A non-line-of-sight stationary personnel positioning method and system

Abstract

The present invention discloses a non-line-of-sight stationary personnel positioning method and system, which belongs to the field of radar sensing technology. A distance-velocity spectrum matrix is ​​obtained according to an intermediate frequency signal, and a constant false alarm rate detection is performed on the processed distance-velocity spectrum matrix to obtain a distance-velocity spectrum matrix of a stationary personnel target. According to the distance and signal strength of the stationary personnel target in polar coordinates, the micro-motion speed of the stationary personnel target, and the angle of the stationary personnel target in polar coordinates, the centroid of the clustered radar point cloud is obtained, and the coordinates corresponding to the centroid of the clustered radar point cloud are used as the position of the stationary personnel target. The present invention can capture the micro-motion characteristics of stationary personnel in a strong background noise environment, overcome the large positioning deviation problem caused by non-line-of-sight occlusion and indoor multipath by decoupling coherent multipath signals, and consider the distribution characteristics of radar point clouds. An adaptive parameter clustering algorithm is designed to avoid ambiguity caused by different environmental variables, and has good robustness, is concise and effective, and has broad application prospects.

Description

Technical Field

[0001] The invention relates to a non-line-of-sight stationary personnel positioning method and system, belonging to the technical field of millimeter wave radar sensing. Background Art

[0002] The positioning of stationary personnel in non-line-of-sight areas has always been a difficult problem. Traditional positioning methods mainly rely on visual sensors, such as cameras and infrared sensors, but these sensors are affected by factors such as ambient light, obstacles, and weather, and cannot achieve accurate positioning. Millimeter-wave radar has the characteristics of high precision, high resolution, and penetration, and has the potential to achieve non-line-of-sight high-precision positioning services.

[0003] At present, some studies have proposed methods for using millimeter-wave radar to locate people in non-line-of-sight scenarios, but most of these methods locate moving people, whose movements are large and can easily be used as indicators for positioning people. However, when people are stationary, they face challenges such as insufficient signal strength, coherent multipath interference, and clustering parameters that rely on expert experience to tune, making it impossible to achieve accurate positioning.

[0004] Specifically, the radar signal changes caused by the micro-movement behaviors (such as breathing, heartbeat, etc.) of stationary personnel are severely attenuated due to non-line-of-sight occlusion, and the micro-movement characteristic signals are easily buried in the strong background clutter and difficult to distinguish; secondly, non-line-of-sight occlusion exacerbates the multipath effect, making it difficult to decouple the coherent signal, making it difficult to solve the angle of the stationary personnel; in addition, the radar echo point cloud processed by the algorithm shows the characteristics of sparseness at a distance and density at a close distance. To achieve accurate clustering and then solve the position information, there are challenges such as relying on expert experience to tune the key parameters of the clustering algorithm. Therefore, it is challenging to achieve accurate positioning of non-line-of-sight stationary personnel.

[0005] Therefore, the present invention aims to propose a feasible solution for locating a stationary person in a non-line-of-sight scenario, so as to solve the problem of tracking a stationary person in a non-line-of-sight scenario. Summary of the invention

[0006] Purpose: In order to overcome the problems in the prior art that radar echo signals of stationary personnel are difficult to distinguish in a strong background noise environment, personnel positioning deviation is large due to non-line-of-sight occlusion and indoor multipath, and the target echo point cloud is sparse due to non-line-of-sight, the present invention provides a non-line-of-sight stationary personnel positioning method, which improves the target signal-to-noise ratio through coherent accumulation and strengthens the micro-motion characteristics of stationary personnel; introduces an enhanced super-resolution algorithm that performs Toeplitz reconstruction of the covariance matrix to solve the coherent multipath problem and achieve accurate angle estimation; in order to avoid the ambiguity caused by different environmental variables, a millimeter wave point cloud adaptive clustering method considering the spatial characteristics of the scene is proposed, which can obtain real location information without relying on expert knowledge to adjust and optimize the key parameters of the clustering algorithm. Based on the above steps, the challenges faced by non-line-of-sight stationary personnel positioning are overcome.

[0007] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is:

[0008] In a first aspect, a non-line-of-sight stationary personnel positioning method specifically includes:

[0009] The intermediate frequency signal of the radar echo signal is obtained, and the range-velocity spectrum matrix is ​​obtained according to the intermediate frequency signal.

[0010] The range-velocity spectrum matrix is ​​preprocessed to obtain a processed range-velocity spectrum matrix.

[0011] The processed distance-velocity spectrum matrix is ​​subjected to constant false alarm rate detection to detect the distance unit index and velocity unit index of the stationary personnel target. According to the distance unit index and velocity unit index of the stationary personnel target, the distance-velocity spectrum matrix of the stationary personnel target is obtained, and the distance and signal strength of the stationary personnel target in polar coordinates are determined.

[0012] Doppler phase compensation is performed on the range-velocity spectrum matrix of the stationary personnel target to obtain the compensated range-velocity spectrum matrix, and the micro-motion velocity of the stationary personnel target is obtained according to the compensated range-velocity spectrum matrix.

[0013] The compensated range-velocity spectrum matrix is ​​reorganized according to the number of array elements, and coherent multipath decoupling is performed to obtain the angle of the stationary personnel target in polar coordinates.

[0014] The distance and angle of the stationary human target in polar coordinates are converted into radar point clouds in Cartesian coordinates. According to the coordinates, micro-motion speed and signal strength of the radar point clouds, the radar point clouds in Cartesian coordinates are clustered to obtain the centroid of the clustered radar point clouds.

[0015] The coordinates corresponding to the centroid of the clustered radar point cloud are used as the position of the stationary person target.

