Capturing Façade Diversity in Urban Settings Using an Automated Window to Wall Ratio Extraction and Detection Workflow

Abstract

Measuring window to wall ratios (WWRs) is key to assessing building performance as façade apertures control the admission of light, wind and heat. However, this data is not always publicly available. This paper details a methodology for automatically extracting and rectifying street-view facade imagery while utilizing a Machine Learning model to detect WWRs with architectural generalization in mind. Although several models of detection have emerged to categorize façade features, some lack robustness when presented with greater design diversity. Hence, the training and validation process of the Convolutional Neural Network (CNN) model utilized is centered around three main data categories; environmental conditions, design diversity and context. The results show that the proposed workflow sufficiently represents the WWRs of buildings in an area in Lisbon under varied design conditions. We find that the distribution of prediction accuracy, tested on 864 facades, shows that 72% of buildings are detected within the 10% error range.