Structural Operating Deflection Shape Estimation via a Hybrid Computer-Vision Algorithm

Non-contact optical measurements are being more commonly used in numerous industrial and research domains to obtain reliable displacement measurements, due to their unique advantageous over other instrumentations. Phase-based motion estimation and motion magnification are target-less methods to process the sequence of images captured from vibrating structures. Within this study, a new hybrid computer vision algorithm is proposed to process the motion-magnified sequence of images in an automatic fashion. Particle filter enhanced point-tracking method is employed to track the feature points in the motion-magnified videos and k-means clustering algorithm as an autonomous statistical learning method is utilized for the segmentation of the particles. The performance of the proposed algorithm is investigated on the experimental data acquired from a cantilever beam subject to vibration. By applying the proposed algorithm, human interference and supervision can be decreased dramatically compared to the state of the methods such as canny edge detectors. Therefore, the sensitivity of the structural dynamics identification is improved and the ODS vector estimation procedure can be completed in shorter time.