[0016] As a preferred solution, the method for acquiring the intermediate frequency signal of the radar echo signal specifically includes:

[0017] The linear frequency modulation signal emitted by the radar to the target is obtained. The expression of the linear frequency modulation signal is as follows:

[0018] S Tx ( t ) = A T c o s [ 2 π f c t + π B T c t 2 + ϕ ( t ) ]

[0019] in, represents the linear frequency modulation signal, t represents the time, A T represents the amplitude of the transmitted signal, is the starting frequency of the transmitted Chirp signal, B is the frequency modulation bandwidth of the transmitted Chirp signal, T c is the duration of transmitting a single Chirp signal, φ(t) is the phase noise of the transmitted signal, is the ratio of pi.

[0020] The echo signal of the target is obtained, and the expression of the echo signal is as follows:

[0021] S Rx ( t ) = α A T c o s { [ 2 π f c ( t − τ ) ] + π B T c ( t − τ ) 2 + ϕ ( t − τ ) }

[0022] in, Represents the echo signal, It represents the attenuation coefficient of the amplitude during signal propagation. It represents the delay of the received signal relative to the transmitted signal. Represents the phase noise of the echo signal.

[0023] The linear frequency modulation signal is coherently mixed with the echo signal, and an intermediate frequency signal is output through a low-pass filter. The expression of the intermediate frequency signal is as follows:

[0024] S ( t ) = α A T s i n [ 2 π ( B τ T c t + f c τ ) ]

[0025] in, Represents the intermediate frequency signal.

[0026] As a preferred solution, the radar adopts a frequency modulated continuous wave millimeter wave radar.

[0027] As a preferred solution, obtaining the range-velocity spectrum matrix according to the intermediate frequency signal specifically includes:

[0028] The intermediate frequency signal is discretized and framed to obtain a matrix of M frequency modulation periodic sequence signals. , the frequency modulation period sequence signal matrix The expression is as follows:

[0029] S i = α A T ( n , m ) s i n { 2 π [ f b ( n ) + ϕ b ( m ) ] n / F s }

[0030] Where n represents the fast time index within a single frequency modulation cycle, n=0,...,N ADC -1, where N ADC Indicates the number of sampling points in a single frequency modulation cycle; m represents the slow time index in a single frame, m=0,...,N chirp -1, where N chirp The number of Chirp signals transmitted in each frame period; F s Indicates the ADC sampling rate of the intermediate frequency signal, Represents the signal matrix The dimensions corresponding to the medium-fast time index n and the slow time index m, Indicates the frequency of the intermediate frequency signal corresponding to the fast time index, Represents the phase of the intermediate frequency signal corresponding to the slow time index, i=1,... ,M.

[0031] For each frequency modulation cycle sequence signal matrix Perform frequency domain transformation in the fast time index direction of the representative column to obtain the two-dimensional distance spectrum matrix SR i , where the fast time index direction is divided into N adc distance intervals, the distance interval is the distance resolution R res .

[0032] For the two-dimensional distance spectrum matrix SR i Perform frequency domain transformation in the slow time index direction of the representative row to obtain the range-velocity spectrum matrix SRD i , where the slow time index direction is divided into N chirp speed intervals, which is the speed resolution.

[0033] As a preferred solution, the preprocessing of the distance-velocity spectrum matrix to obtain the processed distance-velocity spectrum matrix specifically includes:

[0034] Remove each range-velocity spectrum matrix SRD i After filtering out the static clutter in the environment, the DC component interference is removed to obtain the matrix after interference elimination.

[0035] The matrix after interference elimination is coherently accumulated to obtain the processed range-velocity spectrum matrix.

[0036] As a preferred solution, the removal of each range-velocity spectrum matrix SRD i After removing the static clutter in the environment, the DC component interference is removed to obtain the matrix after the interference is eliminated. The matrix after the interference is eliminated is coherently accumulated to obtain the processed range-velocity spectrum matrix, which specifically includes:

[0037] The vector mean elimination method is used to remove each range-velocity spectrum matrix SRD i The static clutter in the environment is removed by using the zero channel zeroing method for each range-velocity spectrum matrix SRD i The DC component interference in is eliminated, and the matrix after interference elimination is obtained.

[0038] Superimpose M interference-eliminating matrices to obtain the processed range-velocity spectrum matrix .

[0039] As a preferred solution, the processed distance-velocity spectrum matrix is ​​subjected to constant false alarm rate detection, the distance unit index and the velocity unit index of the stationary personnel target are detected, the distance-velocity spectrum matrix of the stationary personnel target is obtained according to the distance unit index and the velocity unit index of the stationary personnel target, and the distance and signal strength of the stationary personnel target in polar coordinates are determined, specifically including:

[0040] A set of reference cells is selected from the cells to be tested in the processed range-velocity spectrum matrix x.

[0041] The background noise level of the processed range-velocity spectrum matrix x is estimated using the selected reference cells.

[0042] An adaptive detection threshold is calculated based on the estimated background noise level and a preset false alarm rate.

[0043] The signal strength of the unit to be tested is compared with the calculated adaptive detection threshold. If the signal amplitude of the unit to be tested exceeds the adaptive detection threshold, it is determined to be a stationary human target; otherwise, it is determined to be noise or clutter.

[0044] According to the stationary personnel target, the distance unit index and the speed unit index of the stationary personnel target are obtained, and according to the distance unit index and the speed unit index of the stationary personnel target, the distance-speed spectrum matrix of the stationary personnel target is obtained, and the distance and signal strength of the stationary personnel target in polar coordinates are determined.

[0045] As a preferred solution, the compensated range-velocity spectrum matrix is ​​reorganized according to the number of array elements, and coherent multipath decoupling is performed to obtain the angle of the stationary personnel target in polar coordinates, specifically including:

[0046] The compensated range-velocity spectrum matrix is ​​reorganized according to the number of array elements to form L×N chirp The matrix y of .

