A method for rapid detection of human physiological signals

Abstract

The application relates to a human body signal detection field, in particular to a method for detecting human body signals through facial information, and provides a human body physiological signal rapid detection method, which has a shorter measurement time, after an original IPPG signal is acquired, the IPPG original signal is slidingly intercepted with a truncation time of 1.8s, the sliding frame number is 1 frame frequency, after polynomial fitting and averaging of each segment fragment set, the IPPG signal can be acquired; low-frequency noise caused by relative motion and light change is corrected and weighted, and then high-precision human body physiological signals can be acquired, and the shortest calculation time of the human body physiological signals is 1.8 seconds.

Description

Technical Field

[0001] This invention belongs to the field of human body signal detection, and in particular relates to a method for detecting human body signals using facial information. Background Technology

[0002] While modern medical devices have been dramatically transformed by technological advancements, the monitoring of key physiological parameters remains largely unchanged from 50 years ago. Heart rate is typically monitored using electrocardiograms (ECGs), a clinical practice for over a century. Nurses calculate respiratory rate by counting chest movements over a predetermined timeframe, and blood pressure is measured via an inflatable brace. A more recent major change was the introduction of pulse oximeters in the 1870s for measuring peripheral oxygen saturation. These traditional methods are well-established in clinical application, but they also have drawbacks. For example, prolonged wear of ECG and pulse oximeter probes can cause patient discomfort, restrict movement, and increase the risk of infection. Therefore, a continuous, non-contact method for measuring key physiological parameters holds significant clinical promise. Over the past decade, the field of non-contact key physiological parameter monitoring has emerged, with increasing research demonstrating that imaging optical volumetric plethysmography (IPPG) can estimate key physiological parameters using video cameras.

[0003] In 2014, the Oxford team used an autoregressive model to estimate key signals including respiration and heart rate, finding that the estimated heart rate was comparable to reference values ​​measured simultaneously by ear and finger oximeters. A review of non-contact respiratory measurement published by Kranjec in 2014 compared new non-contact measurement methods and highlighted RGB imaging-based approaches. They argued that the main advantages of this method were its low cost and potential for simultaneous monitoring of multiple targets; however, the complexity of the processing and low temporal resolution were its drawbacks.

[0004] Existing technologies require a relatively long facial video acquisition time to ensure the signal-to-noise ratio of the BVP signal. The typical minimum signal acquisition time is 20 to 30 seconds. However, in the application scenario of smart canteen systems, the real-time detection efficiency of human physiological signals needs to be comparable to the detection efficiency of automatic calculation of meal prices and analysis of meal nutritional components. The long detection time of physiological signals will greatly restrict the operating efficiency of smart canteen systems, thereby reducing the application value of human physiological signal detection.

[0005] Therefore, there is an urgent need for a technology to solve the above problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for addressing the problems existing in the background art.

[0007] A method for rapid detection of human physiological signals, characterized by comprising the following steps, which are performed sequentially:

[0008] S1: Capture facial video. The image acquisition module uses a high-depth camera to record high-depth video data of facial scattered light signals.

[0009] S2: Extract the human facial feature nodes from the high-depth video data obtained in step S1, and automatically divide the high signal-to-noise ratio region based on the blood vessel distribution of the human face, and extract the original IPPG signal, which can be represented as follows;

[0010] (1)

[0011] The variable reflecting the change of the IPPG signal over time is shown in the formula. is the grayscale value of the camera's RGB channel at the k-th pixel, and is a 3×N matrix where N is the length of t. It's noise. It is the vector of the blood flow pulse (BVP), reflecting the response intensity of the camera's RGB channels to the BVP signal. It is the intensity of the incident light. It is a BVP signal, where t is time information;

[0012] S3: Extract the BVP signal from the IPPG signal obtained in S2, which mainly includes the following steps;

[0013] S31: The original signal obtained by formula (1) It is divided into a set of segments, each 1.8 seconds long, connected end to end;

[0014] S32: The original signal after processing S31 The signal after a 1.8s sliding window truncation is represented in discrete form:

[0015] (2)

[0016] in, Represents discrete sampling k ( k The original signal of channels R, G, B, where n is the nth sampling point. , N The length of the original signal; In equation (1) ;in It is a multiplicative signal that is independent of time. It is a time-dependent multiplicative signal; Indicates the BVP signal. Some of these can be considered as weak "noise-like" values. The following data processing methods are used to fit and eliminate this "noise-like" data. ;

[0017] Using a cubic polynomial Fitting Part of, among which It is the mathematical expression of a cubic polynomial; to minimize the error of the fitting result, it is necessary to make... and Minimize the sum of squares of the residuals:

[0018] (3)

[0019] right Finding the differential and making it equal to 0, we obtain the following system of equations:

