Current Research

Construction supervision is essential for ensuring compliance with design documents and specification standards, but current practice relies heavily on manual inspection and subjective judgment. This study proposes a hierarchical context-aware construction supervision support framework integrating a Vision-Language Model (VLM) and SAM3 for reinforced concrete work inspection. The framework consists of macro and micro inspection stages that reflect the practical reasoning process of human supervisors. In the macro stage, the VLM interprets the overall construction context from wide-angle site images and identifies areas requiring further inspection. In the micro stage, SAM3 segments construction elements in close-up images, and the VLM identifies members and evaluates checklist items by jointly analyzing the original and segmented images. The proposed framework supports a more systematic and explainable supervision process and demonstrates the potential of VLM-based approaches for construction supervision.