Research On Non-contact Low Frequency Vibration Perception Method Based On Machine Vision

Compared with the traditional contact vibration measurement technology,the noncontact vibration measurement method based on machine vision has the advantages of full-field measurement and multiple scenarios.This thesis uses machine vision as the main method and non-contact low-frequency vibration perception as the main purpose.Two effective vibration monitoring methods are proposed for two types of measurement targets with different texture characteristics.Aiming at the measurement target with obvious texture characteristics,this thesis proposes a pixel-level vibration feature extraction method to obtain its one-dimensional vibration characteristic signal and studies a vibration monitoring network based on multi-scale adaptive soft thresholding to realize the vibration characteristics perception.Pixel-level vibration feature extraction method uses Weighted Least Squares(WLS)filtering algorithm to smooth the image and maintain the edges.This thesis uses the Canny edge detection algorithm to locate vibration feature points,and extract the gray changes on the vibration feature points to form a one-dimensional vibration feature signal.The vibration monitoring network is based on a lightweight network,using the self-attention mechanism and soft thresholding to filter out redundant information in vibration characteristics,and combining multi-scale convolution to aggregate the characteristic information of different receptive fields.This thesis uses the vibration feature extraction method and the vibration monitoring network to build a lowfrequency vibration perception system,successfully identifying 11 types of pixel-level vibration features with different frequencies for different measurement targets,and it has achieved recognition accuracy of more than 95% on the two types of test targets,realize the vibration monitoring aiming at the target with obvious characteristics.Aiming at the measurement target whose texture feature is not obvious,or even the surface has no feature,this thesis uses a circular arc grating projection vibration measurement system to extract vibration characteristics and achieve a high dynamic amplitude measurement range.The amplitude measurement range of rectangular grating projection is limited,because its periodicity causes phase ambiguity,and the high dynamic amplitude of vibration will cause systematic measurement errors.To alleviate the above-mentioned problems,a vibration measurement method using circular arc grating projection instead of rectangular grating projection is proposed.The vibration measurement method refers to Fourier Transform Profilometry(FTP),which uses a half-circular filter to demodulate the wrapped phase of the arc grating,uses the branch cut method to expand the wrapped phase,and uses the phase-height mapping to achieve the measurement target calculation of relative height.This thesis compares the amplitude measurement range of the circular arc grating projection system and the rectangular grating projection system,and successfully verifies the larger measurement range of the former;and uses the circular arc grating projection system to measure complex signals,which proves that the circular arc grating projection system can extract and reconstruct complex vibration characteristic information.The two low-frequency vibration perception methods based on machine vision proposed in this thesis are respectively used for the vibration monitoring of two targets with different texture characteristics,which can accurately extract and identify the vibration characteristics of the measurement target.