[0047] Calculate the covariance matrix R of matrix y. The expression of covariance matrix R is as follows:

[0048] R = E [ y y H ]

[0049] Where E[-] represents the expected value and the superscript H represents the conjugate transpose.

[0050] Solve the eigenvalues ​​of the covariance matrix R and arrange the eigenvalues ​​in descending order as β1, β2, ..., β Q , β Q+1 , ..., β L-1 , β L , the eigenvectors corresponding to each eigenvalue are u1, u2, ..., u Q 、u Q+1 ,...,u L-1 、u L , where the subscript Q represents the number of coherent signals and the subscript L represents the number of elements in the uniform antenna array.

[0051] The decorrelation is achieved by selecting the eigenvectors u1 and u2 corresponding to the first two largest eigenvalues ​​β1 and β2.

[0052] u 1 = [ u 1 1 , u 2 1 , ⋯ , u ( L − 1 ) 1 , u L 1 ] T

[0053] The first vector u in the above formula 11 As a reference element, u 11 Correlation function between each element It can be expressed as:

[0054] r ( k − 1 ) = E [ u 1 1 u k 1 H ] , k = 1 , 2 , ⋯ , L

[0055] Then we can get formula 1:

[0056] [ r ( 0 ) , r ( 1 ) , ⋯ , r ( L − 1 ) ] = [ E [ u 1 1 u 1 1 H ] , E [ u 1 1 u 2 1 H ] , ⋯ , E [ u 1 1 u L 1 H ] ]

[0057] Similarly, each element is related to u 11 The correlation function between them is expressed as r(-k+1):

[0058] r ( − k + 1 ) = E [ u k 1 u 1 1 H ] , k = 1 , 2 , ⋯ , L

[0059] The second formula is:

[0060] [ r ( − L + 1 ) , r ( − L ) , ⋯ , r ( 0 ) ] = [ E [ u L 1 u 1 1 H ] , E [ u ( L − 1 ) 1 u 1 1 H ] , ⋯ , E [ u 1 1 u 1 1 H ] ]

[0061] Combining equation 1 and equation 2, construct a new Toeplitz matrix Y1:

[0062] Y 1 = [ r ( 0 ) r ( 1 ) ⋯ r ( L − 1 ) r ( − 1 ) r ( 0 ) ⋯ r ( L − 2 ) ⋮ ⋮ ⋮ r ( − L + 1 ) r ( − L ) ⋯ r ( 0 ) ] L × L

[0063] in, . Superscript represents the conjugate transpose.

[0064] For the Toeplitz matrix Y1, construct the corresponding unbiased estimation matrix :

[0065]

[0066] Where: J is an L×L commutative matrix, i.e., a matrix in which the sub-diagonal elements are "1" and all other elements are "0".

[0067] Select the first element u in u2 12 As a reference element, the unbiased estimation matrix YY2 is obtained by the same steps as above, namely:

[0068] [ r ' ( 0 ) , r ' ( 1 ) , ⋯ , r ' ( L − 1 ) ] = [ E [ u 1 2 u 1 2 H ] , E [ u 1 2 u 2 2 H ] , ⋯ , E [ u 1 2 u L 2 H ] ]

[0069] [ r ' ( − L + 1 ) , r ' ( − L ) , ⋯ , r ' ( 0 ) ] = [ E [ u L 2 u 1 2 H ] , E [ u ( L − 1 ) 2 u 1 2 H ] , ⋯ , E [ u 1 2 u 1 2 H ] ]

[0070] Y 2 = [ r ' ( 0 ) r ' ( 1 ) ⋯ r ' ( L − 1 ) r ' ( − 1 ) r ' ( 0 ) ⋯ r ' ( L − 2 ) ⋮ ⋮ ⋮ r ' ( − L + 1 ) r ' ( − L ) ⋯ r ' ( 0 ) ] L × L

[0071]

[0072] According to the unbiased estimation matrix , YY2, construct a new Toeplitz matrix Y:

[0073] .

[0074] Get the eigenvalue decomposition of matrix Y:

[0075]

[0076] Where U represents the eigenvector matrix of matrix Y, U=[u1,…,u L ]; U H represents the conjugate transposed matrix of the eigenvector matrix U; Σ represents the eigenvalue diagonal matrix of the matrix Y, Σ=diag{β1, β2 ... , β Q , β Q+1 ... β L , β L+1}. ss The eigenvector matrix representing the signal subspace, U ss = [u1, u2..., u Q-1 , u Q ]; The conjugate transposed matrix of the eigenvector matrix representing the signal subspace; Σ ss represents the eigenvalue matrix of the signal subspace, Σ ss= diag{β1, β2,..., β Q-1 , β Q};U nn Represents the eigenvector matrix of the noise subspace, U nn = [u Q+1 , ..., u L-1 , u L ]; The conjugate transposed matrix of the eigenvector matrix representing the noise subspace; Σ nn Represents the eigenvalue matrix of the signal-noise subspace, = diag{β Q+1 , ..., β L}.

[0077] According to the eigenvalue decomposition results, the DOA of the coherent incident signal is estimated, and the spatial spectrum function is obtained as follows:

[0078]

[0079] In the formula, The array manifold matrix of the coherent signal, a(θ)=[a(θ1), a(θ2), ..., a(θ Q )], a(θ i ) represents the angle θ of the i-th coherent signal i The corresponding array steering vector, a(θ i ) = [1, e -jπsin(θi) , ..., e -j(L-1)πsin(θi) ] T , 1≤i≤Q; the superscript * indicates the conjugate transpose.

[0080] The peak value of the spatial spectrum function is searched, and the position of the spectrum peak is the angle of the stationary person target in polar coordinates.

