Façade Feature Extraction for Sustainable Urban Planning: Merits and Limitations of Existing Algorithms

44 Pages Posted: 6 Jul 2023

See all articles by Nada of Tarkhan

Nada of Tarkhan

affiliation not provided to SSRN

Christoph Reinhart

Massachusetts Institute of Technology (MIT) - School of Architecture and Planning

Jakub Szcześniak

affiliation not provided to SSRN

Abstract

Urban feature extraction methods have presented multiple opportunities in the field of urban studies. With reference to façades, window coverage largely influences indoor environmental performance. Several methods have been proposed to detect windows from street-view images, yet it is unclear what the respective strengths and limitations are. This paper aims to add clarity on façade extraction-focused computer vision techniques. An automated pipeline to enable the extraction of WWRs (Window-to-Wall ratios) is introduced using two approaches; a grammar-based detection framework (Method 1) and a learning-based method (Method 2) utilizing CNNs (Convolutional Neural Networks). The detection efficacy of both methods is compared, in New York and Lisbon, including their ability to extract additional properties such as floor-to-floor heights. The study finds that Method 2 shows lower error scores. In Lisbon, 69% of conditions under Method 1 and 82.5% of conditions under Method 2 were detected within the 10% error range. In New York, 66% of conditions under Method 1 and 77.5% of conditions under Method 2 were within the 10% error range. Finally, a hybrid method is proposed and higher accuracies are obtained in both cities. With reference to building-height estimation formulated under Method 2, the results show a relative-error of 3.5%.

Keywords: façade feature extraction, urban modeling, city-wide climate assessments, urban data, building Simulation, Computer vision

Suggested Citation

Tarkhan, Nada of and Reinhart, Christoph and Szcześniak, Jakub, Façade Feature Extraction for Sustainable Urban Planning: Merits and Limitations of Existing Algorithms. Available at SSRN: https://ssrn.com/abstract=4502704 or http://dx.doi.org/10.2139/ssrn.4502704

Nada of Tarkhan (Contact Author)

affiliation not provided to SSRN ( email )

No Address Available

Christoph Reinhart

Massachusetts Institute of Technology (MIT) - School of Architecture and Planning ( email )

United States

Jakub Szcześniak

affiliation not provided to SSRN ( email )

No Address Available

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
84
Abstract Views
195
Rank
557,750
PlumX Metrics