Research On Technologies And Applications Of Vision-based Subtle Vibration Magnification

Video magnification technology can magnify subtle but important changes by processing image spatio-temporal information.Video magnification technology can be used to reveal small changes that are difficult to perceive with the naked eye,and is of great value in some fields like mode recognition,heart rate detection,and failure and damage detection.This thesis is based on detailed analysis and research on the video magnification technology.Vibration magnification algorithms are optimized from multiple angles,and the algorithms have achieved great results on non-contact heart rate detection field.It has good engineering application value.Large motion interference can cause the magnified results to appear blurry with dark shadows.To resolve this problem,this thesis proposes an optimized algorithm with acceleration filter based on amplitude selection.Through the acceleration filter,only the second-order acceleration signals can be magnified in order to suppress the influence of linear motion disturbances.On the basis of acceleration filtering,an amplitude selection filter is added according to the different characteristics of the vibration amplitude to suppress large nonlinear disturbances.The proposed algorithm can magnify the subtle vibrations without producing a large number of artifacts and blurs.Traditional image pyramid multi-scale decomposition can only be decomposed in a limited direction and the extraction of feature information is complicated and time-consuming,in view of these problems,this thesis proposes a subtle vibration magnification algorithm based on monogenic wavelet analysis.By combining wavelet multi-scale analysis with monogenic analysis,using the translation,rotation and direction invariance of the monogenic signal to construct an isotropic monogenic wavelet,the optimized algorithm has a more concise filter structure and faster phase decomposition.This algorithm greatly improves video frame decomposition and reconstruction speed.And it can effectively suppress large-scale motion interference and noise in non-linear scenes when adding the amplitude selection filter and the acceleration filter.Traditional subtle vibration magnification methods can only magnify vibrations under the stable condition of constant vibration frequency.For the non-stationary object with unfixed frequencies,the band-pass filtering method based on the time domain is no longer applicable.To resolve this,this thesis proposes an optimized algorithm for subtle vibration amplification based on time-frequency filtering.The time-frequency filtering kernel can be obtained by establishing a global vibration model.Then apply the filtering kernel to each local vibration signal to extract the vibration mode and instantaneous frequency in the time-varying vibration video accurately.Time-frequency filtering can be used to realize the magnification of small vibrations under non-stationary conditions.Based on the optimized vibration magnification algorithm,this thesis proposes a highaccuracy non-contact heart rate detection technology.After magnifying the radial artery pulse by subtle vibration magnification technology,the pulse wave is extracted to calculate the heart rates and the algorithm is compared with the actual heart rate value to verify its accuracy.This thesis studies the influence of different algorithms and parameters,lighting conditions on the detection accuracy and verify the reliability and universality of heart rate detection based on subtle vibration magnification with experiments,and demonstrate the influence of recording time and video frame rate on the results of heart rate detection.Finally,the thesis gives the prospect of future research on this subject.Non-contact heart rate detection is just one of the broad applications of subtle vibration magnification technology.Continuous innovation and optimization of algorithms are the foundation for the development.Unsupervised vibration magnification,real-time vibration magnification and multi-source vibration magnification technologies need to be further explored.