Research On Micro Magnification Amplification Algorithm Based On Video Image

The range of motion seen by the human eye has a boundary,and it is impossible to perceive a very small range of motion.The micro motion magnification algorithm is to magnify the micro motion that is not observed by the human eye,which expands the perception range of the human eye to a certain extent.Video motion magnification algorithms has high practical value in the fields of medical treatment and engineering measurement.According to different processing methods,micro motion amplification algorithms are divided into Lagrangian motion amplification algorithms,Euler video amplification algorithms,and learning-based motion amplification algorithms.Among them,the Euler video magnification algorithm needs to manually set multiple parameters and repeatedly try different parameters to get better magnification results.The learning-based motion magnification algorithm uses deep learning to replace the filter part of the Euler video magnification algorithm,and obtains a better magnification effect.However,the problem is that the video needs to be processed frame by frame,and each frame of image needs to run a nerve.The network model results in slower operation speed.Firstly,in order to solve the problem that the Euler video magnification algorithm needs to manually select multiple parameters,this paper proposes a fully automatic micro motion magnification algorithm.Through a detailed analysis of the impact of the magnification factor,the passband,and the number of pyramid layers on the micro motion amplification algorithm and the determining factors of each parameter.It is determined that the magnification factor is related to the motion frequency and amplitude of the foreground target in the video,and the passband is related to the range of moving frequency of the target in the video,and the number of pyramid layers is related to the motion amplitude of the foreground target.Based on this,a framework for automatically calculating the parameters of the Euler video amplification algorithm from the original video was established by applying foreground detection,time-frequency analysis and other methods,thereby implementing a fully automatic micro motion amplification algorithm.The simulation experiments show that the parameters obtained by the full-automatic micro-motion amplification algorithm proposed in this paper have good amplification effect.Secondly,in order to solve the problem of slow speed of learning-based motion amplification algorithm,this paper proposes Shuffle Net-based pre-cropped micro motion amplification algorithm.On the one hand,this paper analyzes each module of the learning-based video motion magnification algorithm in detail,and modifies the model into a lightweight model with smaller parameter scale,thereby achieving model compression and improving the algorithm's operation speed.On the other hand,before the picture is input to the model,the foreground detection algorithm is used to cut out the foreground part of the picture,and the foreground maximization algorithm is designed to input the largest common area of the foreground of adjacent frames into the model,thereby realizing the algorithm structure optimization.The experiments show that the Shuffle Net-based pre-cropped micro motion amplification algorithm proposed in this paper effectively improves the processing performance of the amplification algorithm and shortens the calculation time.Finally,a heart rate monitoring system based on video images was built to implement the single heart rate measurement and continuous heart rate monitoring functions.This system uses video's non-contact heart rate method,which has a more comfortable measurement experience than traditional devices.Compared with the traditional finger clip type medical heart rate measuring device,the algorithm used by this system can accurately measure the user's heart rate value.