Research On HR Detection From Facial Video And Its Application

Heart rate(HR)is a crucial physiological index that reflects human health and is of great significance in the prevention and treatment of cardiovascular diseases.Accurate heart rate detection is essential for human health assessment and monitoring.However,most heart rate detection methods are contact measurements that require professional equipment to directly contact the tester’s skin.Contact measurement has certain limitations in terms of convenience of operation and portability of equipment,and long-term contact with skin can cause discomfort to the tester,making it difficult to meet the needs of daily heart rate monitoring.Remote photoplethysmography(rPPG)is an emerging non-contact measurement method that is low-cost and easy to control.However,most testers in published relevant research literature completed facial video acquisition in a relatively static state,meaning that the head movement range was small.When the tester’s head moves,the accuracy of heart rate detection decreases rapidly.To improve the anti-interference ability of the algorithm and enable it to accurately measure heart rate in daily scenarios,this paper proposes an innovative method:removing the fitted noise variation trend from the original signal to obtain the pure blood volume pulse(BVP)signal for heart rate analysis.To verify the accuracy of this method for heart rate detection in daily scenarios,this paper also applies the rPPG technique to driver’s heart rate monitoring.The main works are as follows:(1)To address the problem that the accuracy of heart rate detection by rPPG is easily affected by illumination change noise and motion artifact noise,a new idea is proposed to extract the BVP signal.The pure BVP signal can be obtained by removing the fitted noise trend from the original signal refer to high power and easy extraction peculiarity of noise.This method first tracks the ROI(region of interest)of the face and spatially averages the ROI to obtain the original signal.Then it uses the ‘sliding average cubic polynomial’ to obtain the fitted noise trend,removes the trend from the original signal and then undergoes signal post-processing to obtain pure BVP signal with high SNR(signal-to-noise ratio).Finally,it calculates the frequency spectrum of the BVP signal and completes heart rate detection.After experimental verification,this method improves performance by 23.20% and 56.23% compared with the POS algorithm and the ICA algorithm,respectively.In terms of algorithm accuracy,this method improves by 37.19% and 51.75% compared with POS and ICA respectively.(2)To apply rPPG to daily scenarios,this paper attempts to apply this technology to driver’s heart rate monitoring.A system for rPPG heart rate detection is designed for this scenario.The system uses a USB camera modulated by a video capture program to continuously capture images of the driver’s face and uses the algorithm proposed in this paper to extract the BVP signal for heart rate analysis.This paper also designs an algorithm to analyze heart rate in the time domain.The algorithm takes advantage of the fact that the ideal BVP signal has a repetitive wave with a small height and width of peak.This method can effectively eliminate the influence of repetitive waves and get a more accurate heart rate.The analysis shows that driver’s heart rate monitoring scenarios meet a new challenge compared to indoor environments: ambient light spectral variation.It is also more challenging to eliminate both incident light intensity variations and motion artifacts on the face.