Structural damage recognition and residual capacity prediction using computer vision for reinforced concrete members

Abstract

This study presents the results of relating images of superficial crack patterns of shear critical reinforced concrete walls to its damage levels using computer vision technology. The focus of this study is to determine the extent to which computer vision techniques can successfully relate surface observations to quantitative and qualitative damage level estimation in shear critical reinforced concrete walls. This is part of a broader goal to improve current site inspection routines by considering the feasibility of employing computer vision technology in such circumstances. An extensive database of 778 crack images out of 147 specimens were collected from previous experimental studies including information such as load levels, displacement levels, mechanical, geometric properties and experiment details were collected. These images were then segmented to extract the cracks into binary form. Various textural and geometric attributes of surface crack patterns were extracted from the segmented images and used as predictors for both classification and regression prediction models. The accuracy of the classification models were evaluated using True Positive Rate and Area Under Curve values, whereas statistical measures of errors were used for the regression models. This study also discusses the limitations of the work done and recommendations for future studies relating to this field of research.