A contactless vital sign estimation method and system based on FMCW millimeter wave radar

Abstract

The application belongs to the technical field of radar, and particularly relates to a non-contact vital sign estimation method and system based on an FMCW millimeter wave radar, which comprises the following steps: pre-processing a radar echo signal to extract a chest wall displacement signal therefrom; modeling the chest wall displacement signal as a combination of a fundamental frequency and harmonics thereof, establishing a likelihood function about a respiratory rate and a heart rate, constructing a target function of maximum likelihood estimation of a respiratory or heartbeat frequency, expressing the maximum likelihood estimation of the respiratory or heartbeat frequency as an optimization problem of the target function; and solving the optimization problem of the target function through a Newton iteration method to obtain maximum likelihood estimation values of the respiratory rate and the heart rate. The application fully utilizes harmonic energy in a signal, improves the accuracy and reliability of signal processing, and effectively reduces the interference of respiratory harmonics on heart rate estimation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of radar, and particularly relates to a non-contact vital sign estimation method and system based on FMCW millimeter wave radar. BACKGROUND

[0002] With the improvement of living standards, people's health consciousness is also getting higher and higher, and the non-contact vital sign detection technology based on radar is increasingly attracting people's attention. As the main basis for evaluating a person's health condition, the research on the technology of monitoring respiratory and heartbeat signals based on radar sensors is of great significance in medicine. Compared with other vital sign measurement sensors, radar sensors have unique advantages. Traditional vital sign detection methods mainly include electrocardiograph (ECG), phonocardiogram (PCG), airflow induction, etc. These methods are mainly applied to operating rooms and intensive care units in hospitals, and need to be in direct contact with the skin through electrodes, which can cause discomfort, skin damage or pressure necrosis to the patients, and is not conducive to the recovery of the patients. In some special scenarios, such as patients with severe burns, premature infants or patients with severe skin diseases, the skin barrier of special patients is fragile, and long-term contact with the equipment can cause damage to their skin, and even cause infection. In this case, patients cannot wear measurement equipment for a long time, which makes the traditional vital sign detection method unable to meet the needs of all-day detection. For the daily vital sign monitoring of bedridden old people or paralyzed patients, the discomfort caused by the contact type equipment makes it difficult for the target group to accept. The radar measurement system can be monitored for a long time, and can be directly applied to special patients (such as premature infants) without harming the patients. Therefore, it is necessary to study an accurate non-contact vital sign measurement method based on radar.

[0003] It is still a challenging task to achieve accurate non-contact vital sign monitoring, especially the monitoring related to heart rate estimation. The traditional method usually simply simulates the respiratory and heartbeat signals as the superposition of two sine waves, ignoring the harmonic components in the actual signal. The accuracy of respiratory estimation is usually limited by noise, while the heart rate estimation is mainly disturbed by the respiratory signal and its high-order harmonics. The existing technology for processing the respiratory harmonic interference to heart rate includes band-pass filtering and wavelet transform, etc. These methods can effectively suppress the harmonic signals caused by respiration and improve the accuracy of heart rate estimation. However, the effect of the band-pass filter is limited by the close frequency band range of the respiratory and heartbeat signals, and cannot effectively solve the interference problem. The wavelet transform can extract the respiratory and heartbeat signals, but it needs to select an appropriate wavelet basis function, which is a difficulty, and after selection, it cannot be flexibly adjusted according to different signals, lacking adaptability.

[0004] In summary, the prior art has technical challenges and limitations in dealing with the interference of respiratory harmonics on heart rate, and further research and improvement are needed to improve accuracy and practicality. SUMMARY

[0005] In order to solve the problems existing in the prior art, the present application proposes a non-contact vital sign estimation method and system based on FMCW millimeter wave radar, which constructs a respiratory and heartbeat model with harmonics, establishes a likelihood function about respiratory rate and heart rate, and derives the maximum likelihood estimation of respiratory rate and heart rate. The harmonic energy in the signal is fully utilized, the accuracy and reliability of signal processing are improved, and the interference of respiratory harmonics on heart rate estimation is effectively reduced.

