Method for multi-path echo discrimination and personnel detection based on life signal fingerprint characteristics

Abstract

The application discloses a method for multi-path echo discrimination and personnel detection based on life signal fingerprint characteristics, which comprises the following steps: separating echo signals of each target from radar receiving signals, decomposing the frequency phase signals in the echo into several IMF modes by using a successive variation mode decomposition algorithm, and reconstructing breathing signals and heartbeat signals from the IMF modes. Furthermore, based on the multi-target breathing and heartbeat signal fingerprint feature extraction of the millimeter wave radar, the multi-target personnel is distinguished and detected according to the fingerprint feature similarity of the multi-path false target and the corresponding real target, and the multi-path false target is discriminated. The application does not need prior information of the environment, and does not need to measure the size of the environment later, so that the scene limitation is small. The application can be applied to complex indoor scenes, and has high practicability. In the multi-person vital sign detection, the application is effective, and only a single radar is needed to remove the multi-path, so that the whole detection system is simplified. The method does not need to change the configuration of the radar array and other system levels.

Description

Technical Field

[0001] The present invention relates to the field of indoor personnel health monitoring, and in particular to a multipath echo identification and personnel detection method based on life signal fingerprint characteristics. Background Art

[0002] The research and application of non-contact radar-based physiological indicator systems have made significant progress, providing important technical support for fields such as healthcare, health monitoring, and sports science. These systems primarily rely on actively radiated pulse signals, such as millimeter-wave and ultra-wideband radars. These signals, reflected from the chest cavity of living organisms like humans and pets, carry micro-motion information about vital activities such as breathing and heartbeat. Through time-domain or frequency-domain signal processing algorithms, these vital signs, such as heart rate and respiratory rate, can be captured. However, a major challenge facing current radar-based physiological indicator detection systems is multipath propagation. Multipath refers to the phenomenon in which signals from wireless sensors such as millimeter-wave radars, due to obstacles or environmental reflections, generate multiple echoes captured by the receiver. These multiple echoes can lead to the appearance of multipath false targets, complicating the detection and identification of individuals and severely impacting the multi-target vital sign monitoring process.

[0003] Fingerprints of life signals are unique features formed by the diverse life signals of an organism, resulting from individual differences. Due to variations in body shape and chest structure, captured life signals contain a wealth of individual characteristics. Their uniqueness and stability offer new insights into human detection and identification.

[0004] Currently, radar's high range resolution makes it possible to distinguish multipath echoes from true target echoes. Distinguishing true targets from multipath false targets is a critical technical challenge in multi-target personnel detection and physiological indicator monitoring. False targets are misidentified as vital targets due to the multipath echo effect. These false targets share similarities with their true targets in the vital echo signal, which can be reflected in the fingerprint characteristics of the vital echo signal. Distinguishing between different true targets and distinguishing between true targets and their corresponding radar multipath false targets is a key approach to improving the accuracy and reliability of multi-target vital sign detection.

[0005] In the prior art, methods for suppressing multipath include:

[0006] Existing technology 1: Multipath suppression algorithm based on radar signal propagation model

[0007] By analyzing the propagation paths of electromagnetic waves using information from target detection experimental scenarios, a radar signal propagation model is constructed. Multipath elimination is achieved by exploiting the differences in propagation paths between real targets and multipath false targets.

[0008] This method generally offers high accuracy, but it is highly dependent on the environment. In real life, experimental scenarios can change over time, such as furniture placement changes or curtains swaying in the wind. This can alter the radar signal propagation path, requiring further analysis and modeling, which limits the method's applicability. Furthermore, this method requires prior information about the environment.

[0009] Existing technology 2: Multipath suppression algorithm based on machine learning and deep learning network

[0010] Machine learning-based multipath suppression algorithms manually extract features from radar signals and use machine learning algorithms to classify real targets and multipath ghosts. Deep learning-based data suppression algorithms don't require manual feature extraction, but they require a large number of samples and have lower recognition accuracy. Summary of the Invention

[0011] The technical problem to be solved by the present invention is: Based on the multipath ghosts generated by the signal in the complex indoor environment during the non-contact life feature detection by millimeter wave radar, the present invention proposes a multipath echo identification and personnel detection method based on the fingerprint characteristics of life signals. The method mainly includes separating the echo signal of each target from the radar received signal, using the Sequential Variational Mode Decomposition (SVMD) algorithm to decompose the echo intermediate frequency phase signal into several IMF modes, and reconstructing the breathing signal and heartbeat signal therefrom. Then, the body characteristics, breathing and heartbeat fingerprint characteristics of the human body are extracted, the correlation between the signal characteristics is solved, and each real target and its corresponding multipath are associated to obtain a target set containing each real target. Finally, the time delay difference between the real target and the multipath false target is used to suppress the multipath and obtain the breathing and heartbeat information of the real target.

