A New Optical Sensing Device for Real-Time Noncontact Vibration Measurement Considering Light Field Variation

Abstract

Vibration measurement is essential for vibration monitoring and control. Non-contact vibration measurement is more applicable in practice than the contact measuring manner, but the non-contact methods usually pose a high requirement on the light field environment. To solve this issue, a new binocular vision method is proposed to enable non-contact vibration measurement with different light fields. In this new method, the multi-target objects in the same image are firstly recognized by a YOLOv5 model to generate the bounding boxes; meanwhile, a depth image is generated through the binocular vision and kept synchronized with the target image. Then, based on each bounding box and its corresponding depth image, an optimal depth value decision algorithm is developed to determine the three-dimensional (3D) real-time coordinates of each object. As a result, the vibration of multi-target objects can be measured simultaneously. An experimental test system was built to evaluate the performance of the proposed method in indoor and outdoor light fields. The tests demonstrate accurate vibration measuring results and the proposed non-contact method is able to detect very low frequency vibrations.