A non-reference evaluation method for pulse wave signal quality based on heart rate continuity

Abstract

The application discloses a no-reference evaluation method for pulse wave signal quality based on heart rate continuity, and steps of the method comprise the following steps: firstly, acquiring video images and defining a plurality of face regions of interest, then acquiring corresponding pulse wave signals by using a denoising algorithm; then, converting the pulse wave signals into a time-frequency domain by using synchronous compression wavelet transform, extracting time-frequency domain instantaneous heart rate variance features calculated based on the synchronous compression wavelet transform, sorting and screening the quality of the pulse wave signals, fusing time-frequency spectrums of selected high-quality pulse wave signals, extracting a time-frequency ridge on the fused time-frequency spectrum, estimating heart rate again by using a heart rate continuity assumption, and finally calculating uncertainty of the time-frequency spectrum of the high-quality pulse wave signals to evaluate the reliability of predicted heart rate. The application can effectively evaluate, screen and fuse the quality of pulse wave signals obtained by each sub-region, thereby realizing robust extraction of video heart rate.

Description

Technical Field

[0001] This invention belongs to the technical field of biomedical signal processing, and in particular relates to a referenceless evaluation method for pulse wave signal quality based on heart rate continuity. Background Technology

[0002] Heart rate is one of the important vital signs reflecting human health. Traditional heart rate measurement requires additional contact devices or sensors, such as electrocardiograms (ECG) or photoplethysmography (PPG), which increases costs and causes inconvenience to subjects. Recently, remote photoplethysmography (rPPG) technology has attracted attention. It utilizes the minute color changes caused by blood flow in the skin during physiological activities such as heartbeat and respiration to achieve non-contact measurement of heart rate.

[0003] In rPPG technology, the extremely weak video pulse wave signal presents several challenges. First, changes in ambient lighting or facial movements introduce noise far exceeding the target signal, severely impacting the stability of rPPG measurements. Second, many methods fail to adequately consider the physiological characteristics of facial pulse wave signal measurement; due to variations in capillary distribution beneath the skin surface, the intensity of the pulse wave signal differs across different facial skin regions, leading to significant variations in denoising performance across these regions. Finally, the inherent stability of the denoising algorithm itself or differences in the region of interest directly or indirectly affect the stability of pulse wave signal extraction, making robust heart rate detection results difficult to obtain. Therefore, accurately assessing the quality of the pulse wave signal in different regions of interest, filtering and fusing it, and evaluating the reliability of predicted heart rate are crucial for obtaining robust rPPG measurement results. Summary of the Invention

[0004] To address the shortcomings of the aforementioned technologies, this invention provides a referenceless evaluation method for pulse wave signal quality based on heart rate continuity. This method aims to effectively evaluate, screen, and fuse the quality of pulse wave signals, and assess the reliability of predicted heart rate, thereby enabling robust extraction of video heart rate.

[0005] The present invention adopts the following solution to solve the technical problem:

[0006] The present invention provides a referenceless evaluation method for pulse wave signal quality based on heart rate continuity, characterized by the following steps:

[0007] Step 1: Obtain the subject's T totalThe video data is divided into L parts, and each part of the video data has t. total Frame video image;

[0008] Step 2: Use a face feature point detection algorithm to analyze the t-th video data. total Frame-by-frame detection is performed on the video images to remove non-skin regions in each frame, thereby obtaining I facial regions of interest in each frame of the l-th video data, where 1≤l≤L;

[0009] Step 3: Calculate t for the l-th video data. total The average pixel values ​​of the R, G, and B color channels of the i-th facial region of interest in a frame of video image are used to obtain t of the l-th video data. total The color channel signal corresponding to the i-th facial region of interest in a frame of video image. in, t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the R color channel of the i-th facial region of interest in a frame of video image. t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the G color channel of the i-th facial region of interest in a frame of video image. t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the B color channel of the i-th facial region of interest in a frame video image, 1≤i≤I;

[0010] Step 4: Use a denoising algorithm to process the t-th video data. total R signal of the i-th facial region of interest in a frame video image G signal B signal Processing is performed to obtain the t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l，i ;

[0011] Step 5: Extract the instantaneous heart rate variance features in the time-frequency domain calculated based on synchronous compressed wavelet transform:

[0012] Step 5.1: Use synchronous compressed wavelet transform to transform the original pulse wave signal V l，i Transform to the time-frequency domain and obtain t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l，i The corresponding time spectrum S l，i ;

[0013] Step 5.2: For S l，iWindowing is used to obtain the l-th video data t. total The temporal spectrum of H sub-windows of the i-th facial region of interest in a frame video image in, t represents the l-th video data total The temporal spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image;

[0014] Step 5.3: From Extracting the l-th video data t total The time-frequency ridge with the highest energy (K) in the time spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image. in, t represents the l-th video data total The k-th highest energy time-frequency ridge in the time spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image; t of the l-th video data total The k-th highest energy time-frequency ridge in the time spectrum of the first sub-window of the i-th facial region of interest in a frame video image. Starting from the h-th sub-window, the time-frequency ridge with the k-th highest energy in the time spectrum is... and Connect them to obtain the minimum variance of the first to h sub-windows, and connect them to form the k-th minimum variance ridge of the first h sub-windows. This yields the K minimum variance ridges for the first h sub-windows. This leads to the K minimum variance ridges within the first H sub-windows. calculate variance in, express The variance, from The minimum value is selected as t for the l-th video data. total VAR of the time-frequency spectrum of the raw pulse wave signal of the i-th facial region of interest in a frame video image l，i Thus, the t of the l-th video data is obtained. total The variance of the time spectrum of the raw pulse wave signal of I facial regions of interest in a frame video image, {VAR} l，i} 1≤i≤I ;

[0015] Step 6: Extract frequency domain signal-to-noise ratio features calculated based on Fast Fourier Transform:

[0016] Step 6.1: Use Fast Fourier Transform to convert the original pulse wave signal V l，i Convert to the frequency domain and obtain t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video imagel，i The spectrum Q l，i ;

[0017] Step 6.2: Calculate the spectrum Q l，i main frequency q l，i , will the spectrum Q l，i The curve is in q l，i The area within a frequency band with a center line and a bandwidth of w is denoted as e. l，i Spectrum Q l，i The area of ​​the curve over the remaining frequency band is denoted as n. l，i ; Calculate t of the l-th video data total The signal-to-noise ratio (SNR) of the raw pulse wave signal of the i-th facial region of interest in a frame of video image. l，i For e l，i and n l，i The ratio of the values; thus obtaining t of the l-th video data. total The signal-to-noise ratio (SNR) of the raw pulse wave signals of I facial regions of interest in a frame of video image. l，i} 1≤i≤I ;

[0018] Step 7: Transfer {VAR} l，i} 1≤i≤I Sort the variances in ascending order and obtain the sorted variances. Record the position index of the unsorted variances within the sorted variances in the time-frequency domain quality index set {VARrank}. l，i} 1≤i≤I In the middle, among them, VARrank l，i V represents l，i The time-frequency domain quality index;

[0019] {SNR l，i} 1≤i≤I Sort the signals in descending order to obtain the sorted signal-to-noise ratios (SNRs), and record the position index of the unsorted SNR within the sorted SNR in the frequency domain quality index set {SNRrank}. l，i} 1≤i≤I In the middle, SNRrank l，i V represents l，i The frequency domain quality index;

[0020] When VARrank l，i and SNRrank l，i When both are less than Th1, it means V l，i For high-quality pulse wave signals, where Th1 represents the threshold;

[0021] Step 8: Transfer the t data of the lth video data totalThe signal-to-noise ratio (SNR) of all high-quality pulse wave signals in the frame video image is used as the weight. The time-frequency spectra of all high-quality pulse wave signals are then weighted and averaged to obtain the fused average time-frequency spectrum SF. l ;

[0022] Step 9: Follow the procedure in step 5.2 to process SF. l Windowing is performed to obtain the t of the l-th video data. total Average temporal spectrum of H sub-windows in a frame video image in, t represents the l-th video data total The average temporal spectrum of the h-th sub-window in a frame video image;

[0023] Step 10: Follow the procedure in step 5.3. Processing is performed to obtain the t of the l-th video data. total The time-frequency ridge with the highest energy among the K highest values ​​in the average time-spectrum of the h-th sub-window of a frame video image. in, t represents the l-th video data total The time-frequency ridge with the highest energy in the average time-frequency spectrum of the h-th sub-window in a frame video image is used to obtain the K minimum variance ridges within the first H sub-windows. in, This represents the k-th highest energy time-frequency ridge in the time-frequency spectrum of the first sub-window. Starting from the point, connect the k-th minimum variance ridge within H sub-windows;

[0024] calculate variance in, express The variance;

[0025] Select The minimum variance ridge corresponding to the minimum value Minimum variance ridge The median heart rate (hr) of the l-th video data was used as the heart rate frequency. l Thus, the heart rate value (HR) of the lth video data is obtained. l ;

[0026] Step 11: Calculate t for the l-th video data total The uncertainty of the time spectrum corresponding to all high-quality pulse wave signals of the frame video image is obtained as the uncertainty time spectrum Sun. l To evaluate HR l Reliability.

[0027] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the pulse wave signal quality no-reference evaluation method, and the processor is configured to execute the program stored in the memory.

[0028] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the steps of the pulse wave signal quality no-reference evaluation method.

[0029] Compared with existing technologies, the beneficial effects of this invention are reflected in:

[0030] 1. This invention employs synchronous compressed wavelet transform to convert pulse wave signals to the time-frequency domain, and performs effective quality evaluation on the pulse wave signals acquired in each sub-region based on the short-term heart rate continuity characteristics in the time-frequency domain. Compared to other signal quality evaluation criteria, the instantaneous heart rate variance characteristics in the time-frequency domain calculated based on synchronous compressed wavelet transform can comprehensively consider the time-domain, frequency-domain, and spatial-domain information of the pulse wave signal, thereby guiding the identification of distorted signals.

[0031] 2. This invention divides the face into multiple regions of interest (ROIs). Since the pulse wave signal quality may vary in different regions, the optimal ROI is selected through a no-reference evaluation of signal quality. A multi-region fusion method is then used to extract the heart rate component. The uncertainty of the time-spectrum of the high-quality pulse wave signal extracted from the optimal ROI is then calculated, thereby providing a no-reference assessment of the reliability of the predicted heart rate and achieving robust extraction of video heart rate. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method of the present invention;

[0033] Figure 2 This is a schematic diagram illustrating the location of the region of interest (ROI) on a face in a video frame according to the present invention.

[0034] Figure 3 This is a schematic diagram of the signal time-frequency domain feature extraction of the present invention;

[0035] Figure 4 This is a schematic diagram illustrating the location of the optimal region of interest (ROI) on a face within a video frame according to the present invention.

[0036] Figure 5 This is the average time-frequency spectrum of the high-quality pulse wave collection fusion of the present invention;

[0037] Figure 6 This is a time-spectral uncertainty diagram of the high-quality pulse wave set of the present invention. Detailed Implementation

[0038] In this embodiment, a no-reference evaluation method for pulse wave signal quality based on heart rate continuity is described. First, the input video segment is analyzed frame-by-frame, dividing the face into multiple regions of interest. Pixel averaging is performed on the R, G, and B color channels of each region. The POS algorithm is used to denoise the RGB signal to extract the original pulse wave signal. Synchronous compressed wavelet transform and fast Fourier transform are used to convert the pulse wave signal to the time-frequency domain and frequency domain, respectively. Three time-frequency ridges are extracted from the time-frequency spectrum, and the instantaneous heart rate variance feature corresponding to each ridge is calculated. Based on the assumption of instantaneous heart rate continuity, the ridge with the minimum variance is selected as the candidate heart rate ridge. Then, the pulse wave signals obtained from all regions are ranked according to their minimum instantaneous heart rate variance feature and signal-to-noise ratio feature. The time-frequency spectra of the selected high-quality pulse wave signals are fused, and ridges containing heart rate components are extracted from the average time-frequency spectrum. The heart rate is estimated again using the heart rate continuity assumption. Finally, the uncertainty of the time-frequency spectrum of the high-quality pulse wave signal is calculated, and the reliability of the predicted heart rate is evaluated using the time-frequency spectrum of the uncertainty. Specifically, as shown... Figure 1 As shown, the method is performed according to the following steps:

[0039] Step 1: Obtain the subject's T total The video data is divided into L parts, and each part of the video data has t. total Frame video image;

[0040] Step 2: Use the Mediapipe face landmark detection algorithm to analyze the l-th video data. total Frame-by-frame detection is performed on the video images, and the non-skin regions of each frame are removed using the Convex Hull algorithm, such as... Figure 2 As shown, I facial regions of interest are obtained for each frame of video image in the l-th video data, where 1 ≤ l ≤ L. In this embodiment, I = 70.

[0041] Step 3: Calculate t for the l-th video data. total The average pixel values ​​of the R, G, and B color channels of the i-th facial region of interest in a frame of video image are used to obtain t of the l-th video data. total The color channel signal corresponding to the i-th facial region of interest in a frame of video image. in, t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the R color channel of the i-th facial region of interest in a frame of video image. t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the G color channel of the i-th facial region of interest in a frame of video image. t represents the l-th video data totalThe signal obtained by averaging all pixel values ​​in the B color channel of the i-th facial region of interest in a frame video image, where 1 ≤ i ≤ I.

[0042] Step 4: Use the POS algorithm to process the t-th video data. total R signal of the i-th facial region of interest in a frame video image G signal B signal Processing is performed to obtain the t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l，i .

[0043] Step 5: Extract the instantaneous heart rate variance features in the time-frequency domain calculated based on synchronous compressed wavelet transform, such as... Figure 3 As shown:

[0044] Step 5.1: Use synchronous compressed wavelet transform to transform the original pulse wave signal V l，i Transform to the time-frequency domain and obtain t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l，i The corresponding time spectrum S l，i ;

[0045] Step 5.2: For S l，i Windowing is used to obtain the l-th video data t. total The temporal spectrum of H sub-windows of the i-th facial region of interest in a frame video image in, t represents the l-th video data total The temporal spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image.

[0046] Step 5.3: Based on the short-term continuity of heart rate, the frequency component corresponding to heart rate should be a smooth and continuous curve in the time spectrum. However, the frequency components of noise are relatively more complex. Therefore, variance is used to evaluate the signal quality. Extracting the l-th video data t total The time-frequency ridge with the highest energy (K) in the time spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image. in, t represents the l-th video data total The k-th highest energy time-frequency ridge in the time spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image; t of the l-th video data total The k-th highest energy time-frequency ridge in the time spectrum of the first sub-window of the i-th facial region of interest in a frame video image. Starting from the h-th sub-window, the time-frequency ridge with the k-th highest energy in the time spectrum is... and Connect them to obtain the minimum variance of the first to h sub-windows, and connect them to form the k-th minimum variance ridge of the first h sub-windows. This yields the K minimum variance ridges for the first h sub-windows. This leads to the K minimum variance ridges within the first H sub-windows. calculate variance in, express The variance;

[0047] from The minimum value is selected as t for the l-th video data. total VAR of the time-frequency spectrum of the raw pulse wave signal of the i-th facial region of interest in a frame video image l，i Thus, the t of the l-th video data is obtained. total The variance of the time spectrum of the raw pulse wave signal of I facial regions of interest in a frame video image, {VAR} l，i} 1≤i≤I In this embodiment, K = 3.

[0048] Step 6: Extract frequency domain signal-to-noise ratio features calculated based on Fast Fourier Transform:

[0049] Step 6.1: Use Fast Fourier Transform to convert the original pulse wave signal V l，i Convert to the frequency domain and obtain t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l，i The spectrum Q l，i ;

[0050] Step 6.2: Based on the quasi-periodic characteristics of pulse wave signals, high-quality signals should have clear main peaks and harmonics in their spectrum. When the signal is distorted, the main peaks and harmonics will be weakened by the flat spectrum. Therefore, the signal-to-noise ratio is used to evaluate the signal quality, and the spectrum Q is calculated. l，i main frequency q l，i , will the spectrum Q l，i The curve is in q l，i The area within a frequency band with a center line and a bandwidth of w is denoted as e. l，i Spectrum Q l，i The area of ​​the curve over the remaining frequency band is denoted as n. l，i ;

[0051] Calculate t of the l-th video data total The signal-to-noise ratio (SNR) of the raw pulse wave signal of the i-th facial region of interest in a frame of video image.l，i For e l，i and n l，i The ratio of the values; thus obtaining t of the l-th video data. total The signal-to-noise ratio (SNR) of the raw pulse wave signals of I facial regions of interest in a frame of video image. l，i} 1≤i≤I .

[0052] Step 7: Transfer {VAR} l，i} 1≤i≤I Sort the variances in ascending order and obtain the sorted variances. Record the position index of the unsorted variances within the sorted variances in the time-frequency domain quality index set {VARrank}. l，i} 1≤i≤I In the middle, among them, VARrank l，i V represents l，i The time-frequency domain quality index;

[0053] {SNR l，i} 1≤i≤I Sort the signals in descending order to obtain the sorted signal-to-noise ratios (SNRs), and record the position index of the unsorted SNR within the sorted SNR in the frequency domain quality index set {SNRrank}. l，i} 1≤i≤I Among them, SNRrank l，i V represents l，i The frequency domain quality index;

[0054] When VARrank l，i and SNRrank l，i When both are less than Th1, it means V l，i For high-quality pulse wave signals, the corresponding region of interest will be considered the optimal region of interest, such as... Figure 4 As shown, Th1 represents the threshold, and in this embodiment, Th1 = 50.

[0055] Step 8: Transfer the t data of the lth video data total The signal-to-noise ratio (SNR) of all high-quality pulse wave signals in the frame video image is used as the weight. The time-frequency spectra of all high-quality pulse wave signals are then weighted and averaged to obtain the fused average time-frequency spectrum SF. l ,like Figure 5 As shown.

[0056] Step 9: Follow the procedure in step 5.2 to process SF. l Windowing is performed to obtain the t of the l-th video data. total Average temporal spectrum of H sub-windows in a frame video image in, t represents the l-th video data totalThe average temporal spectrum of the h-th sub-window in a frame video image.

[0057] Step 10: Follow the procedure in step 5.3. Processing is performed to obtain the t of the l-th video data. total The time-frequency ridge with the highest energy among the K highest values ​​in the average time-spectrum of the h-th sub-window of a frame video image. in, t represents the l-th video data total The time-frequency ridge with the highest energy in the average time-frequency spectrum of the h-th sub-window in a frame video image is used to obtain the K minimum variance ridges within the first H sub-windows. in, This represents the k-th highest energy time-frequency ridge in the time-frequency spectrum of the first sub-window. Starting from the point, connect the k-th minimum variance ridge within H sub-windows;

[0058] calculate variance in, express The variance;

[0059] Select The minimum variance ridge corresponding to the minimum value Minimum variance ridge The median heart rate (hr) of the l-th video data was used as the heart rate frequency. l Thus, the heart rate value (HR) of the lth video data is obtained. l .

[0060] Step 11: Calculate t for the l-th video data total The uncertainty of the time spectrum corresponding to all high-quality pulse wave signals of the frame video image is obtained as the uncertainty time spectrum Sun. l ,like Figure 6 As shown, to evaluate HR l Reliability.

[0061] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0062] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A referenceless evaluation method for pulse wave signal quality based on heart rate continuity, characterized in that, The procedure is as follows: Step 1: Obtain the subject's T total The video data is divided into L parts, and each part of the video data has t. total Frame video image; Step 2: Use a face feature point detection algorithm to analyze the t-th video data. total Frame-by-frame detection is performed on the video images to remove non-skin regions in each frame, thereby obtaining I facial regions of interest in each frame of the l-th video data, where 1≤l≤L; Step 3: Calculate t for the l-th video data. total The average pixel values ​​of the R, G, and B color channels of the i-th facial region of interest in a frame of video image are used to obtain t of the l-th video data. total The color channel signal corresponding to the i-th facial region of interest in a frame of video image. in, t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the R color channel of the i-th facial region of interest in a frame of video image. t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the G color channel of the i-th facial region of interest in a frame of video image. t represents the l-th video data total The signal obtained by averaging all pixel values ​​in the B color channel of the i-th facial region of interest in a frame video image, 1≤i≤I; Step 4: Use a denoising algorithm to process the t-th video data. total R signal of the i-th facial region of interest in a frame video image G signal B signal Processing is performed to obtain the t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l,i ; Step 5: Extract the instantaneous heart rate variance features in the time-frequency domain calculated based on synchronous compressed wavelet transform: Step 5.1: Use synchronous compressed wavelet transform to transform the original pulse wave signal V l,i Transform to the time-frequency domain and obtain t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l,i The corresponding time spectrum S l,i ; Step 5.2: For S l,i Windowing is used to obtain the l-th video data t. total The temporal spectrum of H sub-windows of the i-th facial region of interest in a frame video image in, t represents the l-th video data total The temporal spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image; Step 5.3: From Extracting the l-th video data t total The time-frequency ridge with the highest energy (K) in the time spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image. in, t represents the l-th video data total The k-th highest energy time-frequency ridge in the time spectrum of the h-th sub-window of the i-th facial region of interest in a frame video image; t of the l-th video data total The k-th highest energy time-frequency ridge in the time spectrum of the first sub-window of the i-th facial region of interest in a frame video image. Starting from the h-th sub-window, the time-frequency ridge with the k-th highest energy in the time spectrum is... and Connect them to obtain the minimum variance of the first to h sub-windows, and connect them to form the k-th minimum variance ridge of the first h sub-windows. This yields the K minimum variance ridges for the first h sub-windows. This leads to the K minimum variance ridges within the first H sub-windows. calculate variance in, express The variance, from The minimum value is selected as t for the l-th video data. total VAR of the time-frequency spectrum of the raw pulse wave signal of the i-th facial region of interest in a frame video image l,i Thus, the t of the l-th video data is obtained. total The variance of the time spectrum of the raw pulse wave signal of I facial regions of interest in a frame video image, {VAR} l,i } 1≤i≤I ; Step 6: Extract frequency domain signal-to-noise ratio features calculated based on Fast Fourier Transform: Step 6.1: Use Fast Fourier Transform to convert the original pulse wave signal V l,i Convert to the frequency domain and obtain t of the l-th video data. total The raw pulse wave signal V of the i-th facial region of interest in the frame video image l,i The spectrum Q l,i ; Step 6.2: Calculate the spectrum Q l,i main frequency q l,i , will the spectrum Q l,i The curve is in q l,i The area within a frequency band with a center line and a bandwidth of w is denoted as e. l,i Spectrum Q l,i The area of ​​the curve over the remaining frequency band is denoted as n. l,i ; Calculate t of the l-th video data total The signal-to-noise ratio (SNR) of the raw pulse wave signal of the i-th facial region of interest in a frame of video image. l,i For e l,i and n l,i The ratio of the values; thus obtaining t of the l-th video data. total The signal-to-noise ratio (SNR) of the raw pulse wave signals of I facial regions of interest in a frame of video image. l,i } 1≤i≤I ; Step 7: Transfer {VAR} l,i } 1≤i≤I Sort the variances in ascending order and obtain the sorted variances. Record the position index of the unsorted variances within the sorted variances in the time-frequency domain quality index set {VARank}. l,i } 1≤i≤I In the middle, among them, VARrank l,i V represents l,i The time-frequency domain quality index; {SNR l,i } 1≤i≤I Sort the signals in descending order to obtain the sorted signal-to-noise ratios (SNRs), and record the position index of the unsorted SNR within the sorted SNR in the frequency domain quality index set {SNRrank}. l,i } 1≤i≤I In the middle, SNRrank l,i V represents l,i The frequency domain quality index; When VARrank l,i and SNRranh l,i When both are less than Th1, it means V l,i For high-quality pulse wave signals, where Th1 represents the threshold; Step 8: Transfer the t data of the lth video data total The signal-to-noise ratio (SNR) of all high-quality pulse wave signals in the frame video image is used as the weight. The time-frequency spectra of all high-quality pulse wave signals are then weighted and averaged to obtain the fused average time-frequency spectrum SF. l ; Step 9: Follow the procedure in step 5.2 to process SF. l Windowing is performed to obtain the t of the l-th video data. total Average temporal spectrum of H sub-windows in a frame video image in, t represents the l-th video data total The average temporal spectrum of the h-th sub-window in a frame video image; Step 10: Follow the procedure in step 5.

3. Processing is performed to obtain the t of the l-th video data. total The time-frequency ridge with the highest energy among the K highest values ​​in the average time-spectrum of the h-th sub-window of a frame video image. in, t represents the l-th video data total The time-frequency ridge with the highest energy in the average time-frequency spectrum of the h-th sub-window in a frame video image is used to obtain the K minimum variance ridges within the first H sub-windows. in, This represents the k-th highest energy time-frequency ridge in the time-frequency spectrum of the first sub-window. Starting from the point, connect the k-th minimum variance ridge within H sub-windows; calculate variance in, express The variance; Select The minimum variance ridge corresponding to the minimum value Minimum variance ridge The median heart rate (hr) of the l-th video data was used as the heart rate frequency. l Thus, the heart rate value (HR) of the lth video data is obtained. l ; Step 11: Calculate t for the l-th video data total The uncertainty of the time spectrum corresponding to all high-quality pulse wave signals of the frame video image is obtained as the uncertainty time spectrum Sun. l To evaluate HR l Reliability.

2. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the pulse wave signal quality no-reference evaluation method of claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium storing a computer program, characterized in that, The computer program, when run by the processor, executes the steps of the pulse wave signal quality no-reference evaluation method of claim 1.

Images