CN114596963A - Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation - Google Patents

Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation Download PDF

Info

Publication number
CN114596963A
CN114596963A CN202210319966.6A CN202210319966A CN114596963A CN 114596963 A CN114596963 A CN 114596963A CN 202210319966 A CN202210319966 A CN 202210319966A CN 114596963 A CN114596963 A CN 114596963A
Authority
CN
China
Prior art keywords
signal
heart rate
sub
dictionary
sparse
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210319966.6A
Other languages
Chinese (zh)
Inventor
欧卫华
陈龙保
韩杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guizhou Education University
Original Assignee
Guizhou Education University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guizhou Education University filed Critical Guizhou Education University
Priority to CN202210319966.6A priority Critical patent/CN114596963A/en
Publication of CN114596963A publication Critical patent/CN114596963A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A61B5/024—Detecting, measuring or recording pulse rate or heart rate

Landscapes

  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Cardiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Physics & Mathematics (AREA)
  • Physiology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Molecular Biology (AREA)
  • Veterinary Medicine (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)

Abstract

The invention discloses a remote non-contact heart rate estimation method, system and equipment based on sparse structure representation, belongs to the field of physiological sign research and application, and solves the problem of inaccurate detection of the existing heart rate signals, and comprises the following steps: step 1, inputting a face video, and dividing a face into a plurality of sub-regions; step 2, constructing an average RGB pulse signal of each subregion, performing motion compensation, calculating a chrominance signal S by utilizing linear combination of RGB different channels, and calculating a signal-to-noise ratio; step 3, solving a coefficient matrix by using a structural dictionary consisting of cosine bases with different frequencies and wavelet bases with different scales, and reconstructing a heart rate signal by using the structural dictionary and the coefficient matrix; and 4, averaging the reconstructed sub-region heart rate signals, and performing power spectrum analysis to calculate the heart rate HR. The optimal sparse representation based on the dictionary is found, the heart rate signal is reconstructed by utilizing the dictionary and the sparse coding, the heart rate signal which is as real as possible is represented by using dictionary atoms as few as possible, and the accuracy of the algorithm is improved.

