AI-Based Marker-Free DIC for Measuring Displacements of Large Structures

Digital image correlation (DIC) technique provides an accurate and efficient solution for measuring both 2-D and 3-D displacements of large structures. However, a successful DIC implementation requires unique patterns or markers on the target surface. Creating artificial markers on large structures is a time-consuming and challenging task. Any error while developing or identifying unique markers could lead to unreliable and inaccurate DIC results. This study introduces a novel artificial intelligence (AI)-based approach to identify and generate distinctive feature-rich natural markers for DIC. The proposed technique includes a crucial preprocessing step, which comprises an instance segmentation model built with the you only look once (YOLOv8) and the segment anything model (SAM) deep learning algorithms. This model ensures that the markers are integral to the structure rather than a part of the background. Further, the developed methodology employs the KAZE feature-based clustering (FBC) approach to identify poly-shaped non-intersecting regions as a DIC marker for achieving strong correlation. This study incorporates a wind turbine tower dataset to validate and demonstrate the proposed methodology. The performance of the developed technique is evaluated with respect to the conventional manual marker selection approach and recently developed marker generation methodologies. It is observed that the proposed methodology is 11 times faster and reduces memory consumption by 63%. Moreover, it excludes feature-less regions and can successfully determine the optimal feature-rich DIC marker (in the form of a non-intersecting poly-shaped marker) for achieving strong correlations.