CN110353648A - A kind of real-time heart rate detection method based on dual camera - Google Patents

Abstract

The invention discloses a real-time heart rate detection method based on dual cameras, comprising the following steps: turn on the main camera and the reference camera, alternately shoot the main video and the reference video containing human faces at the same frame rate; create two human face detectors, respectively Obtain the ROI of the main camera and the ROI of the reference camera; intercept the background area of the main camera and the background area of the reference camera; calculate the characteristics of the ROI and the background area of each frame based on the noise information extracted from the ROI and the background area of the reference camera value, get the one-dimensional feature sequence on the time axis; filter the one-dimensional feature sequence with a low-pass filter to get the heart rate signal; use the peak counting method to calculate the heart rate value; collect data in real time to update the signal sequence, repeat the above steps, and calculate heart rate value. The method proposed in the invention reduces the signal distortion caused by ambient light interference, and greatly improves the accuracy and stability in non-contact heart rate detection.

Description

A real-time heart rate detection method based on dual cameras

technical field

The invention relates to the technical field of digital image processing, in particular to a real-time heart rate detection method based on dual cameras.

Background technique

According to the statistics of the World Health Organization, the mortality rate of cardiovascular disease ranks first, significantly higher than that of tumors and other diseases. However, many people often have insufficient understanding of heart disease, resulting in missing the best time to seek medical advice.

Research shows that most early heart attacks and strokes are preventable. In addition to ensuring good living habits, the most important thing is early detection and early treatment. Palpitation, chest tightness, chest pain, dizziness or fainting, these seemingly insignificant symptoms, are all signals of cardiovascular and cerebrovascular diseases, and most people miss the best time for treatment because of carelessness.

The most common inspection method in current hospitals is to use an electrocardiograph (ECG) to check the patient's heart function, but the monitoring time is relatively short, and the equipment is expensive and difficult to carry. For portability and low cost, people designed and invented the finger-clip heart rate tester. However, wearing it for a long time will lead to poor blood circulation in the user's fingertips, causing discomfort to the user.

In recent years, with the popularization of computer and real-time video image processing and other technologies, the proposal of image PPG (Photoplethysmography) technology provides a practical idea for realizing non-invasive and non-contact real-time heart rate measurement. Image PPG technology means that because the human heart is constantly contracting and relaxing, the blood filling degree in the human blood vessels will also change continuously with the heartbeat, and the absorption of light will change with the volume of blood. The change presents a pulsating change consistent with the heartbeat, and at the same time, the intensity of the light reflected by the skin surface also undergoes a corresponding periodic change, which is manifested as a change in skin color.

At present, the interference and noise of ambient light commonly exist in the heart rate detection method based on video, which leads to the instability of image data, thus increasing the difficulty of extracting the small signal of facial parameters.

Contents of the invention

The purpose of the present invention is to provide a real-time heart rate detection method based on dual cameras, which can effectively improve the accuracy and stability of the camera-based real-time heart rate detection.

The technical scheme that realizes the object of the present invention is:

A method for real-time heart rate detection based on dual cameras, the method comprising the following steps:

(1) Turn on the main camera and the reference camera, and alternately shoot the main video and the reference video containing faces for 20 seconds at the same frame rate;

(2) create two face detectors, obtain the region of interest of each frame of the main video and the reference video respectively, and extract each frame of U channel data, obtain the region of interest of the main camera and the region of interest of the reference camera U channel sequence;

(3) Intercept the main camera background area and the reference camera background area, and obtain the U channel sequence of the main camera background area and the reference camera background area;

(4) According to the noise information extracted from the region of interest and the background region of the reference camera, calculate the feature values of the region of interest and the background region of each frame to obtain a one-dimensional feature sequence on the time axis;

(5) A low-pass filter is used to filter the one-dimensional feature sequence, and the filtered one-dimensional feature sequence is used as a heart rate signal;

(6) Obtain the number of heartbeats by using the peak counting method, and calculate the heart rate value;

(7) Collect data in real time to update the signal sequence, and repeat steps (2)-(6) to calculate the real-time heart rate value.

Described step (1) is specifically:

Create the main camera object and reference camera object, set the same frame rate fs;

set frame counter frame_counter = 0;

First turn on the reference camera, then turn on the main camera after Δt time, Δt<1/fs, and shoot 20s of videos containing faces at the frame rate of fs, record the current frame of the reference camera and the current frame of the main camera as Frame_ref(i) and Frame(i), frame_counter=K=20*fs.

