Automated fire risk assessment and mitigation in building blueprints using computer vision and deep generative models

Building fire risks pose significant threats to individual safety and bear substantial economic consequences. Consequently, developing effective automated fire risk assessment and mitigation solutions has become increasingly crucial to mitigate fire risks and reduce losses. Previous automated fire risk assessment approaches have predominantly relied on structured building design information, such as Industry Foundation Classes (IFC)-based Building Information Modelling (BIM) models, which limited their applicability in scenarios without such data. Additionally, there is a notable absence of a comprehensive approach in existing research for effectively mitigating fire risks identified during the assessment process. This study aims to bridge these gaps by proposing an innovative approach for assessing and mitigating fire risks using raw building blueprints. This approach incorporates advanced computer vision techniques to process both paper-based and digital blueprints. It then employs a knowledge-based algorithm for evaluating fire safety and regulatory compliance within these blueprints. A key innovation is the development of a deep generative model that redesigns unqualified blueprint designs to meet safety standards. This research contributes to the field by providing a more capable, accessible, and flexible approach for automated building safety, introducing Artificial Intelligence (AI)-enabled solutions for risk mitigation. It offers a versatile option applicable to various building types, significantly enhancing fire safety and compliance, especially for buildings without extensive BIM data. This study addresses the limitations of current methodologies and lays the groundwork for future advancements in automated fire risk assessment and mitigation. © 2024 Elsevier Ltd