[0006] A non-contact vital sign estimation method based on FMCW millimeter wave radar in an embodiment of the present application comprises the following steps:

[0007] S1, pre-processing the radar echo signal to extract the chest wall displacement signal therefrom;

[0008] S2, modeling the chest wall displacement signal as a combination of respiratory and heartbeat base frequencies and their harmonics, establishing a likelihood function about respiratory rate and heart rate, constructing a target function for maximum likelihood estimation of respiratory or heartbeat frequency, and representing the maximum likelihood estimation of respiratory or heartbeat frequency as an optimization problem of the target function;

[0009] S3, solving the optimization problem of the target function by Newton iteration method to obtain the maximum likelihood estimation value of respiratory rate and heart rate.

[0010] A non-contact vital sign estimation system based on FMCW millimeter wave radar in an embodiment of the present application comprises the following modules:

[0011] The pre-processing module is configured to pre-process the radar echo signal to extract the chest wall displacement signal therefrom;

[0012] The model construction module is configured to model the chest wall displacement signal as a combination of respiratory and heartbeat base frequencies and their harmonics, establish a likelihood function about respiratory rate and heart rate, construct a target function for maximum likelihood estimation of respiratory or heartbeat frequency, and represent the maximum likelihood estimation of respiratory or heartbeat frequency as an optimization problem of the target function;

[0013] The model solving module is configured to solve the optimization problem of the target function by Newton iteration method to obtain the maximum likelihood estimation value of respiratory rate and heart rate.

[0014] Compared with the prior art, the present application has the following technical effects:

[0015] 1. The present application utilizes the harmonic structure existing in the respiratory and heartbeat signals, and obtains the maximum likelihood estimation of the respiratory rate and the heart rate by constructing the respiratory and heartbeat models with harmonics and solving the maximum likelihood estimation function based on the Newton method. The present application does not need to separate the respiratory and heart rate signals by using a band-pass filter, but only needs to simply divide the search region and independently estimate in the respiratory and heart rate range, so that the maximum likelihood estimation of the respiratory rate and the heart rate can be derived. This method fully utilizes the harmonic energy in the signal, improves the accuracy and reliability of the signal processing, effectively reduces the interference of the respiratory harmonic on the heart rate estimation, and is expected to play an important role in the non-contact vital sign monitoring.

[0016] 2. The present application firstly eliminates the interference of the static strong reflector by the phasor cancellation method, realizes the correct estimation of the target distance, secondly correctly extracts the vital sign signal by using the arctangent method and phase unwrapping, and then enhances the energy of the signal by the slow-time phase correlation, subsequently constructs the respiratory and heartbeat models with harmonics, derives the maximum likelihood estimation of the respiratory rate and the heart rate, and finally solves the respiratory rate and the heart rate by the Newton iteration. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 It is a flow chart of the non-contact vital sign estimation method in the embodiment of the present application;

[0018] Figure 2 It is a distance spectrum diagram in the embodiment of the present application, wherein (a) is a schematic diagram before phasor cancellation, and (b) is a schematic diagram after phasor cancellation;

[0019] Figure 3 It is a chest wall displacement signal curve diagram after phase unwrapping in the embodiment of the present application;

[0020] Figure 4 It is a principle diagram of removing the motion trend by polynomial regression in the embodiment of the present application;

[0021] Figure 5 It is a curve diagram of removing the motion trend by polynomial regression in the embodiment of the present application;

[0022] Figure 6 It is a schematic diagram before and after the slow-time phase correlation processing in the embodiment of the present application, wherein (a) is a time domain comparison schematic diagram, and (b) is a frequency domain comparison schematic diagram;

[0023] Figure 7 It is a comparison schematic diagram of the respiratory rate estimation result and the reference sensor;

[0024] Figure 8 It is a comparison schematic diagram of the heart rate estimation result and the reference sensor. DETAILED DESCRIPTION

[0025] The application will be further described in connection with the embodiments and drawings below, but the embodiments of the application are not limited thereto

[0026] Embodiments

[0027] The embodiment provides a non-contact vital sign estimation method based on FMCW millimeter wave radar, as shown in the following formula (1) : Figure 1 The main steps include:

[0028] S1, pre-processing the radar echo signal to extract the chest wall displacement signal therefrom.