Description

Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation
Technical Field
The invention relates to the technical field of biological sign research and application, in particular to the technical field of remote heart rate estimation based on videos.
Background
The heart rate signal is an important physiological signal, and can be used for calculating physiological parameters such as heart rate, heart rate variability, respiratory rate, blood pressure and the like. Conventional heart rate signal detection methods require skin contact (e.g., electrocardiogram), but can lead to discomfort and unsanitary problems. The remote heart rate estimation method based on the facial video is a non-contact heart rate measurement method and is convenient to use. The beating of the heart causes subtle changes in the color of the skin surface, particularly on the facial skin surface. On the basis, remote photoplethysmography (rPPG) is proposed, and a pure heart rate signal is extracted from weak color change by using technologies such as computer vision, signal processing and the like. Remote non-contact heart rate estimation falls into three broad categories: blind source separation, color subspace projection, and deep learning.
With the successful application of deep learning in the fields of computer vision, natural language processing and the like, the method of an end-to-end deep learning network is introduced into the field of remote heart rate estimation based on facial videos. Spetlik et al propose an HR-CNN network that acquires a pulse signal through a two-dimensional convolutional network based on signal-to-noise ratio constraints. Considering the effects of abnormal lighting and motion, Chen et al introduced an attention mechanism and proposed a deep physics approach by constructing a feature map that learns spatiotemporal information. In order to be able to better represent the color variation information, Niu et al convert the RGB signal into a YUV space, extract the heart rate variation information, and convert it into a time-space diagram for heart rate detection. Deep learning methods show better results on data sets, but the models are complex and weakly generalized. However, the deep learning method faces the challenges of requiring a large number of labeled data sets to train the network, and has a low calculation speed, many parameters and the need of pre-training the network model.
The method of blind source separation and color subspace decomposition in the traditional method has better interpretability and faster running speed. The sparse representation has wide application in computer vision, is introduced into the field of reconstruction of heart rate signals, and achieves good effect. Liu et al propose a signal sparse representation method for adaptive interference orthogonal matching pursuit, which recovers a heart rate signal damaged by face instability by constructing a discrete sine dictionary by using an adaptive interference orthogonal matching pursuit algorithm. The heart rate signal is a physiological signal and includes both a periodic signal and a pulsating signal. Although the method of Liu et al achieves good results, the pulsatility of the signals between the sub-regions and the correlation between the sub-regions are not considered, resulting in inconsistency of the estimated heart rate signals, which results in inaccuracy of the estimation of the heart rate signals.
Disclosure of Invention
The invention aims to: in order to solve the technical problem that the heart rate signal estimation is inaccurate due to the fact that the pulsatility of signals between sub-regions and the correlation between the sub-regions are not considered in the signal sparse representation method, the invention provides a remote non-contact heart rate estimation method, system and equipment based on sparse structure representation.
The invention specifically adopts the following technical scheme for realizing the purpose:
a remote non-contact heart rate estimation method based on sparse structure representation comprises the following steps:
the method comprises the steps of 1, inputting a face video, fixedly selecting forehead and cheek parts as interested areas of a detected face, and partitioning to obtain a plurality of sub-areas (in remote video heart rate detection, the face detection is firstly needed, in different scenes, the face can be processed in aspects of movement, rotation and the like, and the face detection method is timely and accurate, considering the real-time property and the positioning and tracking effect of the detection method, the face detection and tracking are carried out by adopting a PCN (pulse coupled network) deep network face recognition algorithm to obtain coordinate points and width of the face, and because the forehead and cheek areas of the face have rich heart rate information, based on the analysis, the forehead and the cheek are fixedly selected as the ROI by adopting the PCN algorithm, and the forehead and the cheek are partitioned into the plurality of sub-areas);
step 2, constructing an average RGB pulse signal of each subregion, performing motion compensation on the RGB pulse signal (eliminating sudden change of pixels and generating a primary de-noised color signal), calculating a chrominance signal S by utilizing linear combination of different channels of RGB, calculating the signal-to-noise ratio of each region based on the chrominance signal S (the quality of chrominance signals of partial subregions is poor due to the factors of foreign matters on a face shielding surface, uneven light distribution, motion and the like), selecting the subregion with rich heart rate information by calculating the signal-to-noise ratio of the chrominance signal, taking the subregion with the signal-to-noise ratio of the subregion larger than the total signal-to-noise ratio as a high-quality quantum region, defining the corresponding chrominance signal as the high-quality chrominance signal, and generating a high-quality quantum region chrominance signal M;
step 3, for real heart rate signal MpulseSparse structure representation is carried out, and heart rate signals are reconstructed
Figure BDA0003571248220000031
The reconstructed heart rate signal is further computed from a high quality sub-region chrominance signal M, which consists of the chrominance signals of the sub-region with a high signal-to-noise ratio, i.e. the reconstructed heart rate signal is obtained from the high quality sub-region chrominance signal M
Figure BDA0003571248220000032
Wherein the content of the first and second substances,
Figure BDA0003571248220000033
the chrominance signal representing the selected high quality sub-region (the chrominance signal M of the high quality sub-region, which is linearly calculated from the average RGB pulse signal, is derived from the true heart rate signal (M)pulse) And a noise signal (M)noise) Composition, i.e. M ═ Mpulse+MnoiseSparse structural representation reconstructs the heart rate signal
Figure BDA0003571248220000034
As an estimate of the heart rate signal),
Figure BDA0003571248220000035
representing the reconstruction of the heart rate signal for selected high quality quantum regions, X is obtained as follows: the method comprises the steps of utilizing cosine bases with different frequencies and wavelet bases with different scales to form a mixed structure dictionary D (utilizing different cosine bases to reconstruct the periodic characteristics of heart rate signals and utilizing wavelet bases with different scales to reconstruct the fluctuation characteristics of the heart rate signals), solving a coefficient matrix A, selecting proper dictionary atoms from the mixed dictionary by a sparse structure method to represent the heart rate signals, carrying out sparse coding by a greedy algorithm to form the coefficient matrix A, and reconstructing the heart rate signals by an inner product method of the mixed structure dictionary D and the coefficient matrix A, namely X is D.A;
and 4, averaging the reconstructed heart rate signals of all the sub-regions, and then analyzing the power spectrum to obtain the heart rate HR.