[0081] As a preferred solution, the distance and angle of the stationary human target in polar coordinates are converted into a radar point cloud in a Cartesian coordinate system, and the radar point cloud in the Cartesian coordinate system is clustered according to the coordinates, micro-motion speed and signal strength of the radar point cloud to obtain the centroid of the clustered radar point cloud, which specifically includes:

[0082] The distance and angle of the stationary human target in polar coordinates are converted into a radar point cloud in a Cartesian coordinate system. The coordinate value of the radar point cloud in the Cartesian coordinate system is (x, y), and the expression is as follows:

[0083]

[0084] in, represents the distance in polar coordinates, Represents an angle in polar coordinates.

[0085] The horizontal coordinate of the stationary target , vertical coordinate , the micro-motion speed v, the signal strength Strength are normalized to obtain the normalized horizontal coordinate , vertical coordinate , micro-motion speed v, signal strength Strength.

[0086] Using the normalized horizontal coordinate , vertical coordinate , micro-motion speed v, signal strength Strength, calculate the Euclidean distance between two points in the radar echo point cloud.

[0087] Calculate the minimum number of points within the neighborhood radius , the expression is as follows:

[0088]

[0089] Here, dim is the dimension of the input data.

[0090] Calculate neighborhood radius , the expression is as follows:

[0091]

[0092] Among them, Δθ is the angular resolution of the radar, d is the distance from the point cloud target to the radar, and δ is the custom threshold coefficient.

[0093] The Euclidean distance between two points in the radar echo point cloud and the minimum number of points in the neighborhood radius And the neighborhood radius The DBSCAN algorithm is input to cluster the radar echo point cloud of stationary personnel to obtain the centroid of the clustered radar point cloud.

[0094] In a second aspect, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements a non-line-of-sight stationary personnel positioning method as described in any one of the first aspects.

[0095] According to a third aspect, a computer device includes:

[0096] Memory, used to store instructions.

[0097] The processor is used to execute the instructions so that the computer device performs the operations of the non-line-of-sight stationary personnel positioning method as described in any one of the first aspects.

[0098] Beneficial effects: The present invention provides a non-line-of-sight stationary personnel positioning method, which can capture the micro-motion characteristics of stationary personnel in a strong background noise environment. By decoupling coherent multipath signals, the problem of large positioning deviation caused by non-line-of-sight occlusion and indoor multipath is overcome. Considering the distribution characteristics of radar point cloud, an adaptive parameter clustering algorithm is designed to avoid ambiguity caused by different environmental variables. The method of the present invention has good robustness, is concise and effective, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 It is a flow chart of the implementation process of the method of the present invention.

[0100] Figure 2 It is a schematic diagram of constructing a distance-velocity spectrum matrix of discretized data of an intermediate frequency signal during the implementation of the method of the present invention.

[0101] Figure 3 It is a schematic diagram of the vector mean elimination method in the implementation process of the method of the present invention.

[0102] Figure 4 It is a process diagram of the micro-motion feature enhancement method in the implementation process of the method of the present invention to process the original signal spectrum. (a) represents the original spectrum of the radar intermediate frequency signal, (b) represents the signal spectrum of the original signal matrix after static clutter filtering, (c) represents the signal spectrum after the DC component is removed after static clutter filtering, and (d) represents the signal spectrum of the micro-motion feature enhancement obtained after coherent accumulation after the DC component is removed. (e) represents the signal spectrum after the traditional MTI algorithm is performed on the original signal matrix, and (f) represents the signal spectrum after the original signal matrix is ​​zeroed in the zero channel.

[0103] Figure 5 It is a schematic diagram of a radar antenna during the implementation of the method of the present invention.

[0104] Figure 6 The present invention is a flowchart of an enhanced super-resolution algorithm (CD-MUSIC) for Toeplitz reconstruction of a covariance matrix during implementation of the method of the present invention.

[0105] Figure 7 It is a comparison diagram of the calculation results of the enhanced super-resolution algorithm (CD-MUSIC) and other algorithms during the implementation of the method of the present invention.

[0106] Figure 8 The results of the method of the present invention for stationary personnel target comparison are shown in Figure 1. (a) is an experimental scene diagram of stationary personnel positioning using FMCW millimeter wave radar in non-line-of-sight conditions, (b) is a sparse radar point cloud image obtained using the traditional DBSCAN algorithm, and (c) is a radar point cloud image obtained using an adaptive clustering algorithm that considers the spatial characteristics of the scene.

[0107] Fig. 9 These are the common building materials used in the implementation of the method of the present invention and their relative distances from the radar. (a) represents a wooden board, (b) represents glass, (c) represents a PE plastic board, (d) represents a gypsum board, and (e) represents the relative distance between the building shielding material and the FMCW millimeter wave radar.

[0108] Fig.10 The figure is a schematic diagram of the positioning of non-line-of-sight stationary personnel using the method of the present invention under different material occlusion conditions during the implementation of the method of the present invention, as well as the positioning errors under different material and distance conditions. (a) represents the distance of the stationary personnel in the direction relative to the radar normal, which are 0.8m, 1.6m, 2.4m, 3.2m, and 4.0m respectively; (b) represents the error of the positioning of the stationary personnel using the method of the present invention under different obstacle occlusion conditions. DETAILED DESCRIPTION

[0109] The following is a clear and complete description of the technical solutions in the examples of the present invention in conjunction with the accompanying drawings in the examples of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the protection scope of the present invention.

[0110] The present invention will be further described below in conjunction with specific embodiments.

