Research On Heart Rate Detection Algorithm For Face Video From The Perspective Of Signal Reconstruction

Heart rate(HR)is one of the important physiological indicators of the human body,convenient heart rate detection method is an important guarantee for health management and disease prevention.Conventional heart rate detection methods such as electrocardiogram(ECG)and pulse oximeter are contact-based,which are inconvenient to use and easily lead to unhygienic problems.In recent years,non-contact heart rate detection methods based on face video have received extensive attention and indepth research from academia and industry,and have made great progress.However,this technique is easily affected by various factors such as lighting changes,face movement,region of interest(ROI)selection,and its accuracy is low and stability is poor compared to contact detection methods.In this dissertation,we proposed heart rate detection algorithm based on adaptive variational mode decomposition reconstruction and group sparse representation reconstruction from the perspective of heart rate signal reconstruction,and develop a real-time heart rate detection system for community correction centers according to practical application needs,the main works are as follows:1.Adaptive variational mode decomposition reconstruction heart rate detection algorithm.The existing mode decomposition methods suffer from the problem that the number of decomposed modes is difficult to determine,which leads to mixed or over-decomposed and underdecomposed modes.To this end,an adaptive variational mode decomposition reconstruction heart rate detection algorithm is proposed.First,the signal modal number is obtained by decomposing the remote photoplethysmography(r PPG)signal using empirical mode decomposition,and the adaptive variational mode decomposition algorithm is designed to obtain modes with different frequencies and bandwidths.Then,heart rate signal is reconstructed according to heart rate frequency range and mode center frequency.Finally,power spectral density analysis is performed to calculate heart rate.Experimental results on the public datasets UBFC and PURE show that the mean absolute error in heart rate detection is less than5 bpm,and the root mean square error is reduced by 2.19 bpm compared with the empirical mode decomposition method.2.Group sparse representation reconstruction heart rate detection algorithm.The heartbeat of the same human body is consistent,but the different distribution of blood vessels and inconsistent noise interference on the face result in different color variations in the heartbeat formation.Based on this fact,group sparse representation reconstruction heart rate detection method is proposed.First,the region of interest is selected and divided into multiple sub-regions,the sub-region RGB channel signal is extracted for distortion compensation,the chrominance signal is calculated to constitute the raw heart rate signal.Meanwhile,based on the pulse signal periodicity and pulsatility,a mixed dictionary consisting of discrete cosine base signal and wavelet base signal is constructed.Then,the raw heart rate signal is structurally sparsely decomposed.Finally,the heart rate signal is reconstructed for power spectral density analysis to calculate heart rate.Experiments on two public datasets have verified the effectiveness of the method.Compared with the DAOMP method on the UBFC dataset,the mean absolute error is reduced by 3.732 bpm,the root mean square error is reduced by 8.824 bpm,the accuracy is improved by 3.546%,and the Pearson correlation coefficient is improved by 0.223.3.Real-time heart rate detection system design and implementation.Develop a portable real-time heart rate detection system for community correction centers based on group sparse representation reconstruction heart rate detection algorithm.The system is designed with distributed architecture and divided into client-side and server-side.The client-side development is based on the Vue.js framework,with the main functions being the configuration of network communication and the display of real-time heart rate detection results,etc.The server-side development is based on the Python and third-party libraries,with the main functions being face detection and heart rate detection,etc.The effectiveness of the group sparse representation reconstruction heart rate detection algorithm is further verified through practical applications.