[0029] In the embodiment, the step specifically includes:

[0030] S11, performing mixing and filtering processing on the radar echo signal to obtain a radar intermediate frequency signal; the radar intermediate frequency signal is sampled by an analog-to-digital converter (ADC), and the sampled radar intermediate frequency signal can be expressed as:

[0031]

[0032] wherein A b and f IF are the amplitude and frequency of the radar intermediate frequency signal respectively, T f is a frame period, T s is a fast time sampling interval, λ is the wavelength of the radar transmitting signal, R0 is the distance from the human body target to the radar, and r(nT f ) represents the chest displacement at nT f ; n represents the nth radar intermediate frequency signal, also referred to as a slow time (velocity dimension) sampling point; and m represents the mth ADC sampling, also referred to as a fast time dimension (distance dimension) sampling point.

[0033] S12, eliminating the echo interference of the stationary obstacle by using a phasor cancellation method to obtain the radar intermediate frequency signal after eliminating the echo interference.

[0034] Specifically, the average of all the radar intermediate frequency signals is calculated, and the calculated average value is taken as a reference signal; then each received radar intermediate frequency signal is subtracted by the reference signal to eliminate the echo interference of the stationary obstacle and improve the ranging accuracy. The reference signal can be expressed as:

[0035]

[0036] wherein R[m,n] is a matrix of the radar intermediate frequency signal, m is a fast time dimension (distance dimension) sampling point, n is a slow time dimension (velocity dimension) sampling point, and N is the total number of the received radar intermediate frequency signals; the formula of the phasor cancellation algorithm is:

[0037] R a [m,n] = R[m,n]-C[m] (3)

[0038] wherein R a [m,n] is the matrix of the echo-canceled radar intermediate frequency signal.

[0039] The distance spectrum before and after phasor cancellation is shown in FIG. 2, wherein (a) is the distance spectrum before cancellation, and (b) is the distance spectrum after cancellation. Figure 2

[0040] S13, performing distance dimension fast Fourier transform on the echo-canceled radar intermediate frequency signal, and finding the peak value after fast Fourier transform to determine the distance of the human body target from the radar. The expression of the distance R0 of the human body target from the radar is:

[0041]

[0042] S14, demodulating the signal at the distance gate where the human body target is located by using the arctangent and phase unwrapping method to obtain the phase signal of the distance gate where the human body target is located, as the chest wall displacement signal caused by breathing and heartbeat.

[0043] In this step, the phase signal of the distance gate where the human body target is located is extracted on the one-dimensional distance spectrum diagram, and the extracted phase signal can be expressed as:

[0044]

[0045] wherein B I (t) is the in-phase branch of the phase signal; B Q (t) is the quadrature branch of the phase signal, A represents the amplitude of the phase signal, θ represents the fixed phase shift caused by the absolute distance between the human body target and the radar, δ(t) represents the chest wall displacement caused by breathing and heartbeat, and φ is the total phase noise generated by the system.

[0046] By using the orthogonal characteristics of the IQ two-way signal, a phase function containing the chest displacement can be obtained, so as to extract the phase information.

[0047]

[0048] Since the value range of the arctangent function is in the interval [-π, π], the phase value demodulated by the AD may be different from the actual phase. When some values in the actual phase Φ(t) exceed the interval, the phase wrapping phenomenon occurs, that is, the values demodulated by exceeding the range are still in the interval [-π, π], which is equivalent to generating a 2π jump at [-π, π]. In this embodiment, the phase unwrapping is performed by comparing the AD demodulated phase values of adjacent sampling points.

[0049]

[0050] ​where f(x) represents the unwrapped phase information, Φ n+1 represents the phase value at the n+1 sampling moment, Φ n represents the phase value at the n sampling moment.

[0051] The chest wall displacement signal curve obtained after phase unwrapping is shown in Figure 3 .

[0052] S15, subsequently, for the interference caused by slight human body shaking, a polynomial regression is used to remove the motion trend (i.e. remove the offset) of the chest wall displacement signal, eliminate the strong low-frequency component interference in the frequency domain, and improve the accuracy of subsequent parameter estimation.

[0053] The principle of removing the offset by polynomial regression is shown in Figure 4 , based on the k-order polynomial regression, the motion trend signal p(t) of the chest wall displacement signal δ(t) is calculated, and then the chest wall displacement signal δ(t) is subtracted from the motion trend signal p(t) to obtain the vital sign signal after removing the motion trend, i.e. the corrected phase signal y(t), as the chest wall displacement signal after removing the motion trend.

[0054] Figure 5 The signal curve diagram before and after removing the offset by polynomial regression is shown.