[0012] The technical solution adopted in the present invention is:

[0013] Step 1: Obtain radar echo signals of multiple stationary targets in a space. The intermediate frequency echo signals contain not only the target signal of the human chest, which is caused by weak vibrations, but also reflections from static objects such as walls and stationary parts of the human body. Therefore, the mean of the intermediate frequency echo signals of each slow-time frame can be used as an estimate of static clutter. The static clutter is then eliminated using a mean cancellation algorithm.

[0014] Step 2: The radial distance of each target is obtained by performing FFT on the intermediate frequency signal along the fast time dimension, and the minimum variance undistorted response Capon algorithm is used to estimate the azimuth of the target. This algorithm minimizes the variance of the echo signal in the target direction angle, thereby maximally suppressing the noise power.

[0015] Step 3: After eliminating static clutter, the detection environment of this application can be considered a uniform clutter environment. After steps 1 and 2, a two-dimensional range-direction angle diagram is obtained. CASO-CFAR detection is performed on the range and angle dimensions respectively using one-dimensional unit average. The minimum value in the reference unit is selected as the reference background noise for multi-target extraction.

[0016] Step 4: By weighted superposition of the received signals of each channel, the echo signal in a certain direction is enhanced while the energy of the signals in other directions is suppressed, and the target signal is separated using the minimum variance distortionless response (MVDR) beamforming algorithm. This algorithm can enhance the echo signal in a certain direction while suppressing the echoes in other directions. Under the premise of satisfying the distortion-free constraint of the target echo signal, it finds the optimal weight to minimize the output power and ensure the minimization of clutter noise.

[0017] Step 5: After obtaining the number of suspected targets and the intermediate frequency signals of each target's individual echoes, direct inverse tangent demodulation will cause the phase to collapse within the range [-π, π), resulting in confusion. Further phase unwrapping is required to eliminate phase discontinuities. Therefore, differential cross-multiplication is used to extract the phase signal, which contains the target's breathing and heartbeat signals.

[0018] Step 6: Decompose and reconstruct the target breathing signal and heartbeat signal according to the frequency difference between the human breathing and heartbeat signals;

[0019] Step 7: Perform radar detection on static human targets. By analyzing the target echo y(t), the body characteristics, breathing characteristics, and heartbeat characteristics of the human body are extracted to ensure accurate capture of human physiological indicators. All eigenvalues ​​are summarized to obtain the feature matrix corresponding to each target, and the multivariate feature values ​​of each target are integrated.

[0020] Step 8: By accurately calculating the correlations between different targets, a correlation matrix is ​​obtained. The average correlation between targets is obtained by taking the mean of the correlation matrix. In this process, targets include real targets and false targets caused by multipath. Targets with high correlations are considered to be the same target and its corresponding false multipath target, while targets with low correlations are considered to be different targets. Based on this judgment criterion, multiple target sets can be successfully obtained.

[0021] Step 9: After the personnel identity features are associated, the propagation delay differences caused by the differences in the signal propagation paths during the propagation process are used to distinguish the true target and the multipath propagation delay differences in each target set; the target closest to the radar is selected to determine the true target, ultimately achieving effective identification of the true target in multi-person scenarios.

[0022] To further illustrate the technical solution of the present invention, it is now described in detail as follows:

[0023] Step 1: Obtain radar echo signals of multiple target personnel in a static state in space and eliminate static clutter by using the mean cancellation method.

[0024] The FMCW (Frequency Modulated Continuous Wave) millimeter wave radar system transmits a periodic frame signal. Each frame signal contains a chirp signal transmitted by multiple transmitting antennas TX. Its instantaneous frequency changes linearly with time. The initial carrier frequency is f c , the linear frequency modulation slope is γ.

[0025]

[0026] The radiated electromagnetic signal is reflected by the outside world and received by multiple radar antennas RX. The received signal S R (t) There is a time delay τ = 2R / c, where R is the radial distance between the radar and the reflecting target. After passing through the receiver and mixer, an intermediate frequency signal modulated with human chest micro-motion information can be obtained.

[0027]

[0028] The distance information between the reflector and the radar is reflected in the time domain delay τ of the received signal and the frequency value f of the intermediate frequency signal. IF Therefore, the intermediate frequency signal f can be obtained by distance FFT IF The radial distance coordinates of the target can be inferred from the spectrum peak.