In the technical scheme of the application, a section of video containing a face is input, forehead and cheek parts of the detected face are extracted to be used as a region of interest (ROI), a plurality of sub-regions are segmented in the ROI, an original color signal is extracted, the original color signal is analyzed, and sudden change of pixels caused by rigid motion and unstable lighting conditions of the face is eliminated through motion compensation; calculating a chrominance signal and a signal-to-noise ratio of each subregion, taking the subregion of which the signal-to-noise ratio is greater than the total signal-to-noise ratio as a high-quality quantum region, defining the corresponding chrominance signal as a high-quality chrominance signal, and generating the high-quality chrominance signal; a structural dictionary consisting of cosine bases with different frequencies and wavelet bases with different scales is adopted to efficiently reconstruct the period and amplitude of the heart rate signal; solving a sparse matrix, and selecting more proper dictionary atoms from the selected dictionary to represent the heart rate signals by using a sparse structure method; reconstructing a heart rate signal by adopting a dictionary and sparse matrix inner product method; averaging the reconstructed heart rate signal matrix; and finally, performing power spectrum analysis on the average heart rate signals of all the sub-regions to obtain the final heart rate. This application divides into a plurality of subregions with the people's face, extracts the high-quality chrominance signal in subregion, constructs mixed structure dictionary, seeks the best sparse representation based on mixed structure dictionary, utilizes mixed structure dictionary and sparse coding reconsitution heart rate signal, expresses the heart rate signal as far as possible with the dictionary atom as few as possible. The accuracy of the algorithm is improved.
Further, in step 2, calculating the average value, the absolute deviation of the average value and the standard deviation of the RGB pulse signals of each subregion (the standard deviation reflects the degree of deviation of the signals from the average value), and then calculating the deviation amplitude of each atom from the average value, wherein the average value is added to the standard deviation for the atoms exceeding the average value and exceeding the standard deviation; the atoms below the mean and below the standard deviation are motion compensated with the mean minus the standard deviation (color signal contains a lot of noise, heart rate signal appears very weak, heart rate does not change significantly in a short time).
Figure BDA0003571248220000041
Figure BDA0003571248220000042
Figure BDA0003571248220000043
Figure BDA0003571248220000051
Wherein i is the number of channels, i ═ 1, 2, 3 respectively represent RGB channel signals, and specifically i ═ 1 is the R channel; i is 2, which is a G channel; i-3 is B channel, l is signal length, Ci(t) the signal length of the i-th channel at time t, Ci_avgIs the signal average of the i channel, Ci_MeIs the signal mean deviation of the i channel, Ci_SDIs the standard deviation of the signal for the i channel.
Further, in step 2, the chrominance signal S is represented by S ═ Xf-αYfCalculated where α is two signals XfAnd YfA ratio of standard deviation of (a) to (a) (X)f)/σ(Yf)，XsAnd YsIs the fusion of the normalized RGB pulse signals after motion compensation, XsAnd YsThe calculation function of (c) is as follows:
Xs＝3Rn-2Gn
Ys＝1.5Rn+Gn-1.5Bn
wherein R isn，GnAnd BnA color signal for each sub-region;
Xfand YfAre each XsAnd YsIs calculated to obtain XsAnd YsThen, by XsAnd YsPassing through a band-pass filter to obtain XfAnd YfFinally, calculating an S chrominance signal;
the signal-to-noise ratio function for each subregion is as follows:
Figure BDA0003571248220000052
wherein S isfRepresenting spectral values of the chrominance signal after S-Fourier transformation, f representing the frequency limiting the heart rate, f being in the range 0.6, 4]，(f＝[0.4，6])，WtRepresenting the calculated sliding window size.
Further, in step 3, the obtaining method of X is as follows: using cosine basis DcSum wavelet DwConstructing a mixed structure dictionary D (since any periodic discrete signal can be expressed as a linear combination of cosine-based signals with different frequencies, heart rate signals have different periodicities and different amplitudes, discrete cosine-based sparse representation and wavelet-based sparse representation are used, in order to reduce the size of the dictionary to improve the performance of the algorithm, the frequency of the cosine-based signal is limited in a heart rate frequency range (0.6-4HZ), and constructing a dictionary D containing a cosine base and a small baseA dictionary of mixed-structure wave-based),
Figure BDA0003571248220000061
Dw＝WaveDict(Short3，Nb，j，b)
wherein k isiFor the ith in a discrete cosine dictionarythFrequency of atoms, kiAnd k isi+1By a difference of
Figure BDA0003571248220000062
FrFor video frame rate, NbFor the length of the signal generated, j is the level vector, b is the translation factor, fLIn order to be able to generate the length of the signal sequence,
the mixed dictionary model is:
Figure BDA0003571248220000063
furthermore, in step 3, the coefficient matrix is solved, and the method of sparse structure (sparse structure representation method is to true heart rate signal (M)pulse) Sparse structure representation is carried out, and an estimated value of the heart rate signal is obtained. Although each zone contains different heart rate information, its basic heart rate component should be the same and the heart rate signals of the different zones should be similar. Thus, for heart rate signals reconstructed in different sub-regions, by minimizing L of the coefficient matrix1、L2Parameterization to select the same dictionary atoms in the signal-based dictionary, min | | A | | sweet wind1，2. Furthermore, the reconstructed heart rate signal X and the real heart rate signal MpulseThere should be a small error between them. This problem can be solved by minimizing L between the true heart rate signal and the reconstructed heart rate signal2Parameterized to be bounded, i.e. minimum
Figure BDA0003571248220000071
Based on this, the reconstruction of the heart rate signal can be expressed as solving the following model) from the hybrid dictionarySelecting proper dictionary atoms to represent pulse signals, carrying out sparse coding by a greedy algorithm (so as to find the most proper dictionary atoms), and solving a coefficient matrix specifically comprises the following steps:
Figure BDA0003571248220000072
obtaining sparse codes, forming a coefficient matrix A,
wherein λ is used to balance signal reconstruction accuracy
Figure BDA0003571248220000073
And a penalty term | | A | | non-woven phosphor1，2The regularization parameter of (a);
further, in step 3, reconstructing the heart rate signal, and reconstructing the heart rate signal by adopting a method of inner product of a hybrid dictionary D and a coefficient matrix a, wherein an inner product model is as follows:
X＝D·A。
an ideal PPG signal should contain periodicity and fluctuation rate. The periodicity can be expressed as a linear combination of a small number of cosine bases and the amplitude can be expressed as a linear combination of a small number of wavelets. Therefore, the ideal PPG signal can be represented with a small number of mixed-structure dictionaries, i.e. a is a sparse matrix. The heart rate signal is formed by a change in the heartbeat, and the epidermis of the face is changed in different colors due to interference of light, motion, and other noises.
Furthermore, the reconstructed heart rate signals of the sub-regions are averaged to obtain a heart rate signal, which specifically comprises:
Figure BDA0003571248220000074
where N is the total number of selected sub-regions, xnIs the heart rate signal of the selected sub-area.
Further, in step 4, the power spectrum analysis is performed on the averaged heart rate signal, and the component with the highest peak value is selected as the main component of the heart rate signal, that is, the frequency f corresponding to the component with the highest power is adoptedHRFinally, time-to-time conversion to time domainObtaining a final heart rate HR, specifically: HR ═ fHR*60。
A system for remote non-contact heart rate estimation based on sparse structure representation, comprising:
the subregion acquisition module is used for inputting a face video, fixedly selecting forehead and cheek parts of a detected face as regions of interest, and partitioning the regions of interest to obtain a plurality of subregions;
the high-quality quantum region chrominance signal generating module is used for constructing an average RGB pulse signal of each sub-region, performing motion compensation on the RGB pulse signals, calculating a chrominance signal S by utilizing linear combination of different RGB channels, calculating the signal-to-noise ratio of each sub-region based on the chrominance signal S, taking the sub-region with the signal-to-noise ratio of the sub-region larger than the total signal-to-noise ratio as a high-quality quantum region, defining the corresponding chrominance signal as a high-quality chrominance signal, and generating a high-quality quantum region chrominance signal M;
a heart rate signal reconstruction module for reconstructing a real heart rate signal MpulseSparse structure representation is carried out, and heart rate signals are reconstructed
Figure BDA0003571248220000081
The high-quality sub-region chrominance signal M consists of chrominance signals of sub-regions of high signal-to-noise ratio, i.e.