[0111] Embodiment 1:

[0112] This embodiment introduces a non-line-of-sight stationary person positioning method, and the specific steps are as follows:

[0113] Step 1: Obtain the millimeter-wave radar echo signal of a stationary person in a non-line-of-sight scenario, uniformly sample the intermediate frequency signal, construct an M-frame frequency modulation periodic sequence signal matrix, and then perform a two-dimensional fast Fourier transform (2D-FFT) on the frequency modulation periodic sequence signal matrix to obtain a distance-velocity spectrum matrix. Use the vector mean elimination method to eliminate the environmental static clutter interference in the matrix, and then use the zero channel zeroing method to remove the DC component interference in the matrix to obtain the matrix after interference elimination. Finally, perform coherent accumulation on the matrix after interference elimination to obtain the accumulated matrix signal, which is used to enhance the micro-motion characteristics of stationary personnel. Step 1 is used to enhance the micro-motion characteristics of the radar echo signal of a non-line-of-sight stationary person.

[0114] Furthermore, the step 1 specifically includes:

[0115] Step 1.1: Obtain intermediate frequency signal data. Use frequency modulated continuous wave (FMCW) millimeter wave radar to periodically transmit a Chirp signal (linear frequency modulation signal) with a frequency that increases linearly with time to the target human body in the radar field of view. The linear frequency modulation signal model S Tx (t) is:

[0116] S Tx ( t ) = A T c o s [ 2 π f c t + π B T c t 2 + ϕ ( t ) ] (1)

[0117] Where: the subscript Tx represents the signal transmitted by the transmitting antenna, t represents the time, A T represents the amplitude of the transmitted signal, is the starting frequency of the transmitted Chirp signal, B is the frequency modulation bandwidth of the transmitted Chirp signal, T c is the duration of transmitting a single Chirp signal, φ(t) is the phase noise of the transmitted signal, is the ratio of pi.

[0118] Assuming that the distance between the stationary target and the radar is d and the speed of light is c, the delay of the received signal relative to the transmitted signal is τ=2d / c. Therefore, the target echo signal model received by the receiving antenna is as follows:

[0119] S Rx ( t ) = α A T c o s { [ 2 π f c ( t − τ ) ] + π B T c ( t − τ ) 2 + ϕ ( t − τ ) } (2)

[0120] Where: the subscript Rx represents the signal received by the receiving antenna, It represents the attenuation coefficient of the amplitude during signal propagation, and its value is affected by factors such as distance d, reflection cross-sectional area of ​​the target, and material. is the phase noise of the echo signal.

[0121] In the radar receiver, the received signal S Rx (t) and the transmitted signal S Tx (t) is coherently mixed, and then processed by a low-pass filter to output the intermediate frequency signal S(t), which is expressed as:

[0122] S ( t ) = α A T s i n [ 2 π B τ T c t + 2 π f c τ − π B T c τ 2 +Δ ϕ ( t ) ] (3)

[0123] In the formula, the phase πBτ 2 / T c The term is very small and can be omitted; the intermediate frequency signal phase noise Δϕ(t) = ϕ(t)- ​​ϕ(t-τ) can also be ignored when the distance is short. Therefore, the intermediate frequency signal model of formula (3) can be simplified to:

[0124] S ( t ) = α A T s i n [ 2 π ( B τ T c t + f c τ ) ] (4)

[0125] From formula (4), we can see that the frequency of the intermediate frequency signal is f b =2πBτ / T c , the phase is φ b =2πf c τ. The uniform sampling method is used to discretize formula (4), converting the continuous time signal into a discrete time series, and obtaining the signal matrix S of each frame of frequency modulation period sequence:

[0126] S = α A T ( n , m ) s i n { 2 π [ f b ( n ) + ϕ b ( m ) ] n / F s } (5)

[0127] Where n represents the fast time index within a single frequency modulation cycle, n=0,...,N ADC -1, where N ADC Indicates the number of sampling points in a single frequency modulation cycle; m represents the slow time index in a single frame, m=0,...,N chirp -1, where N chirp The number of Chirp signals transmitted in each frame period; F s Indicates the ADC sampling rate of the IF signal. Represents the dimension of the signal matrix S, which is the discretized S(t). Indicates the frequency of the intermediate frequency signal corresponding to the fast time index, Indicates the phase of the intermediate frequency signal corresponding to the slow time index.

[0128] Step 1.2: Two-dimensional FFT. According to step 1, the collected intermediate frequency signal data is processed by frame division to obtain M N adc ×N chirp The frequency modulation period sequence signal matrix S i , i=1,...,M. Matrix S i Each column represents the number of N in the fast time direction (i.e., in the timing direction of a single linear frequency modulation signal). adc sampling points, each row represents the slow time (slow time index) direction (i.e., a single frame sequence transmits N chirp pulse signals) with different N chirp The echo values ​​of the echoes of the pulse signals at the same fast time. For each matrix S i By performing FFT transformation on the fast time dimension, we can get N adc A two-dimensional distance spectrum matrix SR i At this time, the radar field of view (i.e., the two-dimensional range spectrum matrix SR i ) is divided into different distance intervals, each distance interval represents the minimum distance between different targets that the radar can distinguish, that is, the distance resolution Rres :

[0129] (6)

[0130] For the two-dimensional distance spectrum matrix SR i Perform FFT transformation in the slow time direction to obtain the range-velocity spectrum matrix SRD i At this time, the radar field of view (range-velocity spectrum matrix SRD i ) is divided into different speed intervals in the slow time direction. Each speed interval represents the minimum speed at which the radar can distinguish different targets, that is, the speed resolution v res :

[0131] (7)

[0132] Where λ is the wavelength of the radar signal, λ=c / f c .

