• A ‘sandwich’ model of automated window detection workflow for digital twin buildings. • 105 multidisciplinary articles focused on design, construction, and operation phases. • Deep learning overtaking rule-based methods in diverse data processing since 2018. • Windows’ digital twins enable energy simulation, robotics, and semantic models. • A three-step guideline is proposed for automated window detection in digital twin. Windows play a significant role in the livability and sustainability of buildings, such as energy efficiency, natural lighting, and ventilation; therefore, they are increasingly important subjects of surveys and analysis for both new and existing buildings. Many automated window detection (AWD) workflows have been studied to extract windows’ properties; however, a generalized model of the underlying technologies and practical guidelines remains lacking for both newcomers and Industry practitioners. Therefore, this paper aims to address the following questions: (1) the scope and general workflow of AWD; (2) the trends regarding AWD key components; and (3) the pathways of key component (e.g., data source, attribute, storage, and method) selections to fit the application scenario. Four decision trees, trained on 105 technical papers collected in accordance with the PRISMA standard, achieved an average F1-score of 80.3% and accuracy of 88.1%. This paper contributes the first comprehensive review of AWD, with a general ‘sandwich’ model, a comprehensive analysis of advanced technologies, trends, and future directions, as well as three-step guidelines for practitioners.
Meng et al. (2026) studied this question.