Described step (2) is specifically:

Utilize the Viola-Jones algorithm to create a face detector, obtain the starting point coordinates (x, y), (x_ref, y_ref) and the face size ( w, h), (w_ref, h_ref);

According to the proportion of the face, intercept Frame(i) and Frame_ref(i) with a height range of x+0.5*h~x+0.7*h and a width range of y+0.1*w~y+0.3*w, namely the main camera region of interest Interest(i) and reference camera region of interest Interest_ref(i), obtain the RGB channels of the main camera region of interest and the reference camera region of interest in the current frame;

The RGB channels of the main camera region of interest and the reference camera region of interest are converted into YUV channels, and two-dimensional U channels U(i), U_ref(i) are extracted respectively; the U channel calculation formula is as follows:

U＝-0.169*R-0.331*G+0.5*B

The U channel of the region of interest of the main camera in each frame constitutes a sequence U(1), U(2),..., U(K), and the U channel of the region of interest of the reference camera in each frame constitutes a sequence U_ref(1 ), U_ref(2), ..., U_ref(K).

Described step (3) is specifically:

Create n*n background windows, the value of n is 8, 16, 32, 64 or 128, which can be adjusted according to the video resolution;

Use the background window to move row-by-row and column-by-column in the complement area of the main camera ROI and the reference camera ROI in each frame, and take out the temporary pixel matrix covered by the background window each time, which is respectively recorded as TempArray and TempArray_ref, respectively calculate the variance U_Sigma and U_Sigma_ref of its U channel matrix;

If U_Sigma>preset variance threshold Tr, then continue to move the background window and recalculate, otherwise use the current TempArray as the background area of the main camera in the current frame, and extract the U channel U_back(i) of the background area of the main camera in the current frame, each frame The U channel in the background area of the main camera constitutes a sequence U_back(1), U_back(2),..., U_back(K);

If U_Sigma_ref>preset variance threshold Tr, continue to move the background window and recalculate; otherwise, use the current TempArray_ref as the reference camera background area of the current frame, and extract the U channel U_back_ref(i) of the reference camera background area in the current frame, each frame The U channel of the background area of the reference camera constitutes a sequence U_back_ref(1), U_back_ref(2),..., U_back_ref(K);

Described step (4) is specifically:

Calculate the mean value m_ref(i) and standard deviation σ(i) of the two-dimensional U channel matrix U_ref(i) of the reference camera region of interest in the current frame, calculate the difference matrix between U_ref(i) and m_ref(i), and calculate the difference The number of pixels whose absolute value exceeds 3*σ(i) in the value matrix is denoted as Num(i);

Sort the U channel values in the two-dimensional U channel matrix U(i) of the main camera region of interest in the current frame from small to large, remove Num(i)/2 maximum values, and remove Num(i)/2 The minimum value calculates the average value m (i) of the remaining U channel values, as the two-dimensional U channel matrix eigenvalue of the region of interest described in the current frame;

Calculate the mean m_back_ref(i) and standard deviation σ_ref(i) of the two-dimensional U channel matrix of the reference camera background area in the current frame, calculate the difference matrix between U_back_ref(i) and m_back_ref(i), and count the absolute value in the difference matrix The number of pixels exceeding 3*σ_ref(i) is recorded as Num_back(i);

Sort the U channel values in the two-dimensional U channel matrix U_back(i) of the main camera background area in the current frame from small to large, remove Num_back(i)/2 maximum values, and remove Num_back(i)/2 minimum values value, calculate the mean value m_back(i) of the remaining U channel values, as the two-dimensional U channel matrix eigenvalue of the background area of the current frame;

Set the noise removal factor according to the severity of the ambient light change, denoted as β, where the value range of β is 0 to 1, the more severe the ambient light change, the larger the β value, calculate U_character(i)=m(i)-β* m_back(i), as the feature value of the i-th frame;

The feature values of each frame form a one-dimensional feature sequence U_character(1), U_character(2), . . . , U_character(K) on the time axis.