[0055] S16, the chest wall displacement signal after removing the motion trend is processed by slow-time phase correlation.

[0056] When the slow-time phase correlation processing is performed, the correlation between the chest wall displacement signals at the distance gate where the human target is located and the adjacent distance gate is analyzed, and the Pearson correlation coefficient thereof is calculated.

[0057]

[0058] where the human target is located at the dth distance gate, the ith distance gate is adjacent to the dth distance gate, and are the phase signal vectors (i.e. chest wall displacement signals) at the dth distance gate and the ith distance gate, respectively, represents the covariance between the phase signal vectors; is the corresponding standard deviation.

[0059] Figure 6 In the figure, subgraph (a) illustrates the time domain comparison before and after slow-time phase correlation, and subgraph (b) illustrates the frequency domain comparison before and after slow-time phase correlation.

[0060] When the Pearson correlation coefficient exceeds a predetermined threshold, the chest wall displacement signal is summed, and the signal-to-noise ratio is also improved accordingly. This allows the energy to be effectively utilized in the vicinity, so the estimation performance can be improved without increasing the computational burden.

[0061] S2, modeling the chest wall displacement signal as a combination of the respiratory and heartbeat base frequencies and their harmonics, establishing a likelihood function about the respiratory rate and the heart rate, constructing an objective function of the maximum likelihood estimation of the respiratory rate or the heartbeat frequency, representing the maximum likelihood estimation of the respiratory rate or the heartbeat frequency as an optimization problem of the objective function.

[0062] In the embodiment, step S2 comprises:

[0063] S21, constructing a respiratory and heartbeat signal model containing harmonics.

[0064] In most current studies, for the modeling and analysis of vital sign signals, the respiratory and heartbeat signals are generally simulated as the superposition of two sinusoidal signals with different amplitudes and frequencies. However, this does not conform to the actual situation. In fact, the respiratory and heartbeat signals are not single-frequency signals, but contain harmonic components and intermodulation components; that is, the respiratory and heartbeat signals both contain multiple harmonics, which are integer multiples of the respiratory and heartbeat base frequencies.

[0065] In order to utilize the harmonic structure in the respiratory and heartbeat signals, make full use of the harmonic energy, and reduce the interference of respiratory harmonics on the heart rate estimation, the embodiment constructs a respiratory and heartbeat signal model containing harmonics to model the chest wall displacement signal as a combination of the respiratory and heartbeat base frequencies and their harmonics. Through this signal model, solving in the frequency range of the respiratory and heartbeat respectively can more effectively capture the harmonic components of the respiratory and heartbeat signals, accumulate the harmonic energy and the base frequency energy, improve the accuracy and reliability of signal processing, and thus optimize the respiratory rate and heart rate estimation.

[0066] The respiratory and heartbeat signal model containing harmonics constructed in the embodiment can be expressed as:

[0067]

[0068] where f is the respiratory or heartbeat frequency, a1, a2 and a3 represent the amplitudes of the base, second harmonic and third harmonic of the respiratory or heartbeat respectively, and are the phases of the base, second harmonic and third harmonic of the respiratory or heartbeat respectively, and ω[n] represents a Gaussian white noise with a mean of 0 and a variance of σ 2 n represents the nth sampling time.

[0069] That is, the respiratory signal model and the heartbeat signal model in the embodiment are both expressed by the same formula (9), and the formula (9) represents the respiratory signal model when f is the respiratory frequency, and represents the heartbeat signal model when f is the heartbeat frequency.

[0070] S22, obtaining the probability density function of the to-be-estimated variable based on the constructed respiratory and heartbeat signal models.

[0071] According to the above signal model, the probability density function of the to-be-estimated variable can be obtained as follows:

[0072]

[0073] wherein is the to-be-estimated variable; x is a vector form of x[n].

[0074] S23, obtaining the estimated value of the to-be-estimated variable Θ by maximizing the likelihood function.

