Structural motion estimation via Hilbert transform enhanced phase-based video processing

Abstract

The non-contact sensing techniques using image/video processing have flourished in recent years benefited from the rapid development of digital cameras. However, high-fidelity motion extraction from acquired image frames is still challenging. A novel motion estimation method based on phase-domain image processing, named Hilbert phase-based motion estimation, is proposed in this study to identify motions in a more accurate and efficient manner if compared to traditional phase-based motion estimation. The theoretical relationship between phase variation and physical motion is established based on the Hilbert transform; and then in order to reduce the computation cost, the forward and inverse fast Fourier transforms are integrated with the Hilbert transform. Under the proposed framework, phase variations are therefore obtained from the corresponding analytical signal. Furthermore, peak-picking procedure and the Butterworth ideal band-pass filter are employed to decompose the original video to its mono-component signal prior to identifying displacements. The proposed method is verified using a synthetic video containing a bell-shaped surface with randomly assigned motions. The proposed Hilbert phase-based motion estimation approach avoids the influence from manual parameter selection, and the correlation coefficient of the identification results can reach 99.55%. Experimental verification using a simply supported beam is also deployed, and comparison with the state of the art demonstrates the outperformance of the proposed algorithm.