Figure BDA0003571248220000082
Wherein the content of the first and second substances,
Figure BDA0003571248220000083
a chrominance signal representing a selected high quality quantum region,
Figure BDA0003571248220000084
representing the reconstruction of the heart rate signal for selected high quality quantum regions, X is obtained as follows: forming a mixed structure dictionary D by using cosine bases with different frequencies and wavelet bases with different scales, solving a coefficient matrix A, selecting proper dictionary atoms from the mixed dictionary by using a sparse structure method to represent heart rate signals, carrying out sparse coding by using a greedy algorithm to form the coefficient matrix A, and adopting a mixed structure methodReconstructing a heart rate signal by a method of integrating the inner product of the dictionary D and the coefficient matrix A, namely X is equal to D.A;
and the heart rate obtaining module is used for averaging the reconstructed heart rate signals of all the sub-regions and then carrying out power spectrum analysis to obtain the heart rate HR.
A computer device comprising a memory storing a computer program and a processor implementing the steps of the sparse structure representation based remote contactless heart rate estimation method when executing the computer program.
The invention has the following beneficial effects:
1. dividing a human face into a plurality of sub-regions, extracting high-quality chrominance signals of the sub-regions, constructing a mixed structure dictionary, searching for optimal sparse representation based on the mixed structure dictionary, reconstructing a heart rate signal by using the mixed structure dictionary and sparse coding, and representing the heart rate signal which is as real as possible by using dictionary atoms as few as possible, so that the accuracy of the algorithm is improved;
2. under the frame length of 1200, the best result is obtained when the MAE and the RMSE are lower than 5bpm, the accuracy rate of the detection method is higher than 96%, and the Pearson correlation between the detected heart rates is higher than 97%;
3. under the states of steady state, speaking, slow translation, slow rotation, medium rotation and fast translation, the total accuracy of the detection method exceeds 95%, and the Pearson correlation coefficient between the detected heart rate and the true value exceeds 98%. Under the states of steady state, speaking, slow translation, slow rotation and medium rotation, the Pearson correlation coefficient is more than 99 percent, which shows that the detection method has good performance to the states;
4. the comparison of other methods is carried out on the UBFC data set, the MAE and the RMSE of the method are lower than 5bpm, and the Acc and the PCC are higher than 96 percent, which shows that the method has stronger applicability; compared with the best-performing POS method, the MAE is improved by 1.697bpm, the RMSE is improved by 6.075bpm, and the Pearson correlation coefficient is improved by 11.9%, which shows that the method has higher algorithm robustness under the conditions of stable illumination and heart rate variation;
5. the detection method of the present invention detects a high degree of correspondence between heart rate and true HR for all states of the entire data set.
Drawings
FIG. 1 is a block diagram of a sparse structure representation-based remote non-contact heart rate estimation method of the present invention;
FIG. 2 is a flow chart of a method of remote non-contact heart rate estimation based on sparse structure representation according to the present invention;
FIG. 3 is a flow chart of sparse structure representation of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments.
All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
As shown in fig. 1 to 3, the present embodiment provides a remote non-contact heart rate estimation method based on sparse structure representation, including the following steps:
the method comprises the steps of 1, inputting a face video, fixedly selecting forehead and cheek parts as interested areas of a detected face, and partitioning to obtain a plurality of sub-areas (in remote video heart rate detection, the face detection is firstly needed, in different scenes, the face can be processed in aspects of movement, rotation and the like, and the face detection method is timely and accurate, considering the real-time property and the positioning and tracking effect of the detection method, the face detection and tracking are carried out by adopting a PCN (pulse coupled network) deep network face recognition algorithm to obtain coordinate points and width of the face, and because the forehead and cheek areas of the face have rich heart rate information, based on the analysis, the forehead and the cheek are fixedly selected as the ROI by adopting the PCN algorithm, and the forehead and the cheek are partitioned into the plurality of sub-areas);
step 2, constructing an average RGB pulse signal of each subregion, performing motion compensation on the RGB pulse signal (eliminating sudden change of pixels and generating a primary de-noised color signal), calculating a chrominance signal S by utilizing linear combination of different channels of RGB, calculating the signal-to-noise ratio of each region based on the chrominance signal S (the quality of chrominance signals of partial subregions is poor due to the factors of foreign matters on a face shielding surface, uneven light distribution, motion and the like), selecting the subregion with rich heart rate information by calculating the signal-to-noise ratio of the chrominance signal, taking the subregion with the signal-to-noise ratio of the subregion larger than the total signal-to-noise ratio as a high-quality quantum region, defining the corresponding chrominance signal as the high-quality chrominance signal, and generating a high-quality quantum region chrominance signal M;
calculating the mean, the absolute deviation of the mean and the standard deviation of the RGB pulse signals of each subregion (the standard deviation reflects the degree of deviation of the signals from the mean), then calculating the deviation amplitude of each atom from the mean, and the average plus the standard deviation of the atoms exceeding the mean and the standard deviation; the atoms below the mean and below the standard deviation are motion compensated with the mean minus the standard deviation (color signal contains a lot of noise, heart rate signal appears very weak, heart rate does not change significantly in a short time).
Figure BDA0003571248220000111
Figure BDA0003571248220000112
Figure BDA0003571248220000113
Figure BDA0003571248220000114
Wherein i is the number of channels, i ═ 1, 2, 3 respectively represent RGB channel signals, and specifically i ═ 1 is the R channel; i is 2, which is a G channel; i-3 is B channel, l is signal length, Ci(t) the signal length of the i-th channel at time t, Ci_avgIs the signal average of the i channel, Ci_MeIs the signal mean deviation of the i channel, Ci_SDIs the signal standard deviation of the i channel;
the chrominance signal S is defined by S ═ Xf-αYfCalculated where α is two signals XfAnd YfA ratio of standard deviation of (a) to (a) (X)f)/σ(Yf)，XsAnd YsIs the fusion of the normalized RGB pulse signals after motion compensation, XsAnd YsThe calculation function of (a) is as follows:
Xs＝3Rn-2Gn
Ys＝1.5Rn+Gn-1.5Bn
wherein R isn，GnAnd BnA color signal for each sub-region;
Xfand YfAre each XsAnd YsIs calculated to obtain XsAnd YsThen, by XsAnd YsPassing through a band-pass filter to obtain XfAnd YfFinally, calculating an S chrominance signal;
the signal-to-noise ratio function for each subregion is as follows:
Figure BDA0003571248220000121
wherein S isfRepresenting spectral values of the chrominance signal after S-Fourier transformation, f representing the frequency limiting the heart rate, the range of f being [0.6, 4 ]]，(f＝[0.4，6])，WtRepresenting a calculated sliding window size;
step 3, for real heart rate signal MpulsePerforming sparse structure representation, heavilyForm heart rate signal
Figure BDA0003571248220000122
The reconstructed heart rate signal is further computed from a high quality sub-region chrominance signal M, which consists of the chrominance signals of the sub-region with a high signal-to-noise ratio, i.e. the reconstructed heart rate signal is obtained from the high quality sub-region chrominance signal M
Figure BDA0003571248220000123
Wherein the content of the first and second substances,
Figure BDA0003571248220000124
the chrominance signal representing the selected high quality sub-region (the chrominance signal M of the high quality sub-region, which is linearly calculated from the average RGB pulse signal, is derived from the true heart rate signal (M)pulse) And a noise signal (M)noise) Composition, i.e. M ═ Mpulse+MnoiseSparse structural representation reconstructs the heart rate signal
