A visual privacy-aware plan layout generation framework
Journal of Asian Architecture and Building Engineering, 2026 (SCI-Expanded, AHCI, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1080/13467581.2026.2686914
- Dergi Adı: Journal of Asian Architecture and Building Engineering
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Arts and Humanities Citation Index (AHCI), Scopus, Art Source, Compendex, Index Islamicus, Directory of Open Access Journals, Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: analytical hierarchy process, generative adversarial networks, Machine learning, plan layout generation, visual privacy
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
This study presents a visual privacy-aware framework for architectural plan layout generation by integrating Pix2Pix, the Analytic Hierarchy Process (AHP), and isovist-based spatial analysis. Public restrooms in municipal buildings–where visual shielding at the point of entry is a critical design concern–were selected as the study scope. A dataset of 174 public restroom layouts from awarded architectural competitions was reconstructed in a CAD environment. Gender-specific AHP surveys were used to derive component importance weights, which were integrated into an isovist-based model to compute privacy scores and classify male and female layouts into high- and low-privacy groups using data-driven thresholds. Separate Pix2Pix models were trained on privacy-refined datasets and evaluated on unseen samples. An expert review revealed clear differences in perceived privacy between high- and low-privacy outputs. On a 1–9 Likert scale, male layouts achieved median scores of 6.2 and 4.8, while female layouts showed a stronger separation with median scores of 7.1 and 4.3, respectively. These findings demonstrate that embedding spatial privacy metrics into dataset construction influences generated layout characteristics. The proposed workflow offers a replicable approach for incorporating perception-based spatial qualities into machine learning–driven plan layout generation, contributing to privacy-aware generative design processes.