Current Research

Building Defect Detection via Adaptive RGB-Thermal Fusion

This study addresses the inefficiencies and subjective nature of manual building defect inspections, which heavily rely on human interpretation of visual and thermal data. To enable reliable and automated routine inspections, we propose a practical, multimodal framework that intelligently integrates RGB and thermal imagery without the need for manual sensor prioritization.

Instead of relying on predefined rules, our approach utilizes an advanced adaptive fusion technique. The system dynamically learns to balance the specific strengths of different sensor inputs—such as relying more on thermal data for temperature anomalies and RGB data for structural context. This allows the framework to automatically tailor its analysis to various defect types, including leakages and thermal bridges.

Ultimately, the framework automates the entire visual interpretation process. It generates precise pixel-wise defect masks and quantifies defect severity based on key objective metrics (e.g., temperature differences and affected areas). By providing intuitive visualizations and ready-to-use tabular reports, this system supports efficient, objective, and field-ready building diagnostics.

• H. Yun, K. Lee, D. Lee, Performance enhancement of building crack defect detection using super-resolution methods for drone inspection images , in: Proceedings of the 2025 International Conference on Construction Engineering and Project Management (ICCEPM), Seogwipo, South Korea, 2025.

• H. Yun, D. Kim, D. Lee, 3D coordinate normalization method for drone-based building defect detection , in: Proceedings of the 2025 Spring Annual Conference of the Architectural Institute of Korea (AIK), Seoul, South Korea, 2025.

• H. Yun, H. Lee, D. Kim, D. Lee, A performance comparison study of deep learning algorithms for determining wallpaper defects , in: Proceedings of the 2024 National University Student Academic Conference of the Korea Institute of Construction Engineering and Management (KICEM), Seogwipo, South Korea, 2024.