Figure BDA0003571248220000125
As an estimate of the heart rate signal),
Figure BDA0003571248220000126
representing the reconstruction of the heart rate signal for selected high quality quantum regions, X is obtained as follows: the method comprises the steps of utilizing cosine bases with different frequencies and wavelet bases with different scales to form a mixed structure dictionary D (utilizing different cosine bases to reconstruct the periodic characteristics of heart rate signals and utilizing wavelet bases with different scales to reconstruct the fluctuation characteristics of the heart rate signals), solving a coefficient matrix A, selecting proper dictionary atoms from the mixed dictionary by a sparse structure method to represent the heart rate signals, carrying out sparse coding by a greedy algorithm to form the coefficient matrix A, and reconstructing the heart rate signals by an inner product method of the mixed structure dictionary D and the coefficient matrix A, namely X is D.A;
the method for obtaining X is as follows: using cosine basis DcSum wavelet DwA mixed structure dictionary D is constructed (since any periodic discrete signal can be represented as a linear combination of cosine-based signals of different frequenciesDiscrete cosine-based sparse representation and wavelet-based sparse representation are used. In order to reduce the size of the dictionary to improve the performance of the algorithm, the frequency of the cosine-based signal is limited in the heart rate frequency range (0.6-4HZ), and on the basis, a mixed structure dictionary containing cosine bases and wavelet bases is constructed,
Figure BDA0003571248220000131
Dw＝WaveDict(Short3，Nb，j，b)
wherein k isiFor the ith in a discrete cosine dictionarythFrequency of atoms, kiAnd k isi+1By a difference of
Figure BDA0003571248220000132
FrFor video frame rate, NbFor the length of the signal generated, j is the level vector, b is the translation factor, fLIn order to be able to generate the length of the signal sequence,
the mixed dictionary model is:
Figure BDA0003571248220000133
solving the coefficient matrix by means of a sparse structure (sparse structure representation is for the true heart rate signal (M)pulse) Sparse structure representation is carried out, and an estimated value of the heart rate signal is obtained. Although each zone contains different heart rate information, its basic heart rate component should be the same and the heart rate signals of the different zones should be similar. Thus, for heart rate signals reconstructed in different sub-regions, by minimizing L of the coefficient matrix1、L2Parameterization to select the same dictionary atoms in the signal-based dictionary, min | | A | | sweet wind1，2. Furthermore, the reconstructed heart rate signal X and the real heart rate signal MpulseThere should be a small error between them. This problem can be solved by minimizing L between the true heart rate signal and the reconstructed heart rate signal2Parameterized to be bounded, i.e. minimum
Figure BDA0003571248220000134
Based on this, the reconstruction of the heart rate signal can be expressed as solving the following model), selecting an appropriate dictionary atom from a mixed dictionary to represent the pulse signal, performing sparse coding with a greedy algorithm (to find the most appropriate dictionary atom), and solving a coefficient matrix specifically as solving the following model:
Figure BDA0003571248220000141
obtaining sparse codes, forming a coefficient matrix A,
wherein λ is used to balance signal reconstruction accuracy
Figure BDA0003571248220000142
And a penalty term | | A | | non-woven phosphor1，2The regularization parameter of (a);
reconstructing the heart rate signal, and reconstructing the heart rate signal by adopting a method of inner product of a hybrid dictionary D and a coefficient matrix A, wherein an inner product model is as follows:
X＝D·A。
an ideal PPG signal should contain periodicity and fluctuation rate. The periodicity can be expressed as a linear combination of a small number of cosine bases and the amplitude can be expressed as a linear combination of a small number of wavelets. Therefore, an ideal PPG signal can be represented with a small number of mixed-structure dictionaries, i.e. a is a sparse matrix. The heart rate signal is formed by a change in the heartbeat, and the epidermis of the face is changed in different colors due to interference of light, motion, and other noises.
Averaging the reconstructed heart rate signals of all the sub-regions to obtain heart rate signals, which specifically comprises the following steps:
Figure BDA0003571248220000143
where N is the total number of selected sub-regions, xnA heart rate signal for the selected sub-region;
step 4, averaging the reconstructed heart rate signals of all the sub-regions, and performing work againRate spectrum analysis to obtain the heart rate HR, power spectrum analysis of the averaged heart rate signal, and selecting the component with the highest peak value as the main component of the heart rate signal, i.e. adopting the frequency f corresponding to the component with the highest powerHRAnd finally, converting the time frequency into the time domain to obtain the final heart rate HR, which specifically comprises the following steps: HR ═ fHR*60。
Wherein, in fig. 2, I represents: face detection and ROI selection; II represents chroma signal extraction and motion compensation; III denotes the original RGB pulse signal; IV represents a sparse structure representation; v represents the average reconstructed heart rate signal; VI denotes the calculated heart rate; in fig. 3, a denotes a mixed structure dictionary (D); b represents a matching coefficient matrix (a); c represents the reconstructed sub-region heart rate signal; d represents averaging the reconstructed heart rate signals of the individual sub-regions.
In order to verify the superiority of the method in the video-based remote heart rate estimation, two data sets PURE and UBFC are selected for experiments, and in order to evaluate the comprehensive performance of the algorithm, the average absolute error (MAE), the Root Mean Square Error (RMSE), the accuracy (Acc) and the Pearson Correlation Coefficient (PCC) of the heart rate are selected as evaluation indexes. The heart rate error of the nth video sample is defined as follows:
Figure BDA0003571248220000151
wherein n represents the n-ththThe number of the test specimens is determined,
Figure BDA0003571248220000152
representing the heart rate signal extracted from the video,
Figure BDA0003571248220000153
representing the heart rate detected from the PPG, i.e. the true signal.
The mean absolute error of heart rate is defined as follows:
Figure BDA0003571248220000154
n represents the total number of test samples.
The root mean square error of the heart rate is defined as follows:
Figure BDA0003571248220000155
the accuracy of the heart rate is defined as follows:
Figure BDA0003571248220000156
the pearson correlation coefficient between the estimated value and the true value is as follows:
Figure BDA0003571248220000157
where μ denotes the average of the estimated heart rate,
Figure BDA0003571248220000161
is the true average.
(1) Result analysis on UBFC datasets by the invention
TABLE 1 Performance at different video lengths
Frame length MAE(bpm) RMSE(bpm) Acc(％) PCC
300 8.756 12.515 91.315 0.782
450 7.091 10.764 92.979 0.824
600 5.843 8.949 94.199 0.869
750 4.017 5.711 95.925 0.940
900 4.217 6.011 95.790 0.939
1200 3.467 4.071 96.526 0.976
All video segments have a frame rate of 30fps, and the video is divided into different lengths so as to realize heart rate signal recovery. The performance of the detection method at different video lengths is shown in table 1. Under the condition of 1200 frames, the detection method obtains the best result when the MAE and the RMSE are lower than 5bpm, the accuracy rate of the detection method is more than 96%, and the Pearson correlation between the detected heart rates is more than 97%. Below 450 frame lengths, the performance of the detection method drops dramatically. This is because short video in nature does not provide sufficiently stable information for heart rate signal recovery.
(2) Analysis of results on the PURE data set by the present invention
In the PURE data set, the performance of the present invention was tested in different states of the data set, including six different states. The results are shown in Table 2. The overall accuracy of the detection method is over 95%, and the Pearson correlation coefficient between the detected heart rate and the true value is over 98%. Under the states of steady state, speaking, slow translation, slow rotation and medium rotation, the Pearson correlation coefficient is more than 99 percent, which shows that the detection method has good performance to the states.