Described step (5) is specifically:

Use a P-order 3Hz Butterworth filter to filter the one-dimensional feature sequence U_character(1), U_character(2),..., U_character(K) to obtain a length J filter sequence Fil(1), Fil(2) ,...,Fil(J);

Among them, J=K+P-1;

Described step (6) is specifically:

Scan the filter sequence Fil(1), Fil(2),...,Fil(J) whose length is J, and find the peak value: the Fil(1) does not perform calculation processing; the Fil(2), Fil(3 ),…, Fil(J-1) is compared with two adjacent points, if the value of this point is larger than the two adjacent points, then this point is considered as a peak value; that is: if Fil(j)>Fil(j- 1) and Fil(j)>Fil(j+1), then Fil(j) is a peak point;

Count the number of peak points within the 20s, which is recorded as Peak_20s;

Calculate Heart_Rate=Peak_20s*3, that is, the real-time heart rate value corresponding to the 20s.

Described step (7) is specifically:

Discard the first 5s data Umid(1), Umid(2),...Umid(L1) in the 20s, where L1=fs*5, and the last 15s data form a temporary sequence Umid(1), Umid(2),...Umid (L2), wherein L2=fs*15;

Update and collect 5s data, supplement after the temporary sequence Umid(1), Umid(2),...Umid(L2), so as to realize the update of the sequence Umid(1), Umid(2),...Umid(K), K =L1+L2, repeat steps (2)~(6) to get the real-time heart rate value.

The beneficial effects of the technical solution provided by the present invention are: a real-time heart rate detection method based on dual cameras proposed by the present invention, the method determines the range of a specific region of interest according to the facial features in signal extraction, and improves the signal-to-noise ratio , determine the range of the background area according to the background features, and increase the stability of the data; in terms of feature value extraction, the intra-frame noise parameters provided by the reference frame are used to reduce the intra-frame noise, and the inter-frame noise parameters provided by the background area are used to reduce the impact of environmental changes, thereby Improve the accuracy of heart rate value calculation. The present invention is applicable to real-time heart rate detection scenarios of fatigue-prone people such as drivers. The method proposed in the invention reduces signal distortion and jitter caused by ambient light interference, and greatly improves the accuracy and stability in non-contact heart rate detection.

Description of drawings

Fig. 1 is a flowchart of the present invention;

Fig. 2 is the U channel sequence flowchart of the present invention extracting main camera ROI and reference camera ROI;

Fig. 3 is the U channel sequence flowchart of the present invention extracting main camera background area and reference camera background area;

Fig. 4 is the algorithm flowchart based on dual camera of the present invention;

Fig. 5 is the flow chart that the present invention adopts 3Hz Butterworth low-pass filter to filter feature sequence;

Fig. 6 is the flow chart that the present invention adopts peak counting method to calculate heart rate value;

Fig. 7 is a flow chart of real-time heart rate update in the present invention.

Detailed ways

In order to more clearly illustrate the technical means, technical improvements and beneficial effects of the present invention, the present invention will be described in detail below in conjunction with the accompanying drawings.

Example

Referring to Fig. 1-7, present embodiment comprises the following steps:

S101: Turn on the main camera and the reference camera, and alternately shoot the main video and the reference video including the human face for 20 seconds at the same frame rate. Take a 30-year-old healthy woman as an example for illustration.

The steps are specifically:

Create the main camera object and reference camera object, set the same frame rate fs=20Hz;

set frame counter frame_counter = 0;

Turn on the reference camera first, then turn on the main camera after Δt=0.025s, and shoot videos containing human faces at a frame rate of 20Hz for 20s, record the current frame of the reference camera and the current frame of the main camera as Frame_ref(i) and Frame( i), frame_counter=K=400.

S102: Create two face detectors to obtain the ROI of the main camera and the ROI of the reference camera respectively, extract the U channel data based on the RGB data of each frame, and obtain the U of the ROI of the main camera and the ROI of the reference camera Channel sequence, see Figure 2.

The steps are specifically:

Use the Viola-Jones algorithm to create the face detector, and obtain the starting point coordinates (x, y), (x_ref, y_ref) and face size (w, h) of the main area of the face and the reference area of the face respectively , (w_ref, h_ref); For example, the starting point coordinates of the main face area are (400, 500), the face size is (100, 200), the starting point coordinates of the face reference area are (450, 550), the face size is (100, 200).