[0075] The maximization of the likelihood function can be equivalent to the minimization of the following formula:

[0076]

[0077] wherein A = [a1, a2, a3], J(A, f, Φ) is simplified into the following form:

[0078] J(A, f, Φ) = (x - a1c1 - a2s1 - β1c2 - β2s2 - γ1c3 - γ2s3) T × (x - a1c1 - a2s1 - β1c2 - β2s2 - γ1c3 - γ2s3) (12)

[0079] wherein:

[0080] α = [a1, a2, β1, β2, γ1, γ2] T

[0081] H = [c1, s1, c2, s2, c3, s3] T

[0082] a1 = a1cos(2πfn), a2 = -a1sin(2πfn)

[0083] β1 = a2cos(4πfn), β2 = -a2sin(4πfn)

[0084] γ1 = a3cos(6πfn), γ2 = -a3sin(6πfn)

[0085] c1 = [1, cos(2πf), …, cos(2πf(N-1))]T

[0086] s1 = [1, sin(2πf),..., sin(2πf(N-1))] T

[0087] c2 = [1, cos(4πf),..., cos(4πf(N-1))] T

[0088] s2 = [1, sin(4πf),..., sin(4πf(N-1))] T

[0089] c3 = [1, cos(6πf),..., cos(6πf(N-1)] T

[0090] s3 = [1, sin(6πf),..., sin(6πf(N-1))] T

[0091] The solution of minimizing equation (11) is:

[0092]

[0093] where I-H(H T H) -1 H T is an idempotent matrix and I denotes the identity matrix. Thus, in order to estimate the respiratory or heartbeat frequency f, one should compute the minimum of the cost function J based on f, i.e. equivalently one should get the maximum of the following expression:

[0094] x T H(H T H) -1 H T x

[0095] When the normalized frequency of f is not near 0 or 1 / 2, the maximum likelihood MLE estimate of f can be obtained as:

[0096]

[0097] S24. Constructing an objective function of a maximum likelihood estimate of a respiratory or heartbeat frequency, representing the maximum likelihood estimate of the respiratory or heartbeat frequency as an optimization problem of the objective function.

[0098] Let Constructing an objective function of a maximum likelihood estimate of a respiratory or heartbeat frequency f The maximum likelihood estimate (MLE) of the respiratory or heartbeat frequency f can be represented as an optimization problem of the objective function as follows:

[0099]

[0100] Thus, the maximum likelihood estimates of the respiratory rate and the heart rate are obtained by searching for the frequency value that maximizes the objective function within the range of the respiratory rate and the heart rate, respectively.

[0101] In order to provide sufficient signal length while maintaining the update rate of the respiratory rate and the heart rate estimates, the data is processed using overlapping sliding windows with a duration of 15 s, with an overlap of 14 s, so that the estimates of the respiratory and heart rates are updated every Δt = 1 s.

[0102] S3, solving the optimization problem of the objective function by the Newton iteration method to obtain the maximum likelihood estimate values of the respiratory rate and the heart rate.

[0103] This step solves the maximum likelihood estimate values of the respiratory rate and the heart rate by the Newton iteration method:

[0104]

[0105] where and are the first-order and second-order partial derivatives of the objective function , respectively, denotes the estimate value of the respiratory rate or the heart rate obtained in the kth iteration, and the estimate value obtained in the k+1th iteration is taken as the maximum likelihood estimate value.

[0106] In this embodiment, the maximum likelihood estimates of the respiratory rate and the heart rate share the formula (14), and the respiratory rate and the heart rate are obtained respectively by searching for the frequency value that maximizes the objective function within the range of the respiratory rate and the heart rate, respectively, as the maximum likelihood estimate values of the respiratory rate and the heart rate. Figure 7 The comparison between the respiratory rate estimation result of this embodiment and the reference sensor is shown; Figure 8 The comparison between the heart rate estimation result of this embodiment and the reference sensor is shown.

[0107] In addition, the embodiment also provides a non-contact vital sign estimation system based on an FMCW millimeter wave radar, which specifically comprises the following modules:

[0108] A preprocessing module is configured to preprocess the radar echo signal and extract a chest wall displacement signal therefrom;

[0109] A model construction module is configured to model the chest wall displacement signal as a combination of the fundamental frequency and its harmonics, establish a likelihood function about the respiratory rate and the heart rate, construct an objective function of the maximum likelihood estimation of the respiratory or heartbeat frequency, and express the maximum likelihood estimation of the respiratory or heartbeat frequency as an optimization problem of the objective function.

[0110] The model solving module solves the optimization problem of the objective function by a Newton iteration method to obtain maximum likelihood estimation values of the respiratory rate and the heart rate.

[0111] The above modules of the embodiment are respectively used for implementing corresponding steps of the non-contact vital sign estimation method, and the detailed implementation process can refer to the aforementioned steps S1-S3.