[0020] (4)

[0021] Formula 4 can be expressed in matrix form:

[0022] (5)

[0023] in: , , T denotes the transpose of the matrix; The matrix is ​​a Vandermonde matrix, for The matrix QR decomposition yields:

[0024] (6)

[0025] Where Q is an N×N unitary matrix and R is an N×4 upper triangular matrix; by obtaining the coefficients b of the cubic polynomial from equation (6), the fitting trend of the noise component can be obtained. ;

[0026] S33 weighted processing to obtain the BVP signal;

[0027] Further eliminate the information obtained in S32 Residual noise; based on the hemoglobin absorption spectrum, red light carries a large amount of noise information and a small amount of BVP information, green light carries a large amount of BVP information and a small amount of noise information, and blue light carries noise information and a trace amount of BVP information; after the original signal is fitted with polynomial noise and eliminated, a weighted signal processing method is used: the green channel signal is subtracted from the red channel signal, that is:

[0028] (7)

[0029] In the formula, This is the weighted blood flow pulsation signal. and The green and red channel signals of the BVP signal obtained by formula (6) are respectively;

[0030] S4: Using the polynomial fitting and weighting results obtained in step S3. The signals are used to calculate heart rate and blood oxygen saturation, thus completing the detection of human physiological signals.

[0031] In step S31, considering that the lowest human heart rate is around 50 bmp, which means that a complete heart pumping and diastolic activity takes about 1.2 seconds, the length of the signal cannot be less than 1.8 seconds. This ensures that three extreme points appear within a signal segment, so that the trend of BVP signal change can be fitted in a static state and the power of BVP signal can be preserved.

[0032] In step S1, the high bit depth camera is a camera with a bit depth of 8 bits or 10 bits.

[0033] In step S33, red light has a longer wavelength, i.e., 600-700nm, while green light has a shorter wavelength, i.e., about 500-560nm. In human skin tissue, red light penetrates deeper and is mainly absorbed by hemoglobin in the blood, especially deoxyhemoglobin. Green light penetrates shallower and is mainly absorbed in the epidermis and superficial dermis. Since the physiological activities in the superficial layer are more abundant, green light can more sensitively capture this superficial information because of its shallower penetration. Therefore, red light carries a large amount of noise information and a small amount of BVP information, while green light carries a large amount of BVP information and a small amount of noise information.

[0034] In step S2, the high signal-to-noise ratio region is: the forehead, that is, above the brow ridge and the forehead, which is a branch of the internal carotid artery; the sides of the nose and the tip of the nose, which is a major branch of the external carotid artery; the area from the angle of the mandible to the corner of the mouth, where the arteries directly receive the blood flow from the external carotid artery; and the area in front of the tragus, where there are branches of the maxillary artery, so the pulsation is more obvious.

[0035] Through the above design scheme, the present invention provides a rapid detection method for human physiological signals with a shorter measurement time. After acquiring the original IPPG signal, the original IPPG signal is truncated with a truncation time of 1.8s and the number of truncated frames is 1 frame. After polynomial fitting and averaging of each segment set, the IPPG signal can be obtained. Due to the low-frequency noise caused by relative motion and illumination changes, the noise is corrected and weighted to obtain a high-precision human physiological signal. The shortest calculation time for human physiological signals in this application is 1.8 seconds. Attached Figure Description

[0036] Figure 1This is a flowchart of the present invention;

[0037] Figure 2 This is a diagram illustrating the IPPG noise fitting process of the present invention. Detailed Implementation

[0038] The present application will be further described with reference to the accompanying drawings: A method for rapid detection of human physiological signals, characterized by comprising the following steps, which are performed sequentially:

[0039] S1: Capture facial video. The image acquisition module uses a high-depth camera to record high-depth video data of facial scattered light signals.

[0040] S2: Extract the human facial feature nodes from the high-depth video data obtained in step S1, and automatically divide the high signal-to-noise ratio region based on the blood vessel distribution of the human face, and extract the original IPPG signal, which can be represented as follows;

[0041] (1)

[0042] The variable reflecting the change of the IPPG signal over time is shown in the formula. is the grayscale value of the camera's RGB channel at the k-th pixel, and is a 3×N matrix where N is the length of t. It's noise. It is the vector of the blood flow pulse (BVP), reflecting the response intensity of the camera's RGB channels to the BVP signal. It is the intensity of the incident light. It is a BVP signal, where t is time information;

[0043] S3: Extract the BVP signal from the IPPG signal obtained in S2, which mainly includes the following steps;

[0044] S31: The original signal obtained by formula (1) It is divided into a set of segments, each 1.8 seconds long, connected end to end;

[0045] S32: The original signal after processing S31 The signal after a 1.8s sliding window truncation is represented in discrete form:

