From image to fire safety: An image-driven framework for as-is BIM reconstruction and fire risk assessment of existing buildings via semantic guidance

Existing buildings pose significant challenges in urban fire-risk assessment. BIM provides critical spatial and semantic support for fire simulation, yet most BIM models are design-based and fail to represent actual as-is conditions. This study proposes an image-driven framework for as-is BIM reconstruction and fire-risk assessment. The approach uses semantic-guided Gaussian Splatting (S-GS) generates semantic-rich point clouds and rendered images. Point clouds restore building geometry to create IFC-compliant BIM models. The rendered images are used for semantic object recognition and are linked to the geometric space. The resulting BIM models are then used for fire simulation and quantitative risk assessment. Four case studies verify the feasibility and robustness of the method. By using images as the primary data, this study enhances BIM adaptability and semantic richness for fire analysis, introducing a new paradigm of image–point-cloud collaboration for accurate spatial reconstruction and semantic enrichment of existing buildings.