According to the proportion of the face, intercept the area of Frame(i) whose height ranges from 500 to 540 and whose width ranges from 510 to 530, that is, the area of interest of the main camera Interest(i), and intercept Frame_ref(i) whose height range is from 550 to 590 and whose width The area with a range of 560 to 580, that is, the area of interest of the reference camera Interest_ref(i), obtains the RGB channels of the area of interest of the main camera and the area of interest of the reference camera in the current frame;

The RGB channels of the main camera region of interest and the reference camera region of interest are converted into YUV channels, and two-dimensional U channels U(i), U_ref(i) are extracted respectively; the U channel calculation formula is as follows:

U＝-0.169*R-0.331*G+0.5*B

The U channel of the region of interest of the main camera in each frame constitutes a sequence U(1), U(2),..., U(400), and the U channel of the region of interest of the reference camera in each frame constitutes a sequence U_ref(1 ), U_ref(2), ..., U_ref(400).

S103: Intercept the background area of the main camera and the background area of the reference camera, and obtain the U channel sequence of the background area of the main camera and the background area of the reference camera, see FIG. 3 .

The steps are specifically:

Create an 8*8 background window;

Use the background window to move row-by-row and column-by-column in the complement area of the main camera ROI and the reference camera ROI in each frame, and take out the temporary pixel matrix covered by the background window each time, which is respectively recorded as TempArray and TempArray_ref, respectively calculate the variance U_Sigma and U_Sigma_ref of its U channel matrix;

If U_Sigma>10, continue to move the background window and recalculate; otherwise, use the current TempArray as the background area of the main camera in the current frame, extract the U channel U_back(i) of the background area of the main camera in the current frame, and the main camera in each frame The U channel in the background area constitutes the sequence U_back(1), U_back(2),..., U_back(400);

If U_Sigma_ref>10, continue to move the background window and recalculate; otherwise, use the current TempArray_ref as the reference camera background area of the current frame, extract the U channel U_back_ref(i) of the reference camera background area in the current frame, and refer to the reference camera in each frame The U channel in the background area constitutes the sequence U_back_ref(1), U_back_ref(2),..., U_back_ref(400);

S104: Calculate the eigenvalues of the region of interest and the background region of each frame based on the dual cameras, and obtain the one-dimensional feature sequence on the time axis, refer to Fig. 4, in the figure, a is the calculation of the mean and standard deviation of U_ref(i), and b is the calculation It is the calculation of the mean and standard deviation of U_back_ref(i), and the figure c is the calculation of the feature value of the i-th frame.

The steps are specifically:

Calculate the mean value m_ref(i) and standard deviation σ(i) of the two-dimensional U channel matrix of the reference camera region of interest in the current frame, for example, m_ref(i)=133, σ(i)=6; obtain U_ref(i) For the difference matrix with m_ref(i), the number of pixels whose absolute value exceeds 3*σ(i)=18 in the statistical difference matrix is recorded as Num(i), for example, Num(i)=10;

Sort the U channel values in the two-dimensional U channel matrix U(i) of the area of interest of the main camera in the current frame from small to large, remove 5 maximum values, remove 5 minimum values, and calculate U(i) Average value m(i), as the eigenvalue of the two-dimensional U channel matrix of the region of interest in the current frame, for example, m(i)=130;

Calculate the mean m_back_ref(i) and standard deviation σ_ref(i) of the two-dimensional U channel matrix of the reference camera background area in the current frame, for example, m_back_ref(i)=212, σ_ref(i)=2; obtain U_back_ref(i) and The difference matrix of m_back_ref(i), the number of pixels whose absolute value exceeds 3*σ_ref(i) in the statistical difference matrix, is recorded as Num_back(i), for example, Num_back(i)=4;

Sort the U channel values in the two-dimensional U channel matrix U_back(i) of the main camera background area in the current frame from small to large, remove 2 maximum values, remove 2 minimum values, and calculate the average of U_back(i) Value m_back(i), as the eigenvalue of the two-dimensional U channel matrix of the background area, for example, m_back(i)=210;

Set the noise removal factor according to the severity of the ambient light change, denoted as β=0.5, calculate U_character(i)=130-0.5*210=25, as the feature value of the i-th frame;

The feature values of each frame form a one-dimensional feature sequence U_character(1), U_character(2), . . . , U_character(400) on the time axis.

S105: Use a low-pass filter to filter the one-dimensional feature sequence, and use the filtered one-dimensional feature sequence as the heart rate signal, see FIG. 5 .

The steps are specifically:

A 40-order 3Hz Butterworth filter is used to filter the one-dimensional feature sequence U_character(1), U_character(2),..., U_character(400) to obtain a filter sequence Fil(1), Fil(2) with a length of 439 ),...,Fil(439);

S106: Use the peak counting method to obtain the number of heartbeats, and calculate the heart rate value, see FIG. 6 .

