Research On Heart Rate Measurement System Based On Face Video

With the raising attention of early disease prevention,the collection and detection of personal physiological information has been paid more and more attention.Heart rate,as one of the most important physiological indexes of human body,can reflect the physiological health and emotional state,which has special research significance.However,the traditional heart rate measurement methods are not only limited by the measuring equipment and site,but also bring discomfort to the subjects.Therefore,the non-contact heart rate measurement method with high portability and simple operation has attracted more and more attention.Especially,the heart rate measurement method based on face video has great research value and application prospect.However,the changes of environment light,facial expression and head movement bring great challenges to non-contact heart rate measurement.This paper proposes corresponding solutions for each step of heart rate measurement system based on face video.In the step of face ROI selection,we detect the key points of face in face video and define multiple face ROI.Through comparative experiments,we find that the best ROI is to remove the skin area of eyes and mouth.It preserves as much heart rate information as possible,and it also reduces expression noise at the source.In addition,it can track the motion through the face key point algorithm,and then reduce the noise caused by the head motion.In the extraction of heart rate original signal,RGB color model is used in this paper,which can extract heart rate signal with good quality from the green channel of face ROI.The rationale for this approach is that green light is better absorbed by hemoglobin than red and blue light,so it can penetrate deeper into the skin to detect the vascular system and thus better detect heart rate information.In the process of noise processing,according to the illumination in the background area,the illumination noise on the face is calculated by adaptive filtering,so that the illumination noise can be removed.For expression noise,a smoothing method of heart rate signal is proposed to reduce the violent fluctuation of heart rate signal caused by expression.Finally,the detrend filter and band-pass filter are used to further filter out other noises and obtain the signal close to the real heart rate.In heart rate calculation,a weighted average heart rate calculation method is proposed.Compared with the traditional calculation method,it can quickly combine the results of multiple heart rate measurements to improve the accuracy and stability of the final output heart rate.Finally,four groups of comparative tests are carried out in the challenging manob-hci data set to verify the effectiveness of the above scheme.