This article presents a computer vision damage assessment approach that relates surface crack patterns to damage levels and stress state characteristics in conventionally reinforced concrete and steel fiber–reinforced concrete panels. Previous studies have focused on crack patterns for specific structural element types such as beams and columns, but this study considers stress states in a more general framework. In particular, image data from previously published panel test specimens subjected to nominally constant stress have been collected to develop image-based estimation models capable of quantifying damage levels and stress components for full-panel crack patterns, and to investigate subimage sampling strategies to approximate full-panel results using partial-panel images. The objective here is to show that the analog of representative volume elements can be extended to image-based analysis contexts. The image datasets used in this article have been obtained from five different published studies, which provided 189 crack pattern images captured from [Formula: see text] concrete and steel fiber–reinforced concrete shear panel specimens. Given the limited size of the dataset, a feature-based computer vision approach has been used, with various geometric attributes of surface crack patterns used to train the estimation models. Within the limits of the data available, the preliminary results presented here indicate that quantifiable correlations exist such that stress state and damage level estimation models are valid across a range of loadings (i.e. reverse cyclic and monotonic) and materials (reinforced concrete and steel fiber–reinforced concrete), and that with appropriate sampling techniques, it is possible for subsampled images to yield estimations similar to full-panel results. These localized correlations between crack patterns and stress states potentially could be used in broader contexts for damage assessment of more general reinforced concrete and steel fiber–reinforced concrete members.