In the speaking state and the slow translation state, the detection method detects error fluctuation because facial skin wrinkles when a person speaks and motion artifacts are formed when the person slowly translates, and the factors interfere with the extraction of chrominance signals. In the fast pan state, there may be inaccurate ROI selection because face detection has difficulty capturing fast head rotations. Fast rotation causes blurring of the images in the video, thereby blurring the faint heart rate information to a greater extent.
The detection method in the present invention detects a high degree of correspondence between the heart rate and the ground true HR for all states of the entire data set.
TABLE 2 Performance of the Algorithm under different states
Evaluation criteria 01 02 03 04 05 06
MAE(bpm) 2.577 3.046 2.678 2.738 2.246 2.396
RMSE(bpm) 3.282 3.621 3.748 4.163 2.904 2.949
Acc(％) 95.951 95.060 96.708 95.891 96.629 96.365
PCC 0.996 0.991 0.991 0.985 0.996 0.994
01, stable state; 02, speaking; 03, slow translation; 04, fast translation; slow rotation is carried out at 05; and 06, medium-speed rotation.
(3) Experimental comparison of the invention
The present invention performs comparisons of other methods on UBFC datasets. Table 3 shows the experimental results of six different estimation methods on UBFC data sets. The MAE and RMSE of the method are lower than 5bpm, and the Acc and PCC are higher than 96 percent, which shows that the method has stronger applicability. Compared with the best POS method, the MAE is improved by 1.697bpm, the RMSE is improved by 6.075bpm, and the Pearson correlation coefficient is improved by 11.9%. The result shows that the method has higher detection method robustness under the conditions of stable illumination and heart rate variation.
DAOMP also uses sparse representation, and because its dictionary is simple, the correlation between the heart rates of the sub-regions is not considered, resulting in a large error between the reconstructed heart rate signal and the original pulse signal.
POS also gives good results as a subspace decomposition method.
TABLE 3 comparison of six comparison methods on UBFC data sets
Methods MAE(bpm) RMSE(bpm) Acc(％) PCC
ICA 28.455 36.858 72.420 0.277
CHROM 7.627 15.693 92.542 0.714
POS 5.164 10.146 94.872 0.857
LGI 10.570 20.715 89.841 0.553
DAOMP 14.580 22.804 85.850 0.484
Ours 3.467 4.071 96.526 0.976
Example 2
A sparse structure representation-based remote non-contact heart rate estimation system, comprising:
the subregion acquisition module is used for inputting a face video, fixedly selecting forehead and cheek parts as regions of interest of the detected face, and partitioning the regions of interest to acquire a plurality of subregions;
the high-quality quantum region chrominance signal generating module is used for constructing an average RGB pulse signal of each sub-region, performing motion compensation on the RGB pulse signal, calculating a chrominance signal S by utilizing linear combination of different channels of RGB, calculating the signal-to-noise ratio of each sub-region based on the chrominance signal S, taking the sub-region with the signal-to-noise ratio of the sub-region larger than the total signal-to-noise ratio as a high-quality quantum region, defining the corresponding chrominance signal as the high-quality chrominance signal and generating a high-quality quantum region chrominance signal M;
a heart rate signal reconstruction module for reconstructing a real heart rate signal MpulseSparse structure representation is carried out, and heart rate signals are reconstructed
Figure BDA0003571248220000181
The high-quality sub-region chrominance signal M consists of chrominance signals of sub-regions of high signal-to-noise ratio, i.e.
Figure BDA0003571248220000182
Wherein the content of the first and second substances,
Figure BDA0003571248220000183
a chrominance signal representing a selected high quality quantum region,
Figure BDA0003571248220000184
representing the reconstruction of the heart rate signal for selected high quality quantum regions, X is obtained as follows: by making use of differencesForming a mixed structure dictionary D by cosine basis of frequency and wavelet basis of different scales, solving a coefficient matrix A, selecting proper dictionary atoms from the mixed dictionary by a sparse structure method to represent heart rate signals, carrying out sparse coding by a greedy algorithm to form the coefficient matrix A, and reconstructing the heart rate signals by adopting an inner product method of the mixed structure dictionary D and the coefficient matrix A, namely
X＝D·A；
And the heart rate obtaining module is used for averaging the reconstructed heart rate signals of all the sub-regions and then carrying out power spectrum analysis to obtain the heart rate HR.
In the high-quality quantum region chrominance signal generation module:
calculating the average value, the absolute deviation of the average value and the standard deviation of the RGB pulse signals of each subarea, then calculating the deviation amplitude of each atom from the average value, and adding the standard deviation to the average value for the atoms exceeding the average value and exceeding the standard deviation; atoms below the mean and below the standard deviation are motion compensated with the mean minus the standard deviation, as defined in detail below:
Figure BDA0003571248220000191
Figure BDA0003571248220000192
Figure BDA0003571248220000193
Figure BDA0003571248220000194
wherein, l is the channel number, l 1, 2, 3 respectively represent RGB channel signals, and specific l 1 is an R channel; i is 2, which is a G channel; i-3 is B channel, l is signal length, Ci(t) the signal length of the i-th channel at time t, Ci_avgIs the signal average of the i channel, Ci_MeIs a general purpose ofMean deviation of the signal of the track, Ci_SDIs the signal standard deviation of the i channel;
the chrominance signal S is defined by S ═ Xf-αYfCalculated where α is the two signals XfAnd YfA ratio of standard deviation of (a) to (a) (X)f)/σ(Yf)，XsAnd YsIs the fusion of the normalized RGB pulse signals after motion compensation, XsAnd YsThe calculation function of (c) is as follows:
Xs＝3Rn-2Gn
Ys＝1.5Rn+Gn-1.5Bn
wherein R isn，GnAnd BnA color signal for each sub-region;
the signal-to-noise ratio function for each subregion is as follows:
Figure BDA0003571248220000201
wherein S isfRepresenting spectral values of the chrominance signal after S-Fourier transformation, f representing the frequency limiting the heart rate, the range of f being [0.6, 4 ]]，WtRepresenting the calculated sliding window size.
In the heart rate signal reconstruction module:
the method for obtaining X is as follows: using cosine basis DcSum wavelet DwA hybrid dictionary D is constructed which is,
Figure BDA0003571248220000202
Dw＝WaveDict(Short3，Nb，j，b)
wherein k isiFor the ith in a discrete cosine dictionarythFrequency of atoms, kiAnd k isi+1By a difference of
Figure BDA0003571248220000203
FrFor video frame rate, NbFor the length of the signal generated, j is the level vector, b is the translation factor, fLIn order to be able to generate the length of the signal sequence,
the mixed dictionary model is:
Figure BDA0003571248220000204
solving a coefficient matrix, selecting proper dictionary atoms from a mixed dictionary by using a sparse structure method to represent pulse signals, and carrying out sparse coding by using a greedy algorithm, wherein the step of solving the coefficient matrix is specifically to solve the following model:
Figure BDA0003571248220000205
obtaining sparse codes, forming a coefficient matrix A,
wherein λ is used to balance signal reconstruction accuracy
Figure BDA0003571248220000206
And a penalty term | | A | | non-woven phosphor1，2The regularization parameter of (a);
reconstructing the heart rate signal by adopting a method of inner product of a mixed dictionary D and a coefficient matrix A, wherein an inner product model is as follows:
X＝D·A；
averaging the reconstructed heart rate signals of all the sub-regions to obtain heart rate signals, which specifically comprises the following steps:
Figure BDA0003571248220000211
where N is the total number of selected sub-regions, xnIs the heart rate signal of the selected sub-area.
In the heart rate obtaining module:
performing power spectrum analysis on the averaged heart rate signal, selecting the component with the highest peak value as the main component of the heart rate signal, namely adopting the frequency f corresponding to the component with the highest powerHRFinally, the time-frequency conversion is carried out to the time domain to obtain the final heart rateHR, specifically: HR ═ fHR*60。
Example 3
A computer device comprising a memory storing a computer program and a processor implementing the steps of the sparse structure representation based remote non-contact heart rate estimation method when executing the computer program.