[0029] The radar's captured echo signals also include static targets and its own TX leakage signal. The reflected signal from the static target not only obscures the presence of living objects but also generates a DC offset that interferes with subsequent target phase extraction and demodulation. To estimate the DC component of each range unit, the echo values ​​for each channel and range unit in each frame must be averaged.

[0030]

[0031] Where Y[u,m,n] represents the echo signal of the mth range unit n frames of the uth antenna channel, and N is the total number of frames.

[0032] Step 2: Target distance estimation and angle estimation.

[0033] As can be seen above, the IF signal spectrum peaks of targets at different radial distances differ. The radial distance of each target can be determined by performing an FFT along the fast time dimension of the IF signal. However, targets at the same radial distance but at the same angle have overlapping IF signal spectrum peaks, making them indistinguishable. Angle estimation is primarily based on the time delay between the RX antennas, 2Δxsinθ, which causes the same target signal to reach different RXs. This delay is small compared to the echo delay caused by the target's radial distance R, so it is reflected in the phase information of the target echo signal. Generally, an FFT can be performed on the angular dimension of the echo data to obtain the angular frequency value ω, and then the target azimuth angle θ. However, this has low resolution and depends on the number of antenna channels.

[0034]

[0035] The minimum variance distortionless response Capon algorithm is a high-resolution angle measurement algorithm that primarily satisfies the maximum signal-to-interference-and-noise ratio criterion. Under the constraints of the minimum variance criterion, it seeks the optimal weights for the array output to ensure proper reception of signals from the target of interest while suppressing interference from signals in other directions. The objective function of the optimization problem is the spatial spectrum P(θ), where α(θ) is the steering vector, which is related to the number of receiving antenna channels, and R is the covariance matrix of the intermediate frequency signal matrix received by the antenna array.

[0036]

[0037] According to formula (5), the spectrum peak search is performed, and the angle corresponding to the peak is the target wave arrival direction.

[0038] Step 3: Target detection and distance and direction angle positioning.

[0039] Constant False Alarm Rate (CFAR) detection technology is a commonly used and efficient target detection technology for radar systems. After eliminating static clutter, the detection environment of this application can be considered a uniform clutter environment. CFAR detection is performed by taking the minimum CFAR value of one-dimensional unit average in the distance and angle dimensions respectively. That is, the minimum value in the reference unit is selected as the reference background noise for multi-target extraction.

[0040] Step 4: Separate and extract multi-target intermediate frequency signals using the beamforming algorithm.

[0041] Because each radar receiving antenna (TX) simultaneously captures echo signals from all targets, echo signals from targets at the same radial distance but in different directions will overlap, interfering with the extraction of target vital signals. Therefore, a beamforming algorithm is required to separate target signals. This algorithm performs a weighted superposition of the received signals from each channel, thereby enhancing echo signals from certain directions and suppressing signal energy from other directions.

[0042] According to the distance and azimuth information of all suspected targets detected in step 3, the unique weighting coefficient ω of channel u is calculated for each target i. u (θ i ), and accumulate the intermediate frequency signals of the U channel echoes to obtain the individual intermediate frequency signal z(t) of each target echo.

[0043]

[0044] Step 5: DACM phase unwrapping to obtain the target life signal.

[0045] Through the above four steps, the number of suspected targets in the detection environment and the intermediate frequency signals of the individual echoes of the targets have been obtained, from which the phase signal φ(t) is extracted, which includes the target's breathing signal x b (t) and heartbeat signal x h (t).

[0046]

[0047] Among them, the intermediate frequency signal frequency f IF The radial distance R0 from the target is constant, and only the life signal x b (t) and x h (t) changes with time, which reflects the changing trend of the target's breathing and heartbeat over time. In order to obtain the phase information of the intermediate frequency signal and then extract the vital signal waveform, the common I / Q channel inverse tangent demodulation has a phase folding problem and poor noise resistance. The Differential and Cross Multiply (DACM) algorithm has a certain noise suppression capability and does not have discontinuities caused by the phase folding problem. The DACM of the discrete signals of the I and Q channels after sampling is specifically expressed as:

[0048]

[0049] Step 6: Decomposition and reconstruction of target breathing and heartbeat signals.

[0050] Currently, there are numerous algorithms for separating vital signs. Due to the frequency differences between respiratory and heartbeat signals, bandpass filtering can be used to initially separate and reconstruct them. However, the stronger harmonics of the respiratory signal can mask the weaker heartbeat signal. Therefore, wavelet transforms, empirical mode decomposition (EMD), variational mode decomposition (VMD), and their derivative algorithms are widely used in the field of vital sign separation and reconstruction.