[0046] (2)

[0047] in, Represents discrete sampling k ( k The original signal of channels R, G, B, where n is the nth sampling point. , N The length of the original signal; In equation (1) ;in It is a multiplicative signal that is independent of time. It is a time-dependent multiplicative signal; Indicates the BVP signal. Some of these can be considered as weak "noise-like" values. The following data processing methods are used to fit and eliminate this "noise-like" data. ;

[0048] Using a cubic polynomial Fitting Part of, among which It is the mathematical expression of a cubic polynomial; to minimize the error of the fitting result, it is necessary to make... and Minimize the sum of squares of the residuals:

[0049] (3)

[0050] right Finding the differential and making it equal to 0, we obtain the following system of equations:

[0051] (4)

[0052] Formula 4 can be expressed in matrix form:

[0053] (5)

[0054] in: , , T denotes the transpose of the matrix; The matrix is ​​a Vandermonde matrix, for The matrix QR decomposition yields:

[0055] (6)

[0056] Where Q is an N×N unitary matrix and R is an N×4 upper triangular matrix; by obtaining the coefficients b of the cubic polynomial from equation (6), the fitting trend of the noise component can be obtained. ;

[0057] S33 weighted processing to obtain the BVP signal;

[0058] Further eliminate the information obtained in S32 Residual noise; based on the hemoglobin absorption spectrum, red light carries a large amount of noise information and a small amount of BVP information, green light carries a large amount of BVP information and a small amount of noise information, and blue light carries noise information and a trace amount of BVP information; after the original signal is fitted with polynomial noise and eliminated, a weighted signal processing method is used: the green channel signal is subtracted from the red channel signal, that is:

[0059] (7)

[0060] In the formula, This is the weighted blood flow pulsation signal. and The green and red channel signals of the BVP signal obtained by formula (6) are respectively;

[0061] S4: Using the polynomial fitting and weighting results obtained in step S3. The signals are used to calculate heart rate and blood oxygen saturation, thus completing the detection of human physiological signals.

[0062] In step S31, considering that the lowest human heart rate is around 50 bmp, which means that a complete heart pumping and diastolic activity takes about 1.2 seconds, the length of the signal cannot be less than 1.8 seconds. This ensures that three extreme points appear within a signal segment, so that the trend of BVP signal change can be fitted in a static state and the power of BVP signal can be preserved.

[0063] In step S1, the high bit depth camera is a camera with a bit depth of 8 bits or 10 bits.

[0064] In step S33, red light has a longer wavelength, i.e., 600-700nm, while green light has a shorter wavelength, i.e., about 500-560nm. In human skin tissue, red light penetrates deeper and is mainly absorbed by hemoglobin in the blood, especially deoxyhemoglobin. Green light penetrates shallower and is mainly absorbed in the epidermis and superficial dermis. Since the physiological activities in the superficial layer are more abundant, green light can more sensitively capture this superficial information because of its shallower penetration. Therefore, red light carries a large amount of noise information and a small amount of BVP information, while green light carries a large amount of BVP information and a small amount of noise information.

[0065] In step S2, the high signal-to-noise ratio region is: the forehead, that is, above the brow ridge and the forehead, which is a branch of the internal carotid artery; the sides of the nose and the tip of the nose, which is a major branch of the external carotid artery; the area from the angle of the mandible to the corner of the mouth, where the arteries directly receive the blood flow from the external carotid artery; and the area in front of the tragus, where there are branches of the maxillary artery, so the pulsation is more obvious.

[0066] High signal-to-noise ratio (SNR) regions need to be considered in conjunction with anatomy. The strength of facial blood pulsation is mainly related to the density of local arterial distribution, vessel diameter, and whether they directly branch into large arteries. Regions with relatively strong facial blood pulsation include: 1) the forehead (above the brow ridge and forehead), where the internal carotid artery branches; 2) the sides of the nose and the tip of the nose, where the external carotid artery is a major branch; 3) the area from the angle of the mandible to the corner of the mouth, where arteries directly receive blood flow from the external carotid artery; and 4) the area in front of the tragus, where the maxillary artery branches, resulting in more noticeable pulsation. These high SNR regions share the common characteristics of having arterial branches directly originating from large vessels such as the internal / external carotid arteries, being superficial, and having relatively large vessel diameters, making the pulsation more easily perceived on the skin surface. Other facial areas (such as the deep cheeks and inner eyelids) have relatively weaker pulsation due to deeper vessel locations or smaller vessel diameters. Therefore, based on facial feature nodes, these high SNR regions can be located, and the IPPG signals of these regions can be extracted for calculation.

