Scan-to-Fire-Safety (S2FS): An Automated Indoor Fire Risk Assessment Framework based on Knowledge-Informed Semantic Segmentations

Assessing the potential indoor fire risks in aging buildings is critical to their fire safety. However, existing manual assessment methods suffer from subjectivity and inefficiency. Besides, information-technology-based approaches (e.g., computer vision) fail to capture the actual 3D spatial relationships among fire-risk-related indoor components (FRRICs), resulting in omitting distance-based compliance checks (e.g., fire equipment coverage and electrical clearance requirements). Therefore, this study develops a Scan-to-Fire-Safety (S2FS) framework for automated indoor fire risk assessment using 3D point clouds. Three contributions are: (1) designing a Fire Safety-Aware Knowledge Graph (FSKG) to structure explainable rules for determining indoor fire risk; (2) developing a Fire-Safety-Oriented Semantic Segmentation (FSOSS) model to identify potential FRRICs; (3) developing a Knowledge-informed Spatial Risk Mapping (KSRM) algorithm for quantifiable compliance checking. Validation results demonstrate an overall segmentation mIoU of 78.6%, an accuracy of 91.2% for high-priority fire safety categories, and successful automated detection of regulatory compliance violations. • A fire safety-aware knowledge graph for reasoning rules for determining fire risk • A fire safety-oriented semantic segmentation model for identifying risky components • A mapping algorithm for quantifiable and explainable risk compliance checking • The proposed methods were validated in two residential rooms • Results show 78.6% mIoU and 91.2% accuracy for fire safety categories