[0112] The above embodiment is a preferred embodiment of the present application, but the embodiment of the present application is not limited to the above embodiment, and any change, modification, replacement, combination, simplification, etc. made without departing from the spirit and principle of the present application should be an equivalent replacement mode and should be included in the protection scope of the present application.

Claims

1. A contactless vital sign estimation method based on FMCW millimeter wave radar, characterized by, The method comprises the following steps: S1, pre-processing the radar echo signal to extract the chest wall displacement signal therefrom; S2, modeling the chest wall displacement signal as a combination of the respiratory and heartbeat base frequencies and their harmonics, establishing a likelihood function with respect to the respiratory rate and the heart rate, constructing a target function for maximum likelihood estimation of the respiratory or heartbeat frequency, and representing the maximum likelihood estimation of the respiratory or heartbeat frequency as an optimization problem of the target function; S3, solving the optimization problem of the target function by Newton iteration to obtain the maximum likelihood estimation values of the respiratory rate and the heart rate; Step S2 comprises: S21, constructing a respiratory and heartbeat signal model containing harmonics to model the chest wall displacement signal as a combination of the base frequencies and their harmonics; Solving in the frequency range of the respiratory and heartbeat signals respectively through the respiratory and heartbeat signal model, capturing the harmonic components of the respiratory and heartbeat signals, and accumulating the harmonic energy and the base frequency energy; The respiratory and heartbeat signal model is expressed as: where f is the respiratory or heart beat frequency, a1, a2and a3represent the amplitudes of the fundamental, second and third harmonic of the respiratory or heart beat, respectively, and are the phases of the fundamental, second and third harmonic of the respiratory or heart beat, respectively, and ω[n] represents a Gaussian white noise with mean 0 and variance σ 2 n denotes the n-th sampling instant. When f is the respiratory frequency, x[n] represents the respiratory signal model; when f is the heartbeat frequency, x[n] represents the heartbeat signal model; S22, obtaining the probability density function of the to-be-estimated variable based on the constructed respiratory and heartbeat signal model: wherein represents the variable to be estimated; x is in vector form x[n]; N is the total number of received radar intermediate frequency signals.

2. The contactless vital signs estimation method of claim 1, wherein, Step S1 comprises: S11, mixing and filtering the radar echo signal to obtain a radar intermediate frequency signal; the radar intermediate frequency signal is sampled by an analog-to-digital converter; S12, eliminating the echo interference of stationary obstacles by using a phasor cancellation method to obtain a radar intermediate frequency signal after echo interference elimination; S13, performing a range dimension fast Fourier transform on the radar intermediate frequency signal after echo interference elimination to find the peak value after fast Fourier transform to determine the distance from the human body target to the radar; S14, demodulating the signal at the distance gate where the human body target is located by using the arctangent and phase unwrapping methods to obtain the phase signal of the distance gate where the human body target is located as the chest wall displacement signal caused by the respiration and heartbeat; S15, removing the motion trend of the chest wall displacement signal by using a polynomial regression to eliminate the strong low-frequency component interference in the frequency domain; S16, performing slow-time phase correlation processing on the chest wall displacement signal after the motion trend is removed.

3. The contactless vital signs estimation method of claim 2, wherein, In step S14, the phase signal of the distance gate where the human body target is located is extracted on a one-dimensional distance spectrum, and the extracted phase signal is expressed as: where B I (t) is the in-phase branch of the phase signal; B Q (t) is the quadrature branch of the phase signal, A denotes the amplitude of the phase signal, Θ denotes the fixed phase shift resulting from the absolute distance between the human target and the radar, δ(t) denotes the chest wall displacement caused by breathing and heartbeat, and φ is the total phase noise generated by the system; The phase information is extracted by using the orthogonal characteristics of the IQ two-way signal to obtain a phase function containing the chest displacement: Phase unwrapping is performed by comparing the AD demodulation phase values of adjacent sampling points: where f(x) represents unwrapped phase information, Φ n+1 denotes the phase value at the (n+1)th sampling instant, Φ n denotes the phase value at the nth sampling instant.

