Computer vision for infrastructure inspection and hazard detection

Deep learning-based computer vision for automated defect detection and hazard assessment in civil infrastructure.

hazard-detection   computer-vision   object-detection   instance-segmentation

Computer vision is transforming infrastructure inspection by enabling automated detection and localization of structural defects from images and sensing data. We develop deep learning-based computer vision models for infrastructure health monitoring and hazard detection, with an emphasis on robust object detection and instance segmentation for civil engineering applications. By automating labor-intensive and error-prone inspection tasks, these methods provide faster, more consistent, and scalable assessments of infrastructure conditions. Our current research includes automated concrete crack detection from visual images and tunnel lining segmentation from ground-penetrating radar (GPR) images.

Automated concrete crack detection using deep learning-based instance segmentation models (Mask R-CNN (two-stage model) and YOLO (single-stage model)) (Choi et al., 2024).
Automated tunnel lining segmentation from ground-penetrating radar (GPR) images using deep learning-based object detection and segmentation (Mask R-CNN (two-stage model) and YOLO (single-stage model)) (Bae et al., 2025).

References

2025

  1. Tunnel lining segmentation from ground-penetrating radar images using advanced single-and two-stage object detection and segmentation models
    Byongkyu Bae, Yongjin Choi, Hyunjun Jung, and 1 more author
    Computer-Aided Civil and Infrastructure Engineering, 2025
    DOI

2024

  1. Application of Mask R-CNN and YOLOv8 algorithms for concrete crack detection
    Yongjin Choi, Byongkyu Bae, Taek Hee Han, and 1 more author
    IEEE Access, 2024
    DOI