Research On Heart Rate Physiological Signal Monitoring Method Based On RPPG

The heart rate is an important indicator for assessing cardiovascular health,and its measurement plays a crucial role in the prevention and treatment of cardiovascular diseases.However,traditional contact-based methods for measuring heart rate require the use of devices such as electrocardiograms,heart rate belts,and wristbands,which are inconvenient to use and easily affected by factors like motion and external interference,thereby affecting the accuracy of the measurements.Therefore,non-contact methods for heart rate measurement have increasingly promising applications.Remote Photoplethysmography(rPPG),as a primary noncontact method for heart rate measurement,aims to improve the accuracy of detection.This thesis focuses on two steps in the rPPG technique: the selection of the Region of Interest(ROI)and the extraction of the Blood Volume Pulse(BVP)signal,with the goal of enhancing their precision.To address the issue of unstable feature point detection caused by non-rigid facial movements in preprocessing,this thesis proposes a stable training method for facial feature point detection.In the ROI selection step,a Supervision-by-Registration(SBR)algorithm is employed as the basic framework,and a Long Short Term Memory(LSTM)network is incorporated as an effective source of supervision.To enhance the stability of key facial structural feature points,a weight mask function is introduced.Additionally,a smoothness consistency loss function is proposed to improve the model’s stability.Through training on unlabeled videos,this method achieves accurate,stable,and coherent detection of facial feature point sequences.To improve the accuracy of heart rate calculation,this thesis investigates a video-based method for heart rate detection,with a focus on the extraction of the Blood Volume Pulse(BVP)signal.A standardized rPPG signal generation method is proposed.A meta-rPPG network is employed as the generator in a Generative Adversarial Network(GAN),where videos are inputted to the generator,and the predicted rPPG signals are supervised by the labels during training.To optimize the generator’s predictions,a mathematical modeling approach is used to generate standardized rPPG signals,which are then fed to the discriminator for adversarial training.This process enables the generator to learn the morphology and frequency distribution of the standardized signals.As a result,the signals predicted by the generator become closer to the distribution of real signals,which aids in subsequent heart rate calculations.This thesis verifies the accuracy and robustness of the proposed method through multiple datasets.In addition,a Python-based rPPG heart rate monitoring program is implemented on the web,which generates real-time visualized heart rate monitoring results,demonstrating the research results more intuitively.