Heart Rate Detection Method And System Design Based On Machine Vision

Heart rate is an important indicator of cardiovascular function and has significant clinical value and research significance in the assessment of exercise intensity,evaluation of cardiorespiratory adaptability,diagnosis of cardiovascular diseases,and postoperative evaluation.Currently,commonly used heart rate measurement instruments such as electrocardiograms and pulse wave detectors have disadvantages such as complex operation and high cost.They also require sensors to be attached to the skin,are susceptible to interference,and can cause discomfort in people with fragile skin.As a result,a machine vision-based non-contact heart rate detection method based on remote photoplethysmography(rPPG)has been proposed and widely studied.This method uses a camera to capture facial video and calculates heart rate values by analyzing facial video information.It has the advantages of convenient measurement,high efficiency,and no need for skin contact.However,existing machine vision-based non-contact heart rate detection methods also have some problems and limitations in practical applications.During the detection process,it is necessary to ensure the stability of ambient light and shooting angle;otherwise it will cause changes in the amplitude and frequency of facial skin color changes which will affect the accuracy of measurement results.To improve the accuracy and reliability of machine vision-based heart rate detection this study conducts research on non-contact heart rate detection based on rPPG signals through the following work:(1)A Gaussian heat map facial key point prediction method based on U-Net is proposed to obtain the facial region of interest(ROI).The encoder of U-Net extracts high-level features such as edges and textures from facial images.The decoder predicts the corresponding Gaussian heat map for each pixel point,where the value of each point represents the probability distribution of facial key points.According to the pixel position corresponding to the maximum probability value of the heat map,ROI can be determined.Compared with traditional ROI extraction methods such as Viola-Jones based on skin color thresholds and facial geometry,this method does not rely on these feature information and can more flexibly adapt to different ages,skin colors and facial expressions.It can more accurately obtain ROI.In addition,it has been proven on datasets such as UBFC-RPPG that this method can adapt to different lighting and posture changes and has advantages in measurement accuracy,robustness and stability.(2)A comparative study was conducted on heart rate detection based on three traditional methods: Fast-ICA,CHROM and POS.Baseline offset correction and other methods were used to denoise and optimize the original rPPG signal to improve the accuracy of heart rate detection.Heart rate estimation was performed using peak values and power spectral density of rPPG signals respectively.A new model LA-Res2 Net combined with traditional methods for heart rate detection is proposed.It is also the first time that a Gramian angular field algorithm is used to extract feature images of rPPG signals as input to the model.Experiments have shown that this method outperforms existing methods in multiple indicators with a coefficient r reaching 98%.(3)A heart rate detection GUI control system based on PyQt technology was designed.This system integrates the algorithms mentioned above.Experimental results show that this system not only has a friendly human-computer interaction user interface but also can quickly and accurately perform heart rate monitoring which can help medical professionals more conveniently perform heart rate monitoring in practical applications.