Phase-Based Vibration Frequency Measurement From Videos Recorded by Unstable Cameras

Abstract

Cameras can be used as non-contact sensors to measure target vibration conveniently. Recently, spatial phase-based methods have been developed to detect small vibrations with robustness to imaging noise. To deal with videos recorded by unstable cameras, this paper proposes a novel phase-based method, which focuses on separating camera motions from target vibrations in the spatial domain and the temporal domain. The proposed method first builds a spatial representation based on adjacent-frame phase differences, which can maintain the optical flow continuity and subsequently lead to the removal of linear camera motions in the spatial domain. Then, the temporal vibration waveforms, which may still contain distortions caused by unstable camera and noise, are extracted. To restore the target vibration, singular spectrum analysis is further incorporated for characterizing target vibrations and attenuating the distortions. The proposed method is evaluated in various laboratory and outdoor experiments. Compared with the state-of-the-art methods, the proposed method demonstrates substantial improvements on the vibration frequency measurement in the presence of camera motions.