner: A scalable workflow for building geometry and window-to-wall ratio capture for urban building energy modeling

Abstract

Creating building energy models of neighborhoods and cities is becoming increasingly important as cities seek to use simulation to inform their decarbonization strategies for their building stock. To facilitate urban energy modeling, we introduce a scalable method to capture geometric properties, including existing buildings' window-to-wall (WWR) ratios, following five main steps: (1) Use automated drone flight planning and aerial image capture to rapidly collect aerial imagery of neighborhoods or blocks. (2) Use photogrammetry to extract textured 3D models. (3) Isolate buildings and extract texture maps. (4) Use a custom-trained Machine Learning (ML) model to segment texture maps and identify windows. (5) Extract window and envelope areas for Building Energy Modeling. We report a mean absolute error for WWR predictions of our ML model of 8.99% compared to manually labeled images.