A real-time feature-based clustering approach for vibration-based SHM of large structures

Feature matching techniques are widely explored in structural health monitoring (SHM) of large structures due to their invariance to scale and rotational changes, unlike the traditionally used template matching techniques. However, uniformity of feature distribution is a major concern while matching the images, which requires a large number of features for better correspondence, increasing the computational time. Thus, this study proposes an intermediate solution — a KAZE feature-based clustering approach. It creates a cluster of only four interest points, which decreases the processing time and enhances the accuracy of detecting pixel displacements. On the other hand, due to the high uncertainty of the corresponding points in all the subsequent images, the study introduces a concept of centroid calculation and centroid prediction for identifying the coordinates responsible for measuring the displacements of large structures. Further, the present work also demonstrates the developed methodology in a real-time SHM experiment by measuring the dynamic response of a structure in real-time.