Xiang Luan / AECC Hunan Aviation Powerplant Research Institute
Visual-based motion extraction technique has been rapidly developed in recent year, due to its unique capability of producing full-field measurements simultaneously. Video motion magnification technique, pioneered by the MIT researchers, was proposed to magnify and visualize tiny motions of vibrating structures or subtle color variations of human skin from original videos. It provides a potential for full-field vibration measurement and further structural dynamic identification as well. In this study, a phase-based motion estimation method using undecimated double tree-complex wavelet transform is developed. It is employed to perform a multiscale image decomposition and detect features from fine to coarse level in target spatial frequency. It contains several valuable and attractive properties those are desired in the state of the art, such as perfect shift invariant and perfect reconstruction. Spatial frequencies at multiscale decomposition are estimated priori to extract vibration displacement from the phase variations in wavelet complex coefficients using the least squares objective function. A numerical simulation is conducted on artificially designed motions to demonstrate the measurement accuracy.