[0133] Step 1.3: Static clutter elimination. For each frame period, the range-velocity spectrum matrix SRD i The average values ​​of the elements of are as follows:

[0134] q = 1 N chirp ∑ j = 1 N chirp S R D i [ : , j ] (8)

[0135] Among them, q is the reference speed signal obtained by taking the average value of all signals. Then the static clutter elimination process can be expressed as:

[0136] (9)

[0137] in, Q = [ q , q , ⋯ , q ] ∈ ℂ N adc × N chirp , P i is the i-th interference cancellation signal matrix obtained after vector mean cancellation.

[0138] Step 1.4: Remove the DC component. The DC component is caused by the effect of circuit components and insufficient isolation of the transmitting and receiving antennas. The near-zero distance channel nulling method is used to eliminate the signal matrix P. i DC component interference, while retaining the micro-motion characteristics of the target to obtain the i-th interference elimination signal matrix P i '.

[0139] Step 1.5: Coherent accumulation. Since the micro-movement speed of stationary personnel is slow and the displacement is small, the radar echo power is often small, and the strong background environment clutter can easily annihilate the weak target echo signal. In this regard, the coherent accumulation method is used to superimpose M interference elimination signal matrices P i 'Get the accumulated matrix :

[0140] (10)

[0141] Where i=1,...,M.

[0142] The signal quality can be effectively improved without increasing the computational complexity, while the signal amplitude of the micro-motion characteristics of stationary personnel can be highlighted.

[0143] Step 2: Perform a constant false alarm rate (CFAR) test on the data matrix obtained in step 1 to detect the distance and velocity unit index of the stationary personnel target, eliminate suspected targets, and eliminate false detections caused by dynamic clutter interference. Then perform Doppler phase compensation on the data matrix to obtain the compensated data matrix. An enhanced super-resolution algorithm that performs Toeplitz reconstruction on the covariance matrix of the data matrix y is used to solve the coherent multipath problem and achieve accurate angle estimation. Step 2 is used to calculate the distance of the stationary personnel relative to the radar, and perform coherent multipath decoupling on the echo signal to calculate the angle, velocity and signal strength of the stationary personnel relative to the radar.

[0144] As a preferred solution, the step 2 specifically includes:

[0145] Step 2.1: Use CFAR detection to determine the distance unit index. Perform CFAR target detection on the data matrix x obtained in step 1, select a set of reference units around the unit to be tested in the data matrix x, usually including protection units and training units, and use the selected reference units to estimate the background noise level of the data matrix x. Calculate the adaptive detection threshold based on the estimated background noise level and the preset false alarm rate. Compare the signal strength of the unit to be tested with the calculated threshold. If the signal amplitude exceeds the threshold, it is judged as a stationary human target; otherwise, it is judged as noise or clutter. Output the detection result, that is, the distance unit index n in the data matrix x where the signal amplitude exceeds the threshold i and speed unit index m j , the polar coordinate distance d, micro-motion speed v and signal strength (Strength) of the stationary human target relative to the radar can be determined.

[0146] Step 2.2: Doppler phase compensation. By compensating for the phase shift caused by the Doppler effect, the data matrix x of the stationary personnel target is Doppler phase compensated to obtain the compensated distance-velocity spectrum matrix, and the more accurate speed of the stationary personnel target is obtained through the compensated distance-velocity spectrum matrix.

[0147] Step 2.3: Coherent multipath decoupling. Obstacles in non-line-of-sight scenarios exacerbate problems such as multipath propagation, signal superposition, and coherent multipath interference, which increases the difficulty of estimating the direction of arrival (DOA). Traditionally, the Multiple Signal Classification (MUSIC) algorithm is used to construct a spatial spectrum function through the orthogonality of the signal subspace and the noise subspace to estimate the signal DOA. However, coherent multipath signals are highly correlated, which makes the MUSIC algorithm unable to accurately distinguish the signal subspace and the noise subspace, and unable to decoherent, thereby generating pseudo-spectral peak interference, making it impossible to accurately identify and locate the incident direction of the coherent signal. In this regard, an enhanced super-resolution algorithm (CD-MUSIC) that performs Toeplitz reconstruction on the covariance matrix of the data matrix x is used to solve the coherent multipath problem and achieve accurate DOA estimation. Specifically, for a uniform antenna array composed of L array elements (the determination of the L value is determined by the selected radar model and parameter configuration), Q coherent signals are respectively emitted from different directions θ1, θ2, ..., θ Q is incident on the array.

[0148] The compensated range-velocity spectrum matrix is ​​reorganized according to the number of array elements to form L×N chirp The matrix y of .

[0149] Calculate the covariance matrix R of matrix y. The expression of covariance matrix R is as follows:

[0150] R = E [ y y H ] (11)

[0151] In the formula, E[-] represents the expected value. The superscript H represents the conjugate transpose. Solve the eigenvalues ​​of the covariance matrix R and arrange them in descending order as β1, β2, ..., β Q , β Q+1 , ..., β L-1 , β L , the eigenvectors corresponding to each eigenvalue are: u1, u2, ..., u Q 、u Q+1 ,...,u L-1 、u L , where the subscript Q represents the number of coherent signals and the subscript L represents the number of elements in the virtual uniform antenna array. Eigenvalue β1 ≥ β2 ≥ … ≥ β Q ≥β Q+1 ≥…≥β L-1 ≥β L . The largest eigenvectors u1 and u2 corresponding to the first two largest eigenvalues ​​β1 and β2 are selected to achieve decorrelation. Let

[0152] u 1 = [ u 1 1 , u 2 1 , ⋯ , u ( L − 1 ) 1 , u L 1 ] T (12)

[0153] The first vector u in the above formula 11 As a reference element, u 11 Correlation function between each element It can be expressed as:

[0154] r ( k − 1 ) = E [ u 1 1 u k 1 H ] , k = 1 , 2 , ⋯ , L (13)

[0155] Then we can get:

