Measurement Of Non-contact Human Heart Rate Based On Facial Video

Heart rate is an important physiological information reflecting human body and physical and mental conditions,and contains a lot of information about health and emotional activities.Therefore,real-time heart rate measurement has been widely used in disease prevention,physiological health monitoring,nursing and many other fields.In order to solve the physical or psychological inconvenience caused by the traditional heart rate measurement method,in this paper,a non-contact human heart rate measurement algorithm based on facial video is studied using Remote Photo-plethysmography(rPPG).Compared with the traditional heart rate measurement method,this study has the advantages of more simple and comfortable,more convenient,small size,low cost,and can solve the problems of excessive manpower cost,equipment inconvenience and application scene limitations.Most of the existing researches on non-contact extraction of human heart rate based on rPPG are in a relatively static state.However,the accuracy of heart rate measurement will be affected by unstable light or motion in the real scene.Based on this,this paper carried out research and optimization of anti-light interference and anti-motion interference heart rate measurement method,and designed experiments to verify the accuracy and robustness of the non-contact heart rate measurement method designed in this paper.The specific research work of this paper is as follows:1)Based on the physiological and optical basis of rPPG technology,the feasibility of non-contact heart rate measurement and the interference factors affecting the accuracy and efficiency of heart rate measurement are analyzed.To solve the problem that ROI could not be accurately located during pulse source signal collection,the existing face key point detection algorithm is improved.There is a cascade relationship between the face key point detector and SSD face detector,which effectively improves the accuracy of the detection of key points.At the same time,key point detection algorithm is used for ROI to complete the basic positioning operation,and the selection of ROI must be reliable,so as to effectively reduce the noise in the pulse wave signal.2)There are four methods involved in the pretreatment of pulse wave signals.The effectiveness of the four pretreatment methods is compared by experiments,and the method based on chroma combination was integrated with the second derivative differential approximation method to improve the expression ability of heart rate signal,thus improving the accuracy of heart rate measurement.3)The anti-noise level is lower than that of the traditional heart rate extraction algorithm.The heart rate extraction algorithm studied by the author in this paper is based on deep learning.Meanwhile,based on the Convolutional Neural Networks(CNN)as the technical premise,a convolutional neural network(CNN)is used to design a heart rate information feature extractor,which integrates the spectral features and the periodic characteristics of the heart rate signal together to form a time-frequency representation,while taking the parameter as the input.The spatio-temporal characteristic images are mapped to their corresponding heart rate values,and the relationship between adjacent measured values of two video clips is considered.The Gated Recurrent Unit(GRU)is used to simulate the time relationship between subsequent measured values,so that the model can predict accurate heart rate values.4)Through the comparison experiment with some typical rPPG methods on multiple public data sets,it is proved that the improved algorithm has certain improvement in the measurement accuracy and anti-interference.At the same time,this paper also analyzes the key factors that affect the heart rate measurement value in the real scene,and these factors provide a potential way to further improve the proposed method.