Research On Short Time Non-contact Heart Rate Measurement

Heart Rate(HR)is an important monitoring index in medical and other fields,and is widely used in medical health,fatigue detection and other fields.The heart rate measurement includes contact and non-contact methods.The contact heart rate measurement requires the active cooperation of the subjects,and the non-contact method breaks this restriction,improves the use experience of the subjects,and has a broad application prospect.This paper focuses on heart rate extraction based on short-term face video,and focuses on solving the problem of improving the accuracy of short-term non-contact heart rate measurement in real application environment.In order to solve the influence of environmental illumination and other factors on the accuracy of skin detection in practical application scenarios,this paper proposes an adaptive skin detection method based on nonlinear piecewise transformation.Through nonlinear piecewise transformation,the pixels with abnormal illumination are mapped to the normal illumination range,and the skin probability of the mapped pixels is calculated according to the skin color single Gaussian model,and the threshold is determined by Otsu algorithm to achieve adaptive adjustment of the threshold.The results of cross-validation experiments on the open data set Pra THEEPAN show that the adaptive skin detection method proposed in this paper has improved the skin detection rate by 36.12%,22.21% and 23.22% respectively compared with fixed threshold detection,elliptical model and single Gaussian model,and has improved the skin detection rate by 18.54%,5.79% and 0.98% respectively.The comparison experiment of non-contact heart rate measurement using different skin detection methods on the self-collected data set shows that the measurement error of this method in low illumination,normal illumination and strong illumination environment is reduced by 7.11%,2.354% and 4.725% respectively compared with the fixed threshold skin detection,which proves the effectiveness of this method.Aiming at the interference of face occlusion on non-contact heart rate measurement,this paper proposes a multi-ROI selection method based on adaptive skin detection.This method delimits three Regions Of Interest(ROI)from the face:forehead,left cheek and right cheek,and determines whether to select this ROI for signal extraction by calculating the proportion of skin pixels in each ROI.Multi-group comparative experiments were conducted on the face occlusion data set composed of masks and hats of different colors and types.The results showed that compared with the whole-face heart rate measurement method as ROI,the non-contact heart rate measurement error of this method decreased by 2.6% on average under different occlusion conditions,indicating that this method has good adaptability to different occlusion conditions.In order to improve the accuracy of short-term non-contact heart rate measurement,Adaptive Gamma Correction(AGC)algorithm is used to enhance the image of skin area in ROI.This method adaptively determines parameters using pixel mean and standard deviation in 5 * 5 templates γ Value,and gamma correct the skin area of the u channel.A large number of experiments on the self-collected data set show that the heart rate error extracted from the 6-second face video is less than 4bpm after using the AGC image enhancement algorithm.Compared with the non-image enhancement processing,the accuracy of heart rate detection in low illumination,normal illumination and strong illumination has increased by 2.275%,1.625% and1.339% respectively.