The steps are specifically:

Scan the filter sequence Fil(1), Fil(2),...,Fil(439) whose length is J, and find the peak value: the Fil(1) does not perform calculation processing; the Fil(2), Fil(3 ),…, Fil(438) are compared with two adjacent points, if the value of this point is larger than the two adjacent points, then this point is considered to be a peak; that is: if Fil(j)>Fil(j-1) And Fil(j)>Fil(j+1), then Fil(j) is a peak point;

Count the number of peak points within the 20s, recorded as Peak_20s, for example Peak_20s=28;

Calculate the real-time heart rate value Heart_Rate=Peak_20s*3=84 corresponding to the 20s, that is, the real-time heart rate value of the person in the 20s is 84 beats/min.

S107: Collect data in real time to update the signal sequence, and repeat steps (2)-(6) to calculate the real-time heart rate value, see FIG. 7 .

Discard the first 5s data Umid(1), Umid(2),...Umid(100) in the 20s, and the latter 15s data form a temporary sequence Umid(1), Umid(2),...Umid(300);

Update and collect 5s data, supplement after the temporary sequence Umid(1), Umid(2),...Umid(300), so as to realize the update of the sequence Umid(1), Umid(2),...Umid(400), repeat Steps (2) to (6), namely to obtain the real-time heart rate value.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (8)

