Research On Heart Rate Detection Method From Face Videos Under Motion Interference

Heart rate(HR)is an essential sign to reflect the physiological health state of human beings.The traditional methods for HR measurement are based on contact sensors which are not convenient or comfortable for subjects.To address these issues,imaging photoplethysmography(IPPG)technique is presented for contactless HR estimation from the face videos recorded by a camera.It can be widely used in many application fields,such as telemedicine,fitness exercise and affective computing.However,the subjects in realistic environments often have spontaneous head movements and facial expressions which severely degrade the performances of current pulse signal extraction and HR measurement methods.In order to solve the problem of motion interference in HR measurement from face videos,this paper makes a thorough study on the basic principle of IPPG technique.Three motion-resistant methods for HR estmation from face videos are proposed,which are based on a general skin optical model.The final HR measurement results are visualized by video amplification.The main research works are summarized as follows:(1)For the rigid and non-rigid motion interference in face videos,a non-contact HR estimation method based on multi-signal weighting is presented.The discriminative response map fitting method and KLT tracking algorithm combined with the chrominance features are conducted to eliminate the influence of rigid motion.The weights in frequency and spatial domain are assigned by frequency and gradient prior and the non-rigid motion interference is eliminated by multi-signal weighting which realizes accurate pulse wave extraction and HR estimation.(2)Aiming at the sudden motion interference in face videos,a novel HR estimation method based on wavelet time-frequency analysis is established.This method uses wavelet time-frequency analysis to evaluate the quality of sub-pulse wave signals corresponding to different face patches,and fuses these patches by combining the spatio-temporal information to generate high-quality pulse wave signals,which realizes the effective HR estimation under the sudden movement interference.(3)Aiming at the problem that tiny face color change signal is easily submerged by motion artifact noise,a Gaussian derivative filtering method is proposed for HR estimation from face videos.A gaussian derivative temporal filter is designed to separate the pulse signal and motion noise according to the difference of third-order derivatives in the time-domain.By combining with IPPG methods,it realizes the effective HR estimation under motion artifact interference.(4)Based on the Eulerian video magnification method,a visual system for HR measurement from face videos is developed and realized.It reveals the skin color changes as blood volume changes caused by the heart beats and visualizes the subcutaneous blood distribution,which provides a reference for the prevention and diagnosis of cardiovascular disease.The proposed methods in this paper are tested on the video databases of subjects with different ages and genders.The results show that our methods can realize accurate pulse wave extraction and effective HR estimation by suppressing the influence of motion disturbance.