4. The contactless vital signs estimation method of claim 2, wherein, In step S16, the slow-time phase correlation processing is performed, the correlation between the chest wall displacement signals at the distance gate where the human body target is located and the adjacent distance gate is analyzed, and the Pearson correlation coefficient thereof is calculated: where the human target is located at the dth distance gate, the ith distance gate is adjacent to the dth distance gate, and are the phase signal vectors at the dth distance gate and the ith distance gate, respectively, denotes the covariance between the phase signal vectors; is the corresponding standard deviation.

5. The contactless vital signs estimation method of claim 1, wherein, Step S2 further comprises: S23, obtaining the estimation value of the to-be-estimated variable Θ by maximizing the likelihood function; the maximized likelihood function is equivalent to minimizing J(A,f,Φ): where A = [al, a2, a3], J(A, f, Φ) is simplified to: J(A,f,Φ) = (x - α1c1- α2s1- β1c2- β2s2- γ1c3- γ2s3) T ×(x - α1c1- α2s1- β1c2- β2s2- γ1c3- γ2s3) Wherein: a = [a1, a2, b1, b2, g1, g2] T H = [c1,s1,c2,s2,c3,s3] T α1=a1cos(2πfn),α2=-a1sin(2πfn) β1=a2cos(4πfn),β2=-a2sin(4πfn) Y1 = a3 cos(6πfn), y2 = -a3 sin(6πfn) c1=[1,cos(2πf),…,cos(2πf(N-1))] T s1=[1,sin(2πf),…,sin(2πf(N-1))] T c2=[1,cos(4πf),…,cos(4πf(N-1))] T s2=[1,sin(4πf),…,sin(4πf(N-1))] T c3=[1,cos(6πf),…,cos(6πf(N-1))] T s3=[1,sin(6πf),…,sin(6πf(N-1))] T The solution of minimizing J(A,f,Φ) is: where I-H(H T H) -1 H T is an idempotent matrix and I denotes the identity matrix; in order to estimate the respiratory or heartbeat frequency f, the maximum of the following expression should be obtained: x T H(H T H) -1 H T x When the normalized frequency of the respiratory or heartbeat frequency f is not near 0 or 1 / 2, the maximum likelihood estimation of f is obtained: S24, let the maximum likelihood estimate of the breathing or heartbeat frequency f be constructing an objective function for the maximum likelihood estimate of the breathing or heartbeat frequency representing the maximum likelihood estimate of the breathing or heartbeat frequency as an optimization problem of the objective function.

6. The contactless vital signs estimation method of claim 5, wherein, The step S3 solves the maximum likelihood estimation values of the respiratory rate and the heartbeat rate by the Newton iteration method: wherein and are the first and second derivatives of the objective function respectively, denotes the estimate of the respiration rate or heart rate obtained at the kth iteration, with the estimate obtained at the k+1th iteration being the maximum likelihood estimate.

7. A contactless vital sign estimation system based on FMCW millimeter wave radar, characterized by, The method comprises the following modules: The preprocessing module is used for preprocessing the radar echo signal, and extracting the chest wall displacement signal therefrom; The model construction module is used for modeling the chest wall displacement signal as a combination of the base frequency and its harmonics, establishing a likelihood function about the respiratory rate and the heartbeat rate, constructing an objective function of the maximum likelihood estimation of the respiratory or heartbeat frequency, and representing the maximum likelihood estimation of the respiratory or heartbeat frequency as an optimization problem of the objective function; The model solving module solves the optimization problem of the objective function by the Newton iteration method, and obtains the maximum likelihood estimation values of the respiratory rate and the heartbeat rate; The model construction module comprises: The respiratory and heartbeat signal model construction module is used for constructing a respiratory and heartbeat signal model containing harmonics, so as to model the chest wall displacement signal as a combination of the base frequency and its harmonics; the respiratory and heartbeat signal model is used for solving in the frequency range of the respiratory and heartbeat respectively, so as to capture the harmonic components of the respiratory and heartbeat signals, and accumulate the harmonic energy and the base frequency energy; The respiratory and heartbeat signal model is represented as: where f is the respiratory or heart beat frequency, a1, a2and a3represent the amplitudes of the fundamental, second and third harmonic of the respiratory or heart beat, respectively, and are the phases of the fundamental, second and third harmonic of the respiratory or heart beat, respectively, and ω[n] represents a Gaussian white noise with mean 0 and variance σ 2 n denotes the n-th sampling instant. When f is the respiratory frequency, x[n] represents the respiratory signal model; when f is the heartbeat frequency, x[n] represents the heartbeat signal model; The to-be-estimated variable obtaining module is used for obtaining the probability density function of the to-be-estimated variable based on the constructed respiratory and heartbeat signal model: wherein represents the variable to be estimated; x is in vector form x[n]; N is the total number of received radar intermediate frequency signals.