[0156] [ r ( 0 ) , r ( 1 ) , ⋯ , r ( L − 1 ) ] = [ E [ u 1 1 u 1 1 H ] , E [ u 1 1 u 2 1 H ] , ⋯ , E [ u 1 1 u L 1 H ] ] (14)

[0157] Similarly, each element is related to u 11 The correlation function between them is expressed as r(-k+1):

[0158] r ( − k + 1 ) = E [ u k 1 u 1 1 H ] , k = 1 , 2 , ⋯ , L (15)

[0159] We can get:

[0160] [ r ( − L + 1 ) , r ( − L ) , ⋯ , r ( 0 ) ] = [ E [ u L 1 u 1 1 H ] , E [ u ( L − 1 ) 1 u 1 1 H ] , ⋯ , E [ u 1 1 u 1 1 H ] ] (16)

[0161] Combining formulas (14) and (16), a new Toeplitz matrix Y1 can be constructed:

[0162] Y 1 = [ r ( 0 ) r ( 1 ) ⋯ r ( L − 1 ) r ( − 1 ) r ( 0 ) ⋯ r ( L − 2 ) ⋮ ⋮ ⋮ r ( − L + 1 ) r ( − L ) ⋯ r ( 0 ) ] L × L (17)

[0163] in, .

[0164] For the Toeplitz matrix Y1, the forward and backward spatial smoothing idea is used to construct the corresponding unbiased estimation matrix:

[0165] (18)

[0166] Where J is an L×L commutative matrix, i.e., a matrix in which the sub-diagonal elements are "1" and all other elements are "0".

[0167] Select the first element u in u2 12 As a reference element, the matrix YY2 is obtained by the same steps as above, namely:

[0168] [ r ' ( 0 ) , r ' ( 1 ) , ⋯ , r ' ( L − 1 ) ] = [ E [ u 1 2 u 1 2 H ] , E [ u 1 2 u 2 2 H ] , ⋯ , E [ u 1 2 u L 2 H ] ] (19)

[0169] [ r ' ( − L + 1 ) , r ' ( − L ) , ⋯ , r ' ( 0 ) ] = [ E [ u L 2 u 1 2 H ] , E [ u ( L − 1 ) 2 u 1 2 H ] , ⋯ , E [ u 1 2 u 1 2 H ] ] (20)

[0170] Y 2 = [ r ' ( 0 ) r ' ( 1 ) ⋯ r ' ( L − 1 ) r ' ( − 1 ) r ' ( 0 ) ⋯ r ' ( L − 2 ) ⋮ ⋮ ⋮ r ' ( − L + 1 ) r ' ( − L ) ⋯ r ' ( 0 ) ] L × L (twenty one)

[0171] (twenty two)

[0172] Based on formulas (18) and (22), a new Toeplitz matrix Y is constructed:

[0173] (twenty three)

[0174] Since both matrix Y1 and matrix Y2 are Toeplitz matrices, the transformed matrices YY1 and YY2 are also Toeplitz matrices, so matrix Y is also a Toeplitz matrix. The Toeplitz matrix Y constructed by the new method contains the arrival direction signal of each incident signal, so Y can be equivalent to the new covariance matrix of the coherent incident signal. Therefore, the covariance matrix Y can be used to estimate the DOA of the coherent signal, and the rank of the matrix is ​​equal to the number of incident signals, and the corresponding number of array elements remains L, that is, there is no loss of array aperture.

[0175] Then, the DOA of the coherent signal is estimated from the Toeplitz matrix Y using the MUSIC algorithm, and the eigenvalue decomposition of the matrix Y can be obtained:

[0176] (twenty four)

[0177] Where U represents the eigenvector matrix of matrix Y, U=[u1,…,u L ]; U H represents the conjugate transposed matrix of the eigenvector matrix U; Σ represents the eigenvalue diagonal matrix of the matrix Y, Σ=diag{β1, β2 ... , β Q , β Q+1 ... β L , β L+1}, β1 ≥ β2≥... ≥ β Q ≥β Q+1 ≥ … β L-1 ≥β L .U ss The eigenvector matrix representing the signal subspace, U ss =[u1, u2..., u Q-1 , u Q ]; The conjugate transposed matrix of the eigenvector matrix representing the signal subspace; Σ ss represents the eigenvalue matrix of the signal subspace, Σ ss= diag{β1, β2, ..., β Q-1 , β Q};U nn Represents the eigenvector matrix of the noise subspace, U nn = [u Q+1 , ..., u L-1 , u L ]; The conjugate transposed matrix of the eigenvector matrix representing the noise subspace; Σ nn represents the eigenvalue matrix of the signal-noise subspace, = diag{β Q+1 , ..., β L}. ss The signal subspace spanned by the columns of is orthogonal to that of U nn The noise subspace spanned by the columns of .

[0178] Step 2.4: Estimation of DOA of coherent incident signal. The spatial spectrum function of the MUSIC algorithm is:

[0179] (25)

[0180] In the formula, The array manifold matrix of the coherent signal, a(θ)=[a(θ1), a(θ2), ..., a(θ Q )], a(θ i ) represents the array steering vector corresponding to the i-th coherent signal, a(θ i ) = [1, e -jπsin(θi) , ..., e -j(L-1)πsin(θi) ] T ,1≤i≤Q; the superscript * indicates the conjugate transpose.

[0181] Perform a peak search on the result of formula (25), and the position of the spectrum peak is the angle θ of the stationary person relative to the radar.