[0067] Existing technologies require a relatively long facial video acquisition time to ensure the signal-to-noise ratio of the BVP signal. The typical minimum signal acquisition time is 20 to 30 seconds. However, in the application scenario of smart canteen systems, the real-time detection efficiency of human physiological signals needs to be comparable to the detection efficiency of automatic calculation of meal prices and analysis of meal nutritional components. The long detection time of physiological signals will greatly restrict the operating efficiency of smart canteen systems, thereby reducing the application value of human physiological signal detection.

[0068] This invention's technical solution is based on optical volumetric imaging technology using video stream data. It employs a high-depth camera sensitive to visible light to acquire facial video, and uses artificial intelligence and automated algorithms to extract facial feature points. The facial skin and vascular regions are segmented to extract pulse waveforms. The waveforms are analyzed and processed using short-time polynomial fitting, and the original signal is captured using a 1.8s sliding window (sliding one frame at a time). Figure 2 As shown, a short-time polynomial fitting optimization model is established by averaging the overlapping portion of the noise trend obtained after each sliding step, thereby achieving short-time, high-precision measurement of human physiological signals. This invention reduces the measurement time for non-contact human physiological signal detection and provides more application scenarios for non-contact human physiological signal detection based on optical video.

Claims

1. A method for rapid detection of human physiological signals, characterized in that, The steps are as follows, and they are performed in sequence: S1: Capture facial video. The image acquisition module uses a high-depth camera to record high-depth video data of facial scattered light signals. S2: Extract the human facial feature nodes from the high-depth video data obtained in step S1, and automatically divide the high signal-to-noise ratio region based on the blood vessel distribution of the human face, and extract the original IPPG signal, which can be represented as follows; （1） The variable reflecting the change of the IPPG signal over time is shown in the formula. is the grayscale value of the camera's RGB channel at the k-th pixel, and is a 3×N matrix where N is the length of t. It's noise. It is the vector of the blood flow pulse (BVP), reflecting the response intensity of the camera's RGB channels to the BVP signal. It is the intensity of the incident light. It is a BVP signal, where t is time information; S3: Extract the BVP signal from the IPPG signal obtained in S2, which mainly includes the following steps; S31: The original signal obtained by formula (1) It is divided into a set of segments, each 1.8 seconds long, connected end to end; S32: The original signal after processing S31 The signal after a 1.8s sliding window truncation is represented in discrete form: （2） in, Represents discrete sampling k ( k The original signal of channels R, G, B, where n is the nth sampling point. , N The length of the original signal; In equation (1) ;in It is a multiplicative signal that is independent of time. It is a time-dependent multiplicative signal; Indicates the BVP signal. Some of these can be considered as weak "noise-like" values. The following data processing methods are used to fit and eliminate this "noise-like" data. ; Using a cubic polynomial Fitting Part of, among which It is the mathematical expression of a cubic polynomial; to minimize the error of the fitting result, it is necessary to make... and Minimize the sum of squares of the residuals: （3） right Finding the differential and making it equal to 0, we obtain the following system of equations: （4） Formula 4 can be expressed in matrix form: （5） in: , , T denotes the transpose of the matrix; The matrix is ​​a Vandermonde matrix, for The matrix QR decomposition yields: （6） Where Q is an N×N unitary matrix and R is an N×4 upper triangular matrix; by obtaining the coefficients b of the cubic polynomial from equation (6), the fitting trend of the noise component can be obtained. ; S33 weighted processing to obtain the BVP signal; Further eliminate the information obtained in S32 Residual noise; based on the hemoglobin absorption spectrum, red light carries a large amount of noise information and a small amount of BVP information, green light carries a large amount of BVP information and a small amount of noise information, and blue light carries noise information and a trace amount of BVP information; after the original signal is fitted with polynomial noise and eliminated, a weighted signal processing method is used: the green channel signal is subtracted from the red channel signal, that is: （7） In the formula, This is the weighted blood flow pulsation signal. and The green and red channel signals of the BVP signal obtained by formula (6) are respectively; S4: Using the polynomial fitting and weighting results obtained in step S3. The signals are used to calculate heart rate and blood oxygen saturation, thus completing the detection of human physiological signals.

2. The method for rapid detection of human physiological signals according to claim 1, characterized in that: In step S1, the high bit depth camera is a camera with a bit depth of 8 bits or 10 bits.

3. The method for rapid detection of human physiological signals according to claim 1, characterized in that: In step S2, the high signal-to-noise ratio region is: the forehead, that is, above the brow ridge and the forehead, which is a branch of the internal carotid artery; the sides of the nose and the tip of the nose, which is a major branch of the external carotid artery; the area from the angle of the mandible to the corner of the mouth, where the arteries directly receive the blood flow from the external carotid artery; and the area in front of the tragus, where there are branches of the maxillary artery, so the pulsation is more obvious.

Images