Amplitude measurement of small displacement using video magnification

Abstract

Video magnification algorithms show promising results when used to amplify small vibrations. Measuring these small vibrations is integral in modal analysis of an object and is usually done using specialized vibrometers or accelerometers. Computer vision (CV) systems fall short in this task as the magnitude of the vibrations decreases because of a small SNR (Signal-to-Noise Ratio). In this paper we try to further improve the accuracy of the CV approach by adding video magnification into the image pre-processing stage, allowing the algorithms to measure even imperceptibly small vibrations. For this purpose, we have gather ground truth data with a laser vibrometer in tandem with high speed camera footage of a metal cantilever beam, vibrating in its first mode, and have trained a convolutional neural network for regression.