Non Contact Heart Rate Detection Based On Video

Heart rate plays an important role in evaluating human health and preventing diseases-heart disease,hypertension and septic shock.For people who are inconvenient for contact heart rate detection such as burns and scalds,remote photoplethysmography(rPPG)is an effective and non-contact heart rate detection technology.This technology can measure and evaluate the human heart rate by collecting the data of exposed skin video from the camera within a certain distance;However,when the human body detects the heart rate in motion,the noise caused by motion and light change is much stronger than the hidden weak heart rate signal,resulting in large error in heart rate detection due to the influence of light change and motion.Therefore,aiming at the influence of complex situations such as light change and face movement on heart rate detection by remote photoelectricplethysmography,two different heart rate detection methods are proposed when the tester is in static and dynamic conditions.Aiming at the static situation,a denoising method based on the combination of complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and fast independent component analysis(fast ICA)is proposed to realize the accurate detection of non-contact heart rate.Firstly,68 key points of human face in video image are detected,get region of interest(ROI)and get the heart rate detection source signal;then decompose the source signal through CEEMDAN and select the inherent mode function in the appropriate frequency band for reconstruction.After fast ICA of the reconstructed signal,carry out fast Fourier transform(FFT)on the signals of RGB channels,and finally calculate the heart rate by using the frequency corresponding to the spectrum peak For contact heart rate detection,a heart rate detection method based on multi feature facial region and chroma fusion is proposed.Firstly,the nose and cheek ROI are selected from the video image,the RGB three channel pixel values of the two ROI are chroma fused respectively,and three groups of heart rate detection source signals are formed with the green channel pixel values of the nose ROI.After preprocessing such as interpolation and de trending,the source signal is denoised by singular spectrum analysis(SSA)and CEEMDAN decomposition.Finally,the denoised source signal is fused again,and the heart rate value is obtained by FFT.The heart rate detection experiments were carried out on 10 subjects,and the measurement results were compared with the test results of pulse blood oxygen detection reference instrument.When the subjects were still,the root mean square error and average absolute error were 0.72 bpm and 0.60 bpm respectively;With the increase of light change and human movement,the root mean square error and average absolute error of heart rate detection results are less than 5 bpm.Experimental results show that the proposed method can accurately extract heart rate signal from video images in complex environment,and improve the accuracy of heart rate detection.