[0051] The SVMD algorithm is essentially a set of multiple adaptive Wiener filter banks that can adaptively decompose life signals into K modal components (IMFs) while avoiding the modal aliasing and boundary effects caused by the EMD algorithm. It transforms the optimization problem of K modal components into K single-modal optimization problems, resulting in better convergence.

[0052]

[0053] The life signal g(t) is decomposed by SVMD to obtain K-1 modal components IMF i (t) and the high-frequency residual signal component g r (t). Then, according to the spectrum range of human breathing and heartbeat, appropriate IMF components are extracted and superimposed in the time domain to reconstruct the breathing signal x. b (t) and heartbeat signal x h (t).

[0054] Step 7: Extract the life signal fingerprint features of all suspected targets (including multipath ghosts).

[0055] When performing radar detection on static human targets, by utilizing the target's micro-motion characteristics to achieve target correlation matching, a target set containing each real target can be obtained, thereby suppressing the multipath within the target set and obtaining the real target.

[0056] By analyzing the target echo y(t), we extract the body's physical characteristics, including the signal envelope's mean, standard deviation, root mean square value, skewness, kurtosis, and crest factor. These six characteristics effectively reflect the uniqueness of the body's shape, enabling us to distinguish different people.

[0057] On the other hand, radar technology can be used to identify individuals based on respiratory and heartbeat characteristics. Because a complete respiratory / heartbeat cycle is a nonlinear dynamic process, a series of features are selected, including respiratory rate, heart rate, number of respiratory signal peaks, number of heartbeat signal peaks, respiratory signal peaks, heartbeat signal peaks, duration of exhalation and inspiration, and chest contraction and relaxation strengths. Extracting these ten features from the target echo y(t) helps to better identify individual differences and accurately identify different people.

[0058] Count all features of the same target i into the feature matrix

[0059] F i =[M 1i ,M 2i ,...M 16i ,] (10)

[0060] Thus, the feature matrices F1, F2, ..., F corresponding to different targets are obtained. NT, where i∈[1,N T ],N T is the number of detected targets, including the number of real targets and false targets.

[0061] Step 8: Associate the identity features of all suspected targets (including multipath ghosts).

[0062] Calculate the correlation matrix C between the i-th target and the j-th target ij for

[0063] C ij =[cov 11 ,cov 22 ,...,cov 1616 ]

[0064]

[0065] Calculate the correlation between the i-th target and the j-th target,

[0066]

[0067] The correlation matrix of all targets is obtained as

[0068]

[0069] The ones with large correlation are the same target and its corresponding multipath false target, and the ones with small correlation are different targets. Thus, multiple target sets T are obtained. i ,

[0070]

[0071] Among them, i∈[1,N c ],N c is the target set number. i is the real target, G 1i ,G 2i ,...,G Ngi is a multipath false target, N g is the number of multipaths.

[0072] Step 9: Multipath false target suppression.

[0073] During signal propagation, since the signal may pass through different paths, each path has a different length, the time it takes for the signal to reach the radar receiver will also vary. Multipath is caused by reflections from other surfaces such as walls and obstacles during propagation. Its propagation path is longer than that of the real target and has a greater propagation delay. Therefore, the target closest to the radar is selected as the real target R. i , and finally get the real target in the multi-person scenario

[0074] R=[R1,R2,...,R Nc ] (15)

[0075] The multipath echo identification and personnel detection method based on the fingerprint characteristics of life signals proposed in the present invention has significant advantages over the existing technologies in terms of system implementation difficulty and correction accuracy, mainly including: (1) no prior information of the environment is required, nor is subsequent measurement of the environment size required, and the scene restrictions are small; (2) it is applicable to complex indoor scenes and has strong practicality; (3) it is effective in multi-person vital sign detection; (4) only a single radar is needed to remove multipath, which simplifies the entire detection system; (5) this method suppresses multipath at the software algorithm stage, and does not require changes to the configuration at the system level such as the radar array. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 Schematic diagram of multipath effects in radar detection of personnel vital signs.

[0077] Figure 2 Schematic diagram of millimeter-wave radar multi-target life signal fingerprint feature extraction and multipath false target.

[0078] Figure 3(a) is the distance-time diagram of the millimeter-wave radar echo signal before static clutter elimination.

[0079] Figure 3(b) is the distance-time diagram of the millimeter-wave radar echo signal before static clutter elimination.

[0080] Figure 4 Schematic diagram of the angle measurement principle of the multi-RX radar system.