Claims (10)

1. The remote non-contact heart rate estimation method based on the sparse structure representation is characterized by comprising the following steps of:
step 1, inputting a face video, fixedly selecting forehead and cheek parts as regions of interest of a detected face, and partitioning to obtain a plurality of sub-regions;
step 2, constructing an average RGB pulse signal of each subregion, performing motion compensation on the RGB pulse signal, calculating a chrominance signal S by utilizing linear combination of different channels of RGB, calculating the signal-to-noise ratio of each subregion based on the chrominance signal S, taking the subregion of which the signal-to-noise ratio is greater than the total signal-to-noise ratio as a high-quality quantum region, defining the corresponding chrominance signal as a high-quality chrominance signal, and generating a high-quality quantum region chrominance signal M;
step 3, for real heart rate signal MpulseThe heart rate signal is reconstructed by carrying out sparse structure representation
Figure FDA0003571248210000011
The high-quality sub-region chrominance signal M consists of chrominance signals of sub-regions of high signal-to-noise ratio, i.e.
Figure FDA0003571248210000012
Wherein the content of the first and second substances,
Figure FDA0003571248210000013
a chrominance signal representing a selected high quality quantum region,
Figure FDA0003571248210000014
for selected high qualityAnd (3) reconstructing heart rate signals of the sub-areas, wherein the acquisition method of X is as follows: forming a mixed structure dictionary D by using cosine bases with different frequencies and wavelet bases with different scales, solving a coefficient matrix A, selecting a proper dictionary atom from the mixed dictionary by using a sparse structure method to represent a heart rate signal, carrying out sparse coding by using a greedy algorithm to form a coefficient matrix A, and reconstructing the heart rate signal by using an inner product method of the mixed structure dictionary D and the coefficient matrix A, wherein X is D.A;
and 4, averaging the reconstructed heart rate signals of all the sub-regions, and then analyzing the power spectrum to obtain the heart rate HR.
2. The sparse structure representation-based remote non-contact heart rate estimation method according to claim 1, wherein in step 2, the mean, the absolute deviation of the mean and the standard deviation of the RGB pulse signals of each sub-area are calculated, and then the magnitude of each atom deviating from the mean is calculated, and the atoms exceeding the mean and exceeding the standard deviation are added with the mean plus the standard deviation; atoms below the mean and below the standard deviation are motion compensated with the mean minus the standard deviation, as defined in detail below:
Figure FDA0003571248210000021
Figure FDA0003571248210000022
Figure FDA0003571248210000023
Figure FDA0003571248210000024
wherein i is the number of channels, i ═ 1, 2, 3 respectively represent RGB channel signals, and specifically i ═ 1 is the R channel; i ═2, is a G channel; i-3 is B channel, l is signal length, Ci(t) the signal length of the i-th channel at time t, Ci_avgIs the signal average of the i channel, Ci_MeIs the signal mean deviation of the i channel, Ci_SDIs the standard deviation of the signal for the i channel.
3. The sparse-structure-representation-based remote non-contact heart rate estimation method according to claim 1, wherein in step 2, the chrominance signal S is represented by S ═ Xf-αYfCalculated where α is two signals XfAnd YfA ratio of standard deviation of (a) to (a) (X)f)/σ(Yf)，XsAnd YsIs the fusion of the normalized RGB pulse signals after motion compensation, XsAnd YsThe calculation function of (a) is as follows:
Xs＝3Rn-2Gn
Ys＝1.5Rn+Gn-1.5Bn
wherein R isn，GnAnd BnA color signal for each sub-region;
the signal-to-noise ratio function for each subregion is as follows:
Figure FDA0003571248210000031
wherein S isfRepresenting spectral values of the chrominance signal after S-Fourier transformation, f representing the frequency limiting the heart rate, the range of f being [0.6, 4 ]]，WtRepresenting the calculated sliding window size.
4. The sparse structure representation-based remote non-contact heart rate estimation method according to claim 1, wherein in step 3, X is obtained as follows: using cosine basis DcSum wavelet DwA hybrid dictionary D is constructed which is,
Figure FDA0003571248210000032
Dw＝WaVeDict(Short3，Nb，j，b)
wherein k isiFor the ith in a discrete cosine dictionarythFrequency of atom, kiAnd k isi+1By a difference of
Figure FDA0003571248210000033
FrFor video frame rate, NbFor the length of the signal generated, j is the level vector, b is the translation factor, fLIn order to be able to generate the length of the signal sequence,
the mixed dictionary model is:
Figure FDA0003571248210000034
5. the sparse-structure-representation-based remote non-contact heart rate estimation method according to claim 4, wherein in step 3, a coefficient matrix is solved, a sparse-structure method is used for selecting a proper dictionary atom from a mixed dictionary to represent the pulse signal, a greedy algorithm is used for sparse coding, and the solving of the coefficient matrix is specifically to solve the following model:
Figure FDA0003571248210000041
obtaining sparse codes, forming a coefficient matrix A,
wherein λ is used to balance signal reconstruction accuracy
Figure FDA0003571248210000042
And a penalty term | | A | | non-woven phosphor1，2The regularization parameter of (1).
6. The sparse structure representation-based remote non-contact heart rate estimation method according to claim 5, wherein in step 3, the heart rate signal is reconstructed, and the heart rate signal is reconstructed by adopting a method of inner product of a mixed dictionary D and a coefficient matrix A, and an inner product model is as follows:
X＝D·A。
7. the sparse structure representation-based remote non-contact heart rate estimation method according to claim 6, wherein the heart rate signals of the reconstructed sub-regions are averaged to obtain heart rate signals, specifically:
Figure FDA0003571248210000043
where N is the total number of selected sub-regions, xnIs the heart rate signal of the selected sub-area.
8. The sparse-structure-representation-based remote non-contact heart rate estimation method according to claim 1, wherein in step 4, the power spectrum analysis is performed on the averaged heart rate signal, and the component with the highest peak value is selected as the main component of the heart rate signal, that is, the frequency f corresponding to the component with the highest power is adoptedHRAnd finally, converting the time frequency into the time domain to obtain the final heart rate HR, which specifically comprises the following steps: HR ═ fHR*60。
9. A system for remote non-contact heart rate estimation based on sparse structure representation, comprising:
the subregion acquisition module is used for inputting a face video, fixedly selecting forehead and cheek parts of a detected face as regions of interest, and partitioning the regions of interest to obtain a plurality of subregions;
the high-quality quantum region chrominance signal generating module is used for constructing an average RGB pulse signal of each sub-region, performing motion compensation on the RGB pulse signal, calculating a chrominance signal S by utilizing linear combination of different channels of RGB, calculating the signal-to-noise ratio of each sub-region based on the chrominance signal S, taking the sub-region with the signal-to-noise ratio of the sub-region larger than the total signal-to-noise ratio as a high-quality quantum region, defining the corresponding chrominance signal as the high-quality chrominance signal and generating a high-quality quantum region chrominance signal M;
a heart rate signal reconstruction module for reconstructing a real heart rate signal MpulseSparse structure representation is carried out, and heart rate signals are reconstructed
Figure FDA0003571248210000051
The high-quality sub-region chrominance signal M consists of chrominance signals of sub-regions of high signal-to-noise ratio, i.e.
Figure FDA0003571248210000052
Wherein the content of the first and second substances,
Figure FDA0003571248210000053
a chrominance signal representing a selected high quality quantum region,
Figure FDA0003571248210000054
representing the reconstruction of the heart rate signal for selected high quality quantum regions, X is obtained as follows: forming a mixed structure dictionary D by using cosine bases with different frequencies and wavelet bases with different scales, solving a coefficient matrix A, selecting a proper dictionary atom from the mixed dictionary by using a sparse structure method to represent a heart rate signal, carrying out sparse coding by using a greedy algorithm to form a coefficient matrix A, and reconstructing the heart rate signal by using an inner product method of the mixed structure dictionary D and the coefficient matrix A, wherein X is D.A;
and the heart rate obtaining module is used for averaging the reconstructed heart rate signals of all the sub-regions and then carrying out power spectrum analysis to obtain the heart rate HR.
10. A computer device, characterized by comprising a memory storing a computer program and a processor implementing the steps of the sparse structure representation based remote contactless heart rate estimation method of any of claims 1-8 when executing the computer program.
CN202210319966.6A 2022-03-29 2022-03-29 Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation Pending CN114596963A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210319966.6A CN114596963A (en) 2022-03-29 2022-03-29 Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210319966.6A CN114596963A (en) 2022-03-29 2022-03-29 Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation

Publications (1)

Publication Number Publication Date
CN114596963A true CN114596963A (en) 2022-06-07

Family

ID=81810209

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210319966.6A Pending CN114596963A (en) 2022-03-29 2022-03-29 Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation

Country Status (1)

Country Link
CN (1) CN114596963A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116269285A (en) * 2022-11-28 2023-06-23 电子科技大学 Non-contact normalized heart rate variability estimation system
TWI815614B (en) * 2022-08-17 2023-09-11 鉅怡智慧股份有限公司 Device and method of contactless physiological measurement with error compensation function

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI815614B (en) * 2022-08-17 2023-09-11 鉅怡智慧股份有限公司 Device and method of contactless physiological measurement with error compensation function
CN116269285A (en) * 2022-11-28 2023-06-23 电子科技大学 Non-contact normalized heart rate variability estimation system
CN116269285B (en) * 2022-11-28 2024-05-28 电子科技大学 Non-contact normalized heart rate variability estimation system

Similar Documents

Publication Publication Date Title
CN110269600B (en) Non-contact video heart rate detection method based on multivariate empirical mode decomposition and combined blind source separation
Macwan et al. Heart rate estimation using remote photoplethysmography with multi-objective optimization
CN114596963A (en) Remote non-contact heart rate estimation method, system and equipment based on sparse structure representation
Casado et al. Face2PPG: An unsupervised pipeline for blood volume pulse extraction from faces
CN111243739A (en) Anti-interference physiological parameter telemetering method and system
Liu et al. Motion-robust multimodal heart rate estimation using BCG fused remote-PPG with deep facial ROI tracker and pose constrained Kalman filter
Demirezen et al. Heart rate estimation from facial videos using nonlinear mode decomposition and improved consistency check
Macwan et al. Remote photoplethysmography measurement using constrained ica
Yin et al. PulseNet: A multitask learning network for remote heart rate estimation
Yue et al. Non-contact heart rate detection by combining empirical mode decomposition and permutation entropy under non-cooperative face shake
Ryu et al. A measurement of illumination variation-resistant noncontact heart rate based on the combination of singular spectrum analysis and sub-band method
CN113378650B (en) Emotion recognition method based on electroencephalogram source imaging and regularization common space mode
CN114387479A (en) Non-contact heart rate measurement method and system based on face video
Wu et al. Echocardiogram enhancement using supervised manifold denoising
Das et al. Time–frequency learning framework for rPPG signal estimation using Scalogram-based feature map of facial video data
CN115153473B (en) Non-contact heart rate detection method based on multivariate singular spectrum analysis
CN115813409A (en) Ultra-low-delay moving image electroencephalogram decoding method
CN114246570B (en) Near-infrared heart rate detection method by fusing peak signal-to-noise ratio and Peerson correlation coefficient
Kuang et al. Remote photoplethysmography signals enhancement based on generative adversarial networks
CN114580464A (en) Human heart rate variability and respiratory rate measurement method based on variational modal decomposition and constraint independent component analysis
Emara et al. A Hybrid Compressive Sensing and Classification Approach for Dynamic Storage Management of Vital Biomedical Signals
Li A temporal encoder-decoder approach to extracting blood volume pulse signal morphology from face videos
Ma et al. Cdcca-rppgformer: Transformer-like network based on 3d-cdc-st and eca for remote heart rate measurement
Ou et al. Non-contact heart rate measurement from face video based on group sparse representation
Le et al. Remote PPG Estimation from RGB-NIR Facial Image Sequence for Heart Rate Estimation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination