Motion robust non-contact heart rate detection method, electronic device and storage medium

Abstract

The application discloses a motion-robust non-contact heart rate detection method, an electronic device and a storage medium. The method comprises the following steps: firstly, extracting a pixel mean time sequence from a selected video face region of interest, including the selection and positioning of the region of interest, tracking the region of interest and extracting the pixel mean time sequence; secondly, removing non-rigid motion or rigid motion, including signal preprocessing, time delay, conversion from a single channel to a multi-channel, extraction of a bottom layer shared source, selection of a motion noise signal and a pulse signal; finally, heart rate estimation, Fourier transform of the pulse signal and corresponding operation to obtain a heart rate value. The application extracts a heart rate related signal from a video image, removes motion noise by using a denoising algorithm, detects the heart rate in a motion state, provides a new direction for non-contact heart rate detection research, and further promotes the application of the non-contact detection technology in a life scene.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of video-based non-contact physiological parameter detection, in particular to a motion-robust non-contact video heart rate detection method combining chrominance model and time delay joint blind source separation (CHROM-TDJBSS), an electronic device and a storage medium. BACKGROUND

[0002] Currently, heart rate detection is mainly divided into two categories: contact and non-contact. Traditional contact heart rate measurement devices such as electrocardiographs and pulse oximeters usually require patients to wear electrodes or finger clips, which can cause discomfort to individuals. Non-contact remote photoplethysmography (rPPG) detects changes in blood volume in facial capillaries accompanying physiological activities of heartbeats in video images to extract blood volume pulse signals that can reflect heart rate information. Due to the widespread use of cameras, low cost and the absence of the need to wear any equipment, rPPG is widely used, usually in the fields of newborn health detection, home health monitoring and motion monitoring.

[0003] The amplitude of the blood volume pulse signal contained in rPPG is very weak, and in the motion monitoring scene, part of the motion signal itself has periodicity, and the motion amplitude is much larger than that of the blood volume pulse (BVP). Therefore, remote photoplethysmography is easily disturbed by motion noise, causing a major challenge to the accuracy of heart rate measurement. Studies have shown that the green channel contains strong pulsatile information, so single-channel signal decomposition processing technology can be used to improve the quality of the blood volume pulse signal. Since motion will cause changes in the distance or angle between the face and the camera, motion denoising can be modeled as an optical model, and from this, chrominance model and plane orthogonal skin reflectance model are proposed to solve the motion denoising problem. In addition, source separation methods can be used to determine the component containing pulse information, such as blind signal separation methods. The most representative blind signal separation methods are independent component analysis and principal component analysis. However, in some intense motion scenarios, the motion amplitude is too high to meet the prerequisite condition of independent component analysis linear combination. With the development of artificial intelligence technology, deep learning has also been used in remote photoplethysmography. In addition, existing research has also proposed combining different types of methods to improve the quality of the pulse signal.

[0004] The accuracy of heart rate detection can decrease with the intense exercise of the subject, especially when the subject's movements are intense and periodic. On the one hand, the movements can cause the relative displacement of the position and angle between the face region of interest and the camera, and the loss of tracking of the region of interest in the case of too high movement frequency; on the other hand, strong periodic movements can occur in the process of fitness, causing the main frequency of the movement component to fall within the heart rate frequency band of interest and thus interfere with the heart rate estimation, as shown in Figure 1 . Specifically, as shown in Figure 1 (a), the rowing machine movement belongs to forward and backward movement, and the main frequency of the movement of most subjects is only 0.6 Hz, which is lower than the main frequency of the elliptical machine and the stationary bicycle movement, but the movement amplitude is larger. As shown in Figure 1 (b), the elliptical machine movement has a movement frequency of about 2 Hz, and Figure 1 (c) shows the stationary bicycle movement, which has a main frequency of generally 1 Hz, 2 Hz or 2.5 Hz. As can be seen from Figure 1 , in some cases with large movement amplitude, although the movement frequency is low, the multiple frequency is still within the HR frequency band of interest, so that the heart rate main frequency is lost, making it difficult to extract the heart rate. SUMMARY

[0005] The present application is proposed to solve the above-mentioned problems of the prior art, and provides a motion-robust non-contact heart rate detection method, an electronic device and a storage medium, so as to improve the accuracy of video heart rate detection in the scene of intense and periodic movement, thereby providing a new direction for the research of non-contact heart rate detection, and further promoting the application of non-contact detection technology in life scenes.

[0006] To achieve the above-mentioned application purposes, the present application adopts the following technical solutions:

[0007] The motion-robust non-contact heart rate detection method of the present application has the following steps:

[0008] Step 1, extracting the pixel mean time series of the face region of interest from the I-frame video image of the subject, wherein the pixel mean time series contains information related to heart rate and information related to movement;

[0009] Step 1.1, positioning and tracking of the region of interest;

[0010] The face feature point detection and tracking method is used to detect and track the face feature points of the I-frame video image, so as to locate the circumscribed rectangle of the M regions of interest in each video image, and let the width and height of the circumscribed rectangle be w m , h m ; M≥2;

[0011] Step 1.2, extraction of the time series of the pixel mean value of the region of interest;

[0012] The time series of the pixel mean value of the mth region of interest is calculated frame by frame, and the time series of the RGB channel of the I-frame video image is obtained, including the time series of the red channel in the I-frame video image The time series of the green channel in the I-frame video image and the time series of the blue channel in the I-frame video image wherein, represents the red channel pixel mean value of the ith frame in the mth region of interest; represents the green channel pixel mean value of the ith frame in the mth region of interest; represents the blue channel pixel mean value of the ith frame in the mth region of interest;

[0013] Step two, removal of motion noise under different motion scenes;

[0014] Step 2.1, based on the skin reflection model, the RGB channel time series is preprocessed into the single-channel color difference signal C of the mth region of interest using formula (1) [m] ,1≤m≤M:

[0015]

[0016] In formula (1), and are the normalized red channel pixel mean value, green channel pixel mean value and blue channel pixel mean value of the ith frame after band-pass filtering in the mth region of interest; represents the chroma signal defined by the projection of the red channel time series and the green channel time series, represents the chroma signal defined by the projection of the red, green and blue channel time series, and α represents a tuning parameter for separating light intensity changes and pulse signals, and σ(·) represents the standard deviation of a given signal;

[0017] Step 2.2, according to the type of motion scene, set the detrending parameter λ, and use it for detrending processing on the single-channel color difference signal S of the mth region of interest [m] to obtain the detrended single-channel color difference signal S of the mth region of interest wherein, represents the detrended single-channel color difference signal of the ith frame in the mth region of interest;

[0018] Step 2.3, the detrended single-channel color difference signal S of the mth region of interest is delayed to obtain a channel data set and as the delayed mth multi-channel data set wherein, represents the ith column vector of the mth region of interest, and 1≤m≤M; T represents transposition;

[0019] Step 2.4, decompose the multi-channel data set matrix X [m] using the independent vector analysis method in the joint blind source separation algorithm;

[0020] Step 2.4.1, regard each multi-channel data set as a linear mixture of L independent sources respectively, and construct the relationship between the multi-channel data set and the L independent sources using formula (2):

[0021] X [m] = W [m] S [m] , 1≤m≤M (2)

[0022] In formula (2), W [m] is the mixing matrix of the mth region of interest, S [m] ∈R L×1 is the source matrix of the mth region of interest, L represents the number of independent sources of the mth multi-channel data set, L≤τ; and represents the lth source component vector in the source matrix S [m] of the mth region of interest, and T represents the transposition operation;

[0023] Step 2.4.2, obtain the source component vector estimate y [m] of the mth region of interest using formula (2);

[0024]

[0025] In formula (3), (W [m] ) -1 is the inverse matrix of the mixing matrix W [m] ; and the lth group of source component vector estimates wherein, represents the source component estimate of the mth region of interest in the lth group of source component vector estimates.

[0026] Step 2.4.3, solve the inverse matrix (W [m] ) -1 using the independent component analysis method, so as to calculate the source component vector estimate y [m] , and the source component vectors of the source component vector estimate y [m] are sorted in descending order according to the correlation;

[0027] Step three, calculation of heart rate;

[0028] Adopting a Butterworth filter to estimate a target source component vector After filtering, a fast Fourier transform is used to calculate the filtered source component vector estimation The main frequency f of the pulse signal max Thus, the estimated heart rate value is time x f max Wherein, time represents 1 minute.

[0029] The electronic device comprises a memory and a processor, and the memory is used to store a program supporting the processor to execute the non-contact heart rate detection method, and the processor is configured to execute the program stored in the memory.

[0030] The computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the non-contact heart rate detection method are executed.

[0031] Compared with the prior art, the present application has the following beneficial effects:

[0032] 1. In the prior art, red, green and blue channel signals are usually used for heart rate extraction, but when the motion amplitude is large, the main frequency of the pulse signal will be completely submerged in the motion noise, as shown in the following formula: Figure 1 The CHROM method is selected for preprocessing in the present application, and the model-based method (color difference model) has better motion robustness to motion artifacts, and improves the quality of the blood volume pulse signal.

[0033] 2. The present application uses a time delay method to construct the input signal of joint blind source separation: the single-channel color difference signal generated by each region of interest is time-delayed to form a multi-channel signal dataset, which first meets the input of joint blind source separation, and secondly, the time delay highlights variables with quasi-periodicity, such as pulsatile signals and periodic motion variables, thereby increasing the feasibility of the joint blind source separation method for separating pulse signals and motion noise.

[0034] 3. The traditional blind source separation method cannot meet the prior assumption of linear combination when the motion amplitude of the subject is large, cannot solve the heart rate extraction problem when the motion frequency overlaps with the HR frequency band range of interest, and the order of the signal after blind source separation is not fixed. The joint blind source separation method of the present application can extract the source component vector commonly contained in two multi-channel datasets, maximize the correlation between the sources by jointly processing multiple datasets, extract the source group corresponding to each dataset, and ensure the consistent order of the extracted sources, thereby separating the motion noise and the pulse signal and improving the quality of the pulse signal.

[0035] 4、The application considers the relationship between periodic motion noise and periodic heart rate information, according to the first source component vector which is dominated by motion noise and the second source component vector which is often the target BVP signal, a suitable source component vector is selected as the target pulse signal, so as to extract the target pulse signal, remove the motion noise, and complete the video heart rate detection in the severe and periodic motion scene.

[0036] 5、The accuracy of heart rate detection decreases with the severe motion of the subject, especially when the subject's motion is severe and periodic. Therefore, the application uses the color difference signal CHROM combined with joint blind source separation to remove motion noise, wherein the preliminary denoising of CHROM on motion noise improves the quality of the BVP signal, and the time delay operation expands the single-channel CHROM signal into a multi-channel signal on the one hand, which is suitable for the input of JBSS, and on the other hand, highlights the periodicity of the heart rate variable and the regular motion variable, and uses JBSS to extract the coexisting BVP signal from multiple data sets in order of relevance from high to low, which improves the accuracy and motion robustness of heart rate detection in the severe and periodic motion scene. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is the frequency spectrum of the blood volume pulse signal under different motion interference in the prior art;

[0038] Figure 2 is a flowchart of the application;

[0039] Figure 3 is a schematic diagram of the division of the region of interest of the application;

[0040] Figure 4 is a schematic diagram of the blood volume pulse signal extracted by the application in the rowing machine motion scene. DETAILED DESCRIPTION

[0041] In this embodiment, a motion-robust non-contact heart rate detection method (CHROM-TDJBSS) is used to locate and track the region of interest of the I-frame video image, process the obtained color channel time sequence using the CHROM algorithm to complete the preliminary removal of motion noise, and under the action of TD, complete the expansion from a single-channel signal to a multi-channel signal, on the one hand, provide a suitable input for JBSS, and on the other hand, strengthen the periodic variable. Secondly, the source component vector commonly contained in two multi-channel data sets is extracted using JBSS, the blood volume pulse signal is selected, and the calculation of the heart rate is completed. Specifically, as shown in the figure, Figure 2 The method comprises the following steps:

[0042] Step one, extracting the pixel mean time series of the face region of interest from the I-frame video image of the subject, the pixel mean time series contains information related to heart rate and information related to motion; the specific steps include: region of interest positioning and tracking, and pixel mean time series extraction.

[0043] Step 1.1, positioning and tracking of the region of interest;

[0044] The convolution expert constrained local model provided by the OpenFace toolbox is used to detect and track the facial feature points of the I-frame video image. In the process of motion, the region of interest will swing left and right with the subject, and the area and position of the region of interest will change. In order to ensure the stability of the ROI, in this embodiment, the relatively stable feature points of OpenFace are selected to locate the circumscribed rectangle of the M regions of interest in each video image, and the width and height of the circumscribed rectangle are denoted as w m , h m ; M≥2;

[0045] In a specific case, the division of the region of interest is shown in Figure 3 , which is the forehead and the left face, and the number of regions of interest M=2. The video frame tracking requires the subject's face to be within the camera's capture range to ensure the tracking of the region of interest.

[0046] Step 1.2, extraction of the pixel mean time series of the region of interest;

[0047] The pixel mean time series of the mth region of interest is calculated frame by frame to obtain the RGB channel time series of the I-frame video image, including: the time series of the red channel in the I-frame video image The time series of the green channel in the I-frame video image and the time series of the blue channel in the I-frame video image Wherein, represents the red channel pixel mean of the mth region of interest in the ith frame; represents the green channel pixel mean of the mth region of interest in the ith frame; represents the blue channel pixel mean of the mth region of interest in the ith frame.

[0048] Step two, removal of motion noise under different motion scenarios;

[0049] Step 2.1, based on the skin reflection model, use formula (1) to preprocess the RGB channel time series into the single channel color difference signal C [m] of the mth region of interest, 1≤m≤M:

[0050]

[0051] wherein, and are the normalized band-pass filtered red channel pixel mean, green channel pixel mean, and blue channel pixel mean of the i-th frame in the m-th region of interest, respectively; denotes the chrominance signal defined by the projection of the red channel time series and the green channel time series, denotes the chrominance signal defined by the projection of the red, green, and blue channel time series, and a denotes a tuning parameter that separates light intensity variations from pulse signals, and σ(·) denotes the standard deviation of a given signal;

[0052] Step 2.2, according to the category of the motion scene, set the detrending parameter as λ, and use it to perform detrending processing on the single-channel chrominance signal S [m] of the m-th region of interest, to obtain the detrended single-channel chrominance signal of the m-th region of interest; denotes the detrended single-channel chrominance signal of the i-th frame in the m-th region of interest;

[0053] Step 2.3, perform time delay on the detrended single-channel chrominance signal of the m-th region of interest, to obtain τ channel data sets and take them as the time-delayed m-th multi-channel data set 1≤m≤M; research shows that the time delay operation can highlight the quasi-periodic heart rate variable or the periodic motion artifact variable in the rPPG signal. Among them, denotes the i-th column vector of the m-th region of interest, and

[0054] In a specific case, the value of τ is 3, that is, in this embodiment, the chrominance signal is time-delayed into a three-channel data set.

[0055] Step 2.4, decompose the multi-channel data set matrix X [m] using the independent vector analysis method in the joint blind source separation algorithm;

[0056] Step 2.4.1, regard each multi-channel data set as a linear mixture of L independent sources, and use formula (2) to construct the relationship between the multi-channel data set and the L independent sources:

[0057] X [m] = W [m] S [m] , 1≤m≤M (2)

[0058] In formula (2), W [m] is the mixing matrix of the m-th region of interest, S [m] ∈RL×1 is the source matrix of the mth region of interest, L denotes the number of independent sources of the mth multi-channel dataset, L≤τ; and is the source matrix of the mth region of interest [m] is the lth source component vector in the mth region of interest, T denotes the transpose operation;

[0059] Step 2.4.2, the source component vector estimate y [m] of the mth region of interest is obtained by using formula (2)

[0060]

[0061] In formula (3), (W [m] ) -1 is the inverse matrix of the mixing matrix w [m] ; and the lth set of source component vector estimates wherein, denotes the source component estimate of the mth region of interest in the lth set of source component vector estimates.

[0062] Step 2.4.3, the inverse matrix (W [m] ) -1 is solved by using the independent component analysis method, so as to calculate the source component vector estimate y [m] , and the source component vectors of the source component vector estimate y [m] are the same as the original channel number and are sorted from high to low according to the correlation order.

[0063] Step three, calculation of heart rate;

[0064] Step 3.1, through experimental observation, when there is obvious motion artifact, for example, in the ECG-Fitness database, the first ranked source component vector is still dominated by motion noise, and the second ranked source component vector is often the target BVP signal.

[0065] In the specific example, due to the large forehead ROI area, less non-rigid motion, good tracking effect and other factors, the second source component vector recovered from the forehead ROI dataset is selected as the candidate of the target BVP signal.

[0066] Step 3.2, the target source component vector is filtered by using a Butterworth filter, and then the main frequency f of the filtered source component vector estimate max is calculated by using fast Fourier transform, so that the estimated heart rate value is time×f max , wherein time denotes 1 minute.

[0067] In this embodiment, an electronic device includes a memory for storing a program supporting a processor to execute the above method, and the processor configured to execute the program stored in the memory.

[0068] In this embodiment, a computer readable storage medium has a computer program stored thereon, and the computer program is run by a processor to perform the steps of the above method.

[0069] To verify the effectiveness of the method, the performance of the method CHROM-TDJBSS is verified on the public database UBFC-RPPG and ECG-Fitness. The heart rate detection performance is evaluated by using conventional evaluation indicators, specifically including the mean absolute error HR mae (mean absolute error, MAE), the root mean square error HR rmse (root meansquare error, RMSE) and Pearson's correlation coefficient (r).

[0070] Table 1 UBFC-RPPG database comparison test results

[0071] Method HR mae (bpm) HR rmse (bpm) r POS 1.34 4.10 0.98 CHROM 1.65 4.82 0.97 ICA 3.26 9.99 0.86 JBSS 2.77 9.34 0.88 CHROM-TDICA 1.83 5.56 0.96 CHROM-TDJBSS 1.47 3.77 0.98

[0072] Table 2 ECG-Fitness database comparison test results

[0073] Method HR mae (bpm) HR rmse (bpm) r POS 12.87 20.70 0.58 CHROM 13.10 19.17 0.59 ICA 19.76 27.61 0.41 JBSS 18.98 25.79 0.44 CHROM-TDICA 12.99 20.93 0.62 CHROM-TDJBSS 9.93 16.12 0.75

[0074] Table 3 Heart rate measurement results under different exercise scenarios on the ECG-Fitness database

[0075]

[0076] Table 1 shows the results of the method and other methods on the UBFC-RPPG database. As can be seen from Table 1, the CHROM-TDJBSS method proposed in the present application shows the best performance on the other two indicators except that the HR mae value is 0.13bpm lower than the POS method which achieves the best performance, specifically, HR mae is 1.47bpm, and HR rmsedecreased to 3.77 bpm, r increased to 0.98. Model-based methods (CHROM and POS) outperformed blind source separation methods (ICA and JBSS). In addition, ablation experiments are also given in the table, CHROM-TDJBSS achieved the best performance in the comparison of CHROM and CHROM-TDJBSS, which fully demonstrates the effectiveness of the combination of CHROM method and JBSS method. The performance comparison results of ICA and CHROM-TDICA and JBSS and CHROM-TDJBSS prove that CHROM plus time delay can effectively suppress the interference of motion artifacts with relatively small amplitude. Overall, the CHROM-TDJBSS method proposed in the application almost achieves the best result on the UBFC-RPPG database, which shows that the method proposed in the application performs excellently on the relatively static UBFC-RPPG database.

[0077] Table 2 shows the experimental results of the method proposed in the application and other methods on the ECG-Fitness database. As can be seen from Table 2, the CHROM-TDJBSS method proposed in the application outperforms the other five methods in all indicators, HR mae decreased to 9.25 bpm, HR rmse decreased to 16.12 bpm, r increased to 0.75. The performance indicators of model-based methods (CHROM and POS) are much better than blind source separation-based methods (ICA and JBSS), which shows that when the motion amplitude is larger, the combination of motion noise and pulse signals does not conform to the linear model. The ablation experiment results of CHROM and CHROM-TDJBSS also prove the effectiveness of the combination of CHROM method and JBSS method. The comparison of ICA method and JBSS method and CHORM-TDJBSS method and CHROM-TDICA method shows that the JBSS-based method is better than the BSS method, because the BVP signal contained in multiple facial ROIs is better than the BVP signal from a single facial ROI. Overall, the experimental results prove that the method proposed in the application has superiority in removing motion artifacts with larger amplitude or even regular motion artifacts.

[0078] Table 3 details the experimental results of the method proposed in the application and other methods under different motion scenarios (rowing machine, elliptical machine, stationary bicycle, talking) of the ECG-Fitness database. As can be seen from Table 3, the CHROM-TDJBSS method proposed in the application achieves the best result in the four scenarios. At the same time, the performance of the model-based method is better than that of the blind source separation-based method in the four scenarios, and CHROM is better than POS in improving the overall BVP signal quality.

[0079] Figure 4The video clip of the subject exercising on the rowing machine for 30s is displayed, and the reference ECG signal is synchronously collected in the time period.

[0080] In summary, the experiments on the two databases prove that the method proposed in the application is more suitable for heart rate detection under intense exercise. The motion robust non-contact heart rate detection method fusing chromatic aberration model and joint blind source separation, the electronic device and the storage medium proposed in the application improve the accuracy of video heart rate detection under intense and periodic motion scenes, and have good motion robustness.

Claims

1. A motion-robust contactless heart rate detection method, characterized in that, The method comprises the following steps: Step one, extracting the pixel mean time series of the face region of interest from the I-frame video image of the subject, wherein the pixel mean time series contains information related to heart rate and information related to motion; Step 1.1, positioning and tracking of the region of interest; The face feature point detection and tracking method is used to detect and track the face feature points of the I frame video image, so as to locate the circumscribed rectangle of the M regions of interest in each frame of video image, and the width and height of the circumscribed rectangle are denoted as w m , h m ; M≥2. Step 1.2, extraction of the pixel mean time series of the region of interest; The pixel mean time series of the mth region of interest is calculated frame by frame, and the RGB channel time series of the I-frame video image is obtained, including the time series of the red channel in the I-frame video image the time series of the green channel in the I-frame video image and the time series of the blue channel in the I-frame video image wherein, represents the red channel pixel mean of the ith frame in the mth region of interest; represents the green channel pixel mean of the ith frame in the mth region of interest; represents the blue channel pixel mean of the ith frame in the mth region of interest; Step two, removal of motion noise under different motion scenarios; Step 2.

1. Based on the skin reflectance model, pre-process the RGB channel time series into a single channel chrominance signal C for the m-th region of interest using equation (1) [m] , 1≤m≤M: In formula (1), and respectively are the red channel pixel mean, green channel pixel mean, and blue channel pixel mean of the i-th frame of the m-th region of interest after normalization band-pass filtering; denotes a chrominance signal defined using the projection of the red and green channel time series, denotes a chrominance signal defined using the projection of the red, green, and blue channel time series, and a denotes a tuning parameter that separates light intensity variations from the pulse signal, and σ(·) denotes the standard deviation of a given signal; Step 2.2, according to the kind of motion scene, set the detrending parameter λ, and use it to the single channel chrominance signal S of the mth region of interest [m] detrending processing, get the single channel chrominance signal after detrending in the mth region of interest Wherein, indicates the i-th frame of single channel chrominance signal after detrending in the mth region of interest Step 2.

3. De-trend the single-channel color difference signal in the mth region of interest Perform time delay to obtain τ channel data sets and as the mth multi-channel data set after time delay wherein, denotes the ith column vector of the mth region of interest, and 1≤m≤M; T denotes transpose; Step 2.

4. Decompose the multi-channel data set matrix X using the Independent Vector Analysis method in the joint blind source separation algorithm [m] ; Step 2.4.1, regarding each multi-channel data set as a linear mixture of L independent sources respectively, and constructing a relationship between the multi-channel data set and the L independent sources by using formula (2): X [m] = W [m] S [m] , 1≤m≤M (2) In formula (2), W [m] is a mixing matrix of the mth region of interest, S [m] ∈R L×1 is a source matrix of the mth region of interest, L represents the number of independent sources of the mth multi-channel data set, L≤τ; and represents the lth source component vector in the source matrix S [m] of the mth region of interest, T represents a transposition operation; Step 2.4.

2. Obtain the source component vector estimate y for the mth region of interest using formula (2) [m] ; y [m] = (W [m] ) -1 ·X [m] (3) In formula (3), (W [m] ) -1 is the inverse matrix of the mixing matrix W [m] ; and the lth group of source component vector estimates wherein denotes the source component estimate of the mth region of interest in the lth group of source component vector estimates; Step 2.4.3, solving the inverse matrix (W [m] ) -1 , thereby calculating the source component vector estimate y [m] , and the source component vectors of the source component vector estimate y [m] are ordered from high to low in terms of correlation. Step three, calculation of heart rate; Adopting a Butterworth filter to estimate the target source component vector After filtering, the filtered source component vector estimation is calculated using fast Fourier transform The main frequency f of the signal max Thus, the estimated heart rate value is time x f max Wherein, time represents 1 minute.

2. An electronic device comprising a memory and a processor, characterized in that The memory is configured to store a program supporting the processor to execute the non-contact heart rate detection method of claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to perform the steps of the non-contact heart rate detection method of claim 1.

Images