[0081] Figure 5(a) is a radial range-direction angle multi-target detection diagram.

[0082] Figure 5(b) shows the multi-target detection results after CASO-CFAR detection.

[0083] Figure 6(a)-(e) are the waveforms of the target breathing and heartbeat signals obtained by SVMD decomposition.

[0084] Figure 7 It is the target result graph of fingerprint feature similarity matching.

[0085] Figure 8 This is the target radial range-lateral range diagram after multipath false target suppression. DETAILED DESCRIPTION

[0086] The following combination Figure 1-8 And specific embodiments further illustrate the technical solutions of the present invention.

[0087] Based on the extracted multi-target radar life signal fingerprint features, the present invention proposes a multipath ghost identification and personnel detection method. Figure 1 As shown in Figure 1, the multipath effect produces ghost targets, which interferes with human target detection and affects physiological index monitoring. The identification of multipath echoes is of great significance. Figure 2 As shown, it mainly includes two aspects: multi-target personnel detection and signal separation, life signal fingerprint feature extraction and multipath identification. The specific steps are as follows:

[0088] Step 1: Eliminate static clutter in multi-channel radar echoes.

[0089] Based on a multi-transmitter, multi-receiver FMCW radar system, each transmitting antenna (TX) and receiving antenna (RX) form a set of channels, each of which captures target reflections from all directions. The radar used has three transmitting antennas and four receiving antennas, with eight channels used for horizontal azimuth measurement.

[0090] Multiple transmitting antennas TX radiate linear frequency modulation (LFM) signals in sequence with a small time difference within each frame. The instantaneous frequency of the signal changes linearly with time. Its time domain expression is:

[0091]

[0092] Among them, the initial carrier frequency is f c =77GHz, the frequency modulation slope is γ=60.012MHz / μs, that is, the frequency change rate of the LFM signal, is the initial phase. The signal reflected back after the transmitted signal encounters the target in space can be modeled as the time delay signal S of the transmitted signal T (t-τ), where the time delay τ = 2R / c, R is the radial distance between the target and the radar, and the speed of light is 3×10 8 m / s.

[0093] After all TX transmission signals are reflected, they are captured by each RX separately. After mixing with the transmission signal and its -90° phase shift signal in the super receiver, the I and Q channel low-frequency components of the frequency difference between the transmission signal and the reception signal are obtained, namely the intermediate frequency signal y(t).

[0094]

[0095] Among them, it is assumed that the echo strength of the transmitted signal and the received signal are A T =1 and A R = 1. The frequency of the intermediate frequency signal reflects the radial distance information R of the target.

[0096] f IF=γτ=2γR / c (18)

[0097] The echo intermediate frequency signal contains not only the human target signal caused by the weak vibration of the chest cavity, but also the reflected echoes from static objects such as walls and stationary parts of the human body. The time delay caused by these echoes is very stable, and the echo intermediate frequency signal does not change with time. However, the echo delay caused by the human chest cavity will have slight changes, and the phase change can be considered to follow a uniform distribution U(0,2π). Therefore, the mean value of the echo intermediate frequency signal of each frame can be used as an estimate of static clutter. After A / D conversion, the static clutter elimination processing of the discrete echo intermediate frequency signal can be expressed as:

[0098]

[0099] Where Y[u,m,n] represents the echo signal from the mth range cell of antenna channel u in the nth frame. N is the total number of frames, representing the total measurement time. This operation is performed separately for the I and Q channel echoes. The range-time plots of the echo signals before and after static clutter removal are shown in Figures 3(a) and 3(b). Each test collects 1000 frames of radar echo signals, with each frame containing one chirp, and each chirp is sampled at 128 points.

[0100] Step 2: Target distance estimation and angle estimation.

[0101] The radial distance R of the target containing micro-motion information can be estimated by the frequency of the echo intermediate frequency signal. However, there may be multiple targets located at different direction angles θ (the angle between the echo direction and the normal of the receiving antenna plane) but with the same radial distance. Their echo intermediate frequency signals will be mixed together. It is impossible to distinguish the life signals of multiple targets based on radial distance alone. Therefore, target angle estimation is also required.

[0102] like Figure 4 As shown in the figure, because the target echo direction is at an angle θ with the RX antenna plane, there is a spatial delay between the received signals of the same target echo at different RX antennas. The magnitude of this delay reflects the target's arrival angle. Therefore, performing an FFT on the channel dimension of the received signal can determine the target direction angle, but this method has low resolution. The Capon algorithm achieves higher angular resolution without increasing the number of antenna arrays. It minimizes the variance of the echo signal at the target direction angle, thereby minimizing noise power.

