Aziz, AngelinaAngelinaAzizGard, NiklasNiklasGardKönig, MarkusMarkusKönigEisert, PeterPeterEisertHeinbach, Jan HendrikJan HendrikHeinbachTrost, LukasLukasTrost2025-08-112025-08-112025https://publica.fraunhofer.de/handle/publica/49038010.1061/JCCEE5.CPENG-64922-s2.0-105009459600The integration of building information modeling (BIM) and computer vision in fire safety inspections offers a promising avenue to enhance efficiency and accuracy. This study presents a proof of concept for a machine learning (ML)-based framework to automate fire safety inspection tasks, including detecting fire safety equipment (FSE), analyzing inspection tags, and localizing components within as-is BIM models. Across all services, the framework achieves a mean average precision (mAP@0.5) score of 0.944-0.995, validating its robustness in detecting and analyzing FSE components under diverse conditions. The methodology leverages synthetic data tailored to fire safety challenges, such as occluded equipment and rare inspection tag combinations, enhancing ML performance and robustness. The application of an innovative multimodal image-to-model alignment directly registers images to BIM without requiring point clouds or structure-from-motion techniques, simplifying the integration process. These results demonstrate the framework's ability to streamline inspection workflows, improve data accuracy, and address practical challenges in fire safety management. While focused on fire safety, the framework is scalable to other facility management tasks, providing a foundation for automated, data-driven processes in the built environment.enfalseBuilding information modeling (BIM)Computer vision (CV)Fire safety inspectionInspection automationInstance segmentationObject detectionjournal article