Pixel-Level Vibration Source Detection Using Undersampled Video Sequences

This study proposes a concept of vision-based vibration source localization that can extract objects vibrating at hundreds of Hz with pixel-level accuracy using images captured at an undersampled frame rate that is much lower than the vibration frequency. We discuss how to determine the minimum frame rate of undersampled images so that different vibration frequencies, including DC background components, are separable when the vibration frequencies are larger than the Nyquist frame rate of the images. By executing short-time Fourier transforms for brightness signals at all pixels, we demonstrate the performance of vibration source localization using undersampled images when observing flying multicopters with propellers rotating at hundreds of Hz.