8. The contactless vital signs estimation system of claim 7, wherein, The model construction module further comprises: The maximum likelihood estimation module obtains the estimation value of the to-be-estimated variable Θ by maximizing the likelihood function; and the maximum likelihood function is equivalent to minimizing J(A,f,Φ): where A = [al, a2, a3], J(A, f, Φ) is simplified as: J(A,f,Φ) = (x - α1c1- α2s1- β1c2- β2s2- γ1c3- γ2s3) T ×(x - α1c1- α2s1- β1c2- β2s2- γ1c3- γ2s3) Wherein: a = [a1, a2, b1, b2, g1, g2] T H = [c1,s1,c2,s2,c3,s3] T a1 = a1 cos(2πfn), a2 = -a1 sin(2πfn) β1 = a2 cos(4πfn), β2 = -a2 sin(4πfn) γ1 = a3 cos(6πfn), γ2 = -a3 sin(6πfn) c1=[1,cos(2πf),…,cos(2πf(N-1))] T s1=[1,sin(2πf),…,sin(2πf(N-1))] T c2=[1,cos(4πf),…,cos(4πf(N-1))] T s2=[1,sin(4πf),…,sin(4πf(N-1))] T c3=[1,cos(6πf),…,cos(6πf(N-1))] T s3=[1,sin(6πf),…,sin(6πf(N-1))] T The solution of minimizing J(A,f,Φ) is: where I-H(H T H) -1 H T is an idempotent matrix, I denotes the identity matrix; in order to estimate the respiratory or heartbeat frequency f, the maximum of the following expression should be obtained: x T H(H T H) -1 H T x When the normalized frequency of the respiratory or heartbeat frequency f is not near 0 or 1 / 2, the maximum likelihood estimation of f is obtained: a target function that constructs a maximum likelihood estimate of the breathing or heartbeat frequency f as a target function that constructs a maximum likelihood estimate of the breathing or heartbeat frequency a maximum likelihood estimate of the breathing or heartbeat frequency as an optimization problem of a target function.

9. The contactless vital signs estimation system of claim 8, wherein, The model solving module solves the maximum likelihood estimation values of the respiratory rate and the heartbeat rate by the Newton iteration method: wherein and are the first and second derivatives of the objective function respectively, denotes the estimate of the respiration rate or heart rate obtained at the kth iteration, and the estimate obtained at the k+1th iteration is the maximum likelihood estimate.

10. The contactless vital signs estimation system of claim 7, wherein, The process in which the preprocessing module preprocesses the radar echo signal comprises: The radar echo signal is mixed and filtered to obtain a radar intermediate frequency signal; the radar intermediate frequency signal is sampled by an analog-to-digital converter; The echo interference of the stationary obstacle is eliminated by using the phasor cancellation method, to obtain a radar intermediate frequency signal after echo interference elimination; The radar intermediate frequency signal after echo interference elimination is subjected to range dimension fast Fourier transform, and a peak value after fast Fourier transform is searched to determine the distance from the human body target to the radar; The signal at the distance gate where the human body target is located is demodulated by using the arctangent and phase unwrapping methods, to obtain a phase signal of the distance gate where the human body target is located, as the chest wall displacement signal caused by the respiration and the heartbeat; The method comprises the following steps: S1, preprocessing the radar echo signal to extract the chest wall displacement signal; S2, modeling the chest wall displacement signal as a combination of the base frequency and its harmonics, and establishing a likelihood function about the respiratory rate and the heartbeat rate; S3, constructing an objective function of the maximum likelihood estimation of the respiratory or heartbeat frequency, and representing the maximum likelihood estimation of the respiratory or heartbeat frequency as an optimization problem of the objective function; S4, solving the optimization problem of the objective function by the Newton iteration method, and obtaining the maximum likelihood estimation values of the respiratory rate and the heartbeat rate. The chest wall displacement signal is processed by polynomial regression to remove motion trend, so as to eliminate the strong low-frequency component interference in the frequency domain; The chest wall displacement signal after removing motion trend is processed by slow-time phase correlation.

Images