[0103] For each antenna channel composed of TX and RX, the matrix Y composed of the intermediate frequency signals of the total U antenna channels is Y = {y1(t),y2(t),...,y U (t)} T , the covariance is expressed as R Y =Y×Y H For any direction angle θ, the steering vector can be expressed as α(θ)={1,ej2πΔxsinθ ,e j2πΔxsinθ×2 ,...,e j2πΔxsinθ×(U-1)} T , then the signal power at this direction angle can be expressed as,

[0104]

[0105] By traversing all angles within [-π / 2,π / 2], we can obtain the sum of the power reaching the antenna at all directional angles, and extract the directional angle corresponding to each spectrum peak and the directional angle of each target.

[0106] Step 3: Target detection and distance and direction angle positioning.

[0107] The frequency spectrum can be obtained by fast time dimension FFT of the intermediate frequency signal. The spectrum peaks of the intermediate frequency signal of targets at different radial distances are different, and the radial distance range bins of the targets can be located in sequence.

[0108]

[0109] Based on the target azimuth angle obtained by searching the P(θ) spectrum peak obtained in step 3, a two-dimensional plot of the target's radial range and azimuth angle can be obtained, as shown in Figure 5(a). Although there is still some weak ambient noise, to ensure a low target miss rate, one-dimensional CASO-CFAR detection is performed on both the radial range and azimuth angle dimensions. The smaller of the average values ​​of reference cells outside a certain range at both ends of the target cell is used as the background noise estimate, as shown in Figure 5(b). After CASO-CFAR target detection, the peak values ​​corresponding to the radial range and azimuth angle in the main lobe of each target range are recorded.

[0110] Step 4: Separate the intermediate frequency signals of multiple target echoes.

[0111] For the receiving antenna (RX), the echo information of each target is mixed together and needs to be separated. Obviously, radial distance alone cannot successfully separate the echoes of targets with the same radial distance but different angles. Therefore, a beamforming algorithm is required to separate the echoes from all directions.

[0112] The Minimum Variance Distortionless Response (MVDR) beamforming algorithm is applied, which can enhance the echo signal in one direction while suppressing the echo in other directions. It finds the optimal weight W under the premise of satisfying the distortion constraint of the target echo signal. H , which minimizes the output power and ensures that the clutter noise is minimized.

[0113]

[0114] The optimal weight can be obtained by constructing the Lagrangian function.

[0115]

[0116] Combined with the target distance range, this coefficient matrix can be used to separate and extract the echo signals of all targets i detected in step 3.

[0117] Z i =W i H Y (24)

[0118] Step 5: Extraction of target life signals.

[0119] The phase of the echo intermediate frequency signal can be expressed as,

[0120]

[0121] in, is the residual noise of mixing, which can be ignored for short-distance measurement. In the human target detection scenario, the echo delay is at the nanosecond level, so -πγτ 2 The influence of the term on the phase can be directly ignored. Considering the weak vibration x(t) caused by the breathing and heartbeat signals in the human chest, the intermediate frequency signal phase is converted to,

[0122]

[0123] Since the human chest vibration x(t) is usually less than 1 cm and can be ignored relative to the radial distance of the target, the change trend of the intermediate frequency signal phase can reflect the target breathing and heartbeat signals.

[0124] To obtain the phase signal, inverse tangent demodulation is usually performed on the I and Q channels in the receiver. However, since the phase change caused by chest vibration x(t) is approximately 4πx(t) / λ=4π×1cm / 3.9mm≈10.26π, direct inverse tangent demodulation will cause the phase to fold within the range [-π,π) and become chaotic. Phase discontinuities need to be eliminated through a phase unwrapping algorithm. This method is not only cumbersome to implement in hardware but also has poor noise immunity. The phase change is described by the differential and cross-multiplication of the echo intermediate frequency signal. After A / D sampling conversion, the discrete form is expressed as,

[0125]

[0126] Step 6: Decomposition and reconstruction of breathing and heartbeat signals.

[0127] The phase signal obtained after DACM demodulation contains the target's breathing and heartbeat indicators.

[0128]

[0129] Among them, xb (t) and x h (t) are respectively the vibrations caused by breathing and heartbeat, and both can be characterized by a sinusoidal vibration model, and the frequency of both changes stably in a short time. In view of the difference in the frequency range of breathing and heartbeat, various signal decomposition algorithms can be applied to separate and extract the life signals.