1. A real-time heart rate detection method based on dual cameras, characterized in that the method comprises the following specific steps: Step 1: Turn on the main camera and the reference camera, and alternately shoot the main video and the reference video containing faces for 20 seconds at the same frame rate; Step 2: Create two face detectors, obtain the region of interest of each frame of the main video and the reference video respectively, and extract the U channel data of each frame, and obtain the region of interest of the main camera and the region of interest of the reference camera U channel sequence; Step 3: Intercept the background area of the main camera and the background area of the reference camera, and obtain the U channel sequence of the background area of the main camera and the background area of the reference camera; Step 4: According to the noise information extracted from the region of interest and the background region of the reference camera, calculate the feature values of the region of interest and the background region of each frame, and obtain a one-dimensional feature sequence on the time axis; Step 5: Use a low-pass filter to filter the one-dimensional feature sequence, and use the filtered one-dimensional feature sequence as the heart rate signal; Step 6: Use the peak counting method to obtain the number of heartbeats, and calculate the heart rate value; Step 7: Collect data in real time to update the signal sequence, and repeat steps (2)-(6) to calculate the real-time heart rate value. 2. The real-time heart rate detection method according to claim 1, wherein said step 1 is specifically: Create the main camera object and reference camera object, set the same frame rate fs; set frame counter frame_counter = 0; First turn on the reference camera, then turn on the main camera after Δt time, Δt<1/fs, and shoot 20s of videos containing faces at the frame rate of fs, record the current frame of the reference camera and the current frame of the main camera as Frame_ref(i) and Frame(i), frame_counter=K=20*fs. 3. The real-time heart rate detection method according to claim 1, wherein said step 2 is specifically: Utilize the Viola-Jones algorithm to create a face detector, obtain the starting point coordinates (x, y), (x_ref, y_ref) and the face size ( w, h), (w_ref, h_ref); According to the proportion of the face, intercept Frame(i) and Frame_ref(i) with a height range of x+0.5*h~x+0.7*h and a width range of y+0.1*w~y+0.3*w, namely the main camera region of interest Interest(i) and reference camera region of interest Interest_ref(i), obtain the RGB channels of the main camera region of interest and the reference camera region of interest in the current frame; The RGB channels of the main camera region of interest and the reference camera region of interest are converted into YUV channels, and two-dimensional U channels U(i), U_ref(i) are extracted respectively; the U channel calculation formula is as follows: U＝-0.169*R-0.331*G+0.5*B The U channel of the region of interest of the main camera in each frame constitutes a sequence U(1), U(2),..., U(K), and the U channel of the region of interest of the reference camera in each frame constitutes a sequence U_ref(1 ), U_ref(2), ..., U_ref(K). 4. The real-time heart rate detection method according to claim 1, wherein said step 3 is specifically: Create n*n background windows, the value of n is 8, 16, 32, 64 or 128, and adjust according to the video resolution; Use the background window to move row-by-row and column-by-column in the complement area of the main camera ROI and the reference camera ROI in each frame, and take out the temporary pixel matrix covered by the background window each time, which is respectively recorded as TempArray and TempArray_ref, respectively calculate the variance U_Sigma and U_Sigma_ref of its U channel matrix; If U_Sigma>preset variance threshold Tr, then continue to move the background window and recalculate, otherwise use the current TempArray as the background area of the main camera in the current frame, and extract the U channel U_back(i) of the background area of the main camera in the current frame, each frame The U channel in the background area of the main camera constitutes a sequence U_back(1), U_back(2),..., U_back(K); If U_Sigma_ref>preset variance threshold Tr, continue to move the background window and recalculate; otherwise, use the current TempArray_ref as the reference camera background area of the current frame, and extract the U channel U_back_ref(i) of the reference camera background area in the current frame, each frame The U channel of the background area of the reference camera constitutes a sequence U_back_ref(1), U_back_ref(2), . . . , U_back_ref(K). 5. The real-time heart rate detection method according to claim 1, wherein said step 4 is specifically: Calculate the mean value m_ref(i) and standard deviation σ(i) of the two-dimensional U channel matrix U_ref(i) of the reference camera region of interest in the current frame, calculate the difference matrix between U_ref(i) and m_ref(i), and calculate the difference The number of pixels whose absolute value exceeds 3*σ(i) in the value matrix is denoted as Num(i); Sort the U channel values in the two-dimensional U channel matrix U(i) of the main camera region of interest in the current frame from small to large, remove Num(i)/2 maximum values, and remove Num(i)/2 The minimum value calculates the average value m (i) of the remaining U channel values, as the two-dimensional U channel matrix eigenvalue of the region of interest described in the current frame; Calculate the mean value m_back_ref(i) and standard deviation σ_ref(i) of the two-dimensional U channel matrix of the reference camera background area in the current frame, calculate the difference matrix between U_back_ref(i) and m_back_ref(i), and count the absolute value in the difference matrix The number of pixels exceeding 3*σ_ref(i) is recorded as Num_back(i); Sort the U channel values in the two-dimensional U channel matrix U_back(i) of the main camera background area in the current frame from small to large, remove Num_back(i)/2 maximum values, and remove Num_back(i)/2 minimum values value, calculate the mean value m_back(i) of the remaining U channel values, as the two-dimensional U channel matrix eigenvalue of the background area of the current frame; Set the noise removal factor according to the severity of the ambient light change, denoted as β, where the value range of β is 0 to 1, the more severe the ambient light change, the larger the β value, calculate U_character(i)=m(i)-β* m_back(i), as the feature value of the i-th frame; The feature values of each frame form a one-dimensional feature sequence U_character(1), U_character(2), . . . , U_character(K) on the time axis. 6. The real-time heart rate detection method according to claim 1, wherein the step 5 is specifically: Use a P-order 3Hz Butterworth filter to filter the one-dimensional feature sequence U_character(1), U_character(2),..., U_character(K) to obtain a length J filter sequence Fil(1), Fil(2) ,...,Fil(J); Wherein, J=K+P-1. 7. The real-time heart rate detection method according to claim 1, wherein said step 6 is specifically: Fil(1) in the filter sequence Fil(1), Fil(2),...,Fil(J) whose length is J is not calculated; the Fil(2), Fil(3),..., Fil(J-1) is compared with two adjacent points, if the value of this point is larger than the two adjacent points, then this point is considered to be a peak value; that is, if Fil(j)>Fil(j-1) and Fil( j)>Fil(j+1), then Fil(j) is a peak point; Count the number of peak points within the 20s, which is recorded as Peak_20s; Calculate Heart_Rate=Peak_20s*3, that is, the real-time heart rate value corresponding to the 20s. 8. The real-time heart rate detection method according to claim 1, wherein said step 7 is specifically: Discard the first 5s data Umid(1), Umid(2),...Umid(L1) in the 20s, where L1=fs*5, and the last 15s data form a temporary sequence Umid(1), Umid(2),...Umid (L2), wherein L2=fs*15; Update and collect 5s data, supplement after the temporary sequence Umid(1), Umid(2),...Umid(L2), so as to realize the update of the sequence Umid(1), Umid(2),...Umid(K), K =L1+L2, repeat steps (2)~(6) to get the real-time heart rate value.