Shifting research from defect detection to defect modeling in computer vision-based structural health monitoring

Abstract

The last decade has witnessed a plethora of studies on the applications of computer vision (CV) in structural health monitoring (SHM). While effort has been primarily focused on defect detection, increasing studies are tapping into a new area called defect modeling. It remains unclear whether the shifting focus constitutes a systematic transition. This article aims to answer the question by conducting a critical review of CV-based SHM. It is found that the turning of limelight to defect modeling coincides with the proliferation of deep learning (DL) in defect detection. The shift to defect modeling does not mean a resolution of defect detection, but poses higher requirements on its performance in realistic settings (e.g., complex background, instance differentiation). A roadmap is proposed to synergize future defect detection/modeling research. The research contributes to understanding the rapidly evolving landscape of CV-based SHM, and laying out an overarching framework to guide future research.