Image-based Vibration Signal Measurement and Calibration Using 1D CNN

Vibration monitoring based on an accelerometer has been used in the past to monitor structural health. Recently, vibration monitoring technology based on non-contact shooting cameras have been studied. However, there are certain issues, such as the additional cost for more sensor attachment and the need for a location without disturbance. This study presents a method for solving such problems by combining signal extraction based on a contact shooting camera with a convolution neural network (CNN) for error calibration. Using the contact shooting method, the existing camera can be reused without adding new ones. Additionally, this also makes it possible to overcome the accuracy limits of the contact shooting camera. The suggested methodology was verified by determining the accuracy and performance of a real sensor using the data from the quadcopter experiment.