[0130] The SVMD algorithm does not need to give the number of decomposed IMF modes in advance, and adaptively decomposes the target echo phase signal into a plurality of mode components of different frequencies. The constraint conditions of the SVMD decomposition optimization model include: (1) the sum of all mode components and the residual signal is equal to the input signal; (2) in order to resist mode aliasing, it is required that all mode components are guaranteed to be compactly surrounded around the center frequency thereof, and the criterion thereof is represented; (3) the current mode component has as small energy as possible in the vicinity of the center frequency of all previous modes, that is, the frequency spectrum repetition is minimized; and (4) the residual signal has as small energy as possible.

[0131] According to the difference in the frequency range of human body breathing and heartbeat (the normal human body breathing frequency range is 0.1-0.8 Hz, and the heartbeat frequency range is 0.9-2.0 Hz), the time domain superposition of the components in the corresponding frequency range is performed for signal reconstruction. For all target radial distances and azimuth angle information obtained in step 3, the phase information is extracted, and the DACM is disentangled to perform SVMD signal decomposition and reconstruction to obtain the breathing and heartbeat waveforms as shown in FIGS. 6(a)-(e). FIGS. 6(a)-(e) are respectively the life signal waveforms obtained by radar line of sight from right to left and from near to far.

[0132] Step 7, fingerprint feature extraction.

[0133] By analyzing the target echo y(t), the features of the human body are extracted, including the mean, standard deviation, root mean square value, skewness, kurtosis and peak factor of the signal envelope, a total of six.

[0134] On the other hand, on the basis of the breathing and heartbeat features, different individuals can be identified through radar technology. A series of features are selected, including breathing frequency, heartbeat frequency, number of breathing signal peaks, number of heartbeat signal peaks, breathing signal peak value, heartbeat signal peak value, duration of exhalation and inhalation, chest contraction strength and relaxation strength, a total of ten features.

[0135] Among them, the duration of exhalation and inhalation: the exhalation and inhalation process is manifested as the expansion and contraction of the chest of the human body, which belongs to periodic motion, and the duration is obtained by measuring the time interval of the maximum value and the minimum value of the radar signal. Chest contraction strength and relaxation strength: the duration is obtained by measuring the displacement of the radar signal from the minimum value to the maximum value.

[0136] The above 16 unique fingerprint eigenvalues ​​of the same target i are counted into the feature matrix,

[0137] F i =[M 1i ,M 2i ,...M 16i ] (29)

[0138] Thus, the feature matrix corresponding to different targets is obtained where i∈[1,N T ],N T is the number of detected targets, including the number of real targets and false targets.

[0139] Step 8: Target identity feature association.

[0140] Calculate the correlation matrix C between the i-th target and the j-th target ij for

[0141] C ij =[cov 11 ,cov 22 ,...,cov 1616 ]

[0142]

[0143] Calculate the correlation between the i-th target and the j-th target

[0144]

[0145] The correlation matrix of all targets is obtained as

[0146]

[0147] The ones with large correlation are the same target and its corresponding multipath false target, while the ones with small correlation are different targets.

[0148] Thus, multiple target groups T are obtained i

[0149]

[0150] Among them, i∈[1,N c ],N c is the target group number. i For the real goal, is a multipath false target, N g is the number of multipaths. The measured results are as follows Figure 7 As shown in the figure, there are three target groups in the scene, but the life signal of one of the targets is abnormal and there is no heartbeat signal frequency component, so it is identified as a clutter group.

[0151] Step 9: Multipath false target suppression.

[0152] The multipath propagation path is longer than the propagation path of the real target and has a larger propagation delay, so the target closest to the radar is selected as the real target R. i , and finally get the real target in the multi-person scenario,

[0153]

[0154] After selection, multipath false targets are suppressed, and the final human target detection result is as follows: Figure 8 shown.

Claims

1. A multipath echo identification and personnel detection method based on life signal fingerprint characteristics, characterized by: The specific steps of this method are as follows: Step 1: Obtain radar echo signals of multiple target personnel in a static state in the space and eliminate static clutter by using the mean cancellation method; Step 2: The radial distance of each target is obtained by performing FFT on the intermediate frequency signal along the fast time dimension, and the minimum variance undistorted response Capon algorithm is used to estimate the azimuth of the target. This algorithm minimizes the variance of the echo signal in the target direction angle, thereby maximally suppressing the noise power. Step 3: After steps 1 and 2, a two-dimensional range-direction angle map is obtained. CASO-CFAR detection is performed on the distance and angle dimensions respectively. The minimum value in the reference unit is selected as the reference background noise for multi-target extraction. Step 4: By weighted superposition of the received signals of each channel, the echo signal in a certain direction is enhanced while the energy of the signals in other directions is suppressed, and the target signal is separated using the minimum variance distortionless response (MVDR) beamforming algorithm; Step 5: Use differential cross multiplication to extract the phase signal, which contains the target's breathing signal and heartbeat signal; Step 6: Decompose and reconstruct the target breathing signal and heartbeat signal according to the frequency difference between the human breathing and heartbeat signals; Step 7: Perform radar detection on static human targets, and detect the target echo. Analyze and extract the body's physical characteristics, breathing characteristics, and heartbeat characteristics to ensure accurate capture of human physiological indicators. Summarize all characteristic values ​​to obtain the characteristic matrix corresponding to each target and integrate the multivariate characteristic values ​​of each target. Step 8: By accurately calculating the correlation between different targets, a correlation matrix is ​​obtained; by taking the mean of the correlation matrix, the average correlation between the targets is obtained; Step 9: After the personnel identity features are associated, the propagation delay differences caused by the differences in the signal propagation paths are used to distinguish the true target and the multipath propagation delay differences in each target set. The target closest to the radar is selected to determine the true target, ultimately achieving effective identification of the true target in multi-person scenarios. Among them, in step 6, the SVMD algorithm is essentially a group of multiple adaptive Wiener filters. The modal component optimization problem is transformed into A single-mode optimization problem, ； life signal After SVMD decomposition, we get modal components and high-frequency residual signal components ; Then extract the appropriate IMF component according to the human breathing and heartbeat spectrum range and reconstruct the breathing signal by time domain superposition. With heartbeat signal ; Among them, in step eight, calculate the Target and The correlation matrix between targets for: ； Calculate the Target and The correlation between the targets, ； Target Group Contains the real target and its corresponding multipath false target, which can be expressed as: ； in, , is the target group number; For the real goal, is a multipath false target, is the number of multipaths.

2. The multipath echo identification and personnel detection method based on the life signal fingerprint feature according to claim 1 is characterized in that: In step 1, the echo signal captured by the radar contains the quasi-static target and its own TX-end leakage signal. It is necessary to take the average of the echo of each range unit in each channel of each frame to estimate the DC component of each range unit. ； in, Indicates the Antenna Channel distance units The echo signal of the frame, is the total number of frames.

3. The multipath echo identification and personnel detection method based on life signal fingerprint characteristics according to claim 1 is characterized in that: In step 2, the high-resolution angle measurement minimum variance undistorted response Capon algorithm is used to calculate the spatial spectrum ,in is the steering vector, which is related to the number of receiving antenna channels. is the covariance matrix of the intermediate frequency signal matrix received by the antenna array; ； By searching for the peak of the spatial spectrum, the angle corresponding to the peak is the target arrival direction. .

4. The multipath echo identification and personnel detection method based on life signal fingerprint characteristics according to claim 1 is characterized in that: In step 4, based on the distance and azimuth information of all suspected targets detected in step 3, Calculate its channel Unique weighting coefficients , and accumulate The intermediate frequency signal of each channel echo is used to obtain the individual intermediate frequency signal of each target echo , 。 5. The multipath echo identification and personnel detection method based on life signal fingerprint characteristics according to claim 1 is characterized in that: In step five, the phase signal is expressed as: ； Among them, the intermediate frequency signal frequency Radial distance to target are all constants, including the target's breathing signal With heartbeat signal Changes over time; The DACM algorithm for differential cross-multiplication of the I and Q channel discrete signals after sampling is specifically expressed as: 。 6. The multipath echo identification and personnel detection method based on life signal fingerprint characteristics according to claim 1 is characterized in that: In step seven, the body characteristics include the mean, standard deviation, root mean square value, skewness, kurtosis and crest factor of the target echo intermediate frequency signal envelope; In terms of respiratory and heartbeat features, including respiratory rate, heartbeat rate, number of respiratory signal peaks, number of heartbeat signal peaks, respiratory signal peak, heartbeat signal peak, duration of exhalation and inhalation, and chest contraction strength and relaxation strength, respiratory and heartbeat signals are extracted.

7. The multipath echo identification and personnel detection method based on life signal fingerprint characteristics according to claim 1 or 6, characterized in that: In step seven, the same target All features of are counted into the feature matrix, ； So as to obtain the feature matrix corresponding to different targets ,in , is the number of detected targets, including the number of real targets and false targets.

8. The multipath echo identification and personnel detection method based on life signal fingerprint characteristics according to claim 1 is characterized in that: In step nine, the target closest to the radar is selected as the real target. , and finally get the real target in the multi-person scenario, 。

Images