[0182] Step 3: Convert the distance and angle of the stationary human target in polar coordinates obtained in step 2 into a radar point cloud converted to the Cartesian coordinate system, and then cluster the point cloud using the improved density-based spatial clustering of applications with noise (DBSCAN) algorithm. Based on the traditional DBSCAN algorithm, the similarity metric is improved to improve the relevance of the target point cloud, and then the point cloud is clustered by adaptively determining the key parameters of the clustering algorithm to obtain the centroid of the clustered radar point cloud. Step 3 is used to adaptively cluster the target radar point cloud.

[0183] As a preferred solution, the step 3 specifically includes:

[0184] Step 3.1: Coordinate conversion. The distance d and angle θ of the suspected target relative to the radar in the polar coordinate system obtained in step 2 are converted to the two-dimensional coordinates (x, y) in the Cartesian coordinate system, that is:

[0185] (26)

[0186] The two-dimensional coordinate (x, y) set obtained according to the above formula is the radar echo point cloud of the stationary person in the non-line-of-sight scenario.

[0187] Step 3.2: Improve the similarity metric. Since the radar echo point cloud of stationary personnel is sparse in non-line-of-sight scenarios, it is difficult to effectively cluster the point cloud and determine the target center of mass position. To this end, the relevance of the target point cloud is improved by improving the similarity metric between point clouds. The two-dimensional position coordinate information (x, y), micro-motion speed v and signal strength (Strength) of the stationary personnel target are used to enhance the relevance of the point cloud. Since there are absolute value differences between these four parameters, in order to ensure that each parameter has a significant impact on clustering, each parameter is normalized. The processing method is shown in the following formula:

[0188] (27)

[0189] Where: z norm represents the normalized value, z represents the original value; μ represents the mean of the data set, and σ represents the standard deviation of the data set.

[0190] For a single feature after normalization, the DBSCAN clustering algorithm uses the Euclidean distance to measure the distance between single point clouds, which is expressed as:

[0191] d = [ ( x n o r m _ 1 − x n o r m _ 2 ) 2 + ( y n o r m _ 1 − y n o r m _ 2 ) 2 + ( v n o r m _ 1 − v n o r m _ 2 ) 2 + ( S t r e n g t h n o r m _ 1 − S t r e n g t h n o r m _ 2 ) 2 ] 1 / 2 (28)

[0192] Among them, x, y, v and Strength represent the horizontal and vertical coordinates of the target, the micro-motion speed and the target signal strength respectively.

[0193] Step 3.3: Adaptively determine clustering parameters. The traditional DBSCAN algorithm does not take into account the uneven sampling density of millimeter-wave radars. It uses a fixed neighborhood radius Eps and a minimum number of points within the neighborhood radius minPts for core targets, which leads to significant differences in the clustering effects of targets at different distances. In response to this, a method for adaptively determining key clustering parameters that considers scene spatial characteristics is designed, where the minimum number of points within the neighborhood radius minPts is determined by the following formula:

[0194] (29)

[0195] Where dim is the dimension of the input data. Here, the input data includes the horizontal and vertical coordinates of the target, the micro-motion speed, and the target signal strength, so the value of dim is 4.

[0196] For the neighborhood radius Eps, as the distance between the target and the radar increases, the number of detected stationary human target points also decreases. Therefore, an adaptive threshold based on the distance of the target point cloud and the target positioning and direction characteristics is adopted, as shown in the following formula:

[0197] (30)

[0198] Among them, Δθ is the angular resolution of the radar, d is the distance from the point cloud target to the radar, and δ is the custom threshold coefficient, which ranges from 0.1 to 0.5.

[0199] Step 3.4: Point cloud clustering. Considering the spatial characteristics of the non-line-of-sight scene and adaptively determining the key parameters of the clustering algorithm, the DBSCAN algorithm is used to cluster the radar echo point cloud of the stationary person, and the centroid of the clustered radar point cloud is determined, which is the position of the stationary person in the non-line-of-sight target scene.

[0200] Step 4: Determine the position of the stationary human target in the non-line-of-sight target scene according to the coordinates corresponding to the centroid of the clustered radar point cloud. Step 4 is used to determine the target position.

[0201] Embodiment 2:

[0202] This embodiment introduces a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a non-line-of-sight stationary personnel positioning method as described in any one of Embodiments 1 is implemented.

[0203] Embodiment 3:

[0204] This embodiment introduces a computer device, including:

[0205] Memory, used to store instructions.

[0206] The processor is used to execute the instructions so that the computer device performs the operations of a non-line-of-sight stationary personnel positioning method as described in any one of Example 1.

[0207] Embodiment 4:

[0208] This embodiment introduces the specific implementation process of the method of the present invention as follows: Figure 1 As shown:

[0209] Enhance the micro-motion features of stationary personnel.

[0210] First, the intermediate frequency signal data of the millimeter-wave radar echo signal of the stationary person in the radar field of view is obtained. There is an obstacle between the radar and the stationary person. The person remains stationary except for normal breathing, and can maintain postures such as standing, sitting, squatting, and lying flat.

[0211] Secondly, if Figure 2-3 As shown, each frame of the acquired intermediate frequency signal data is constructed into a matrix form, and a two-dimensional FFT transformation is performed on it to obtain a distance-speed diagram. The matrix after the two-dimensional FFT transformation is subjected to phasor mean cancellation to eliminate static clutter interference. Schematic diagram of the phasor mean cancellation method.

[0212] Secondly, the near-zero distance channel zeroing method is used to remove DC component interference.

[0213] Secondly, if Figure 4 As shown in the figure, the coherent accumulation method is used to effectively improve the signal quality without increasing the computational complexity, while highlighting the micro-motion characteristic signal, thereby determining the distance unit where the dynamic target is located. As a comparison, it is compared with the traditional MTI (moving target indication) and zero channel zeroing method. Compared with (d) and (e) and (f), it can be seen that the noise level in the signal spectrum matrix processed by the micro-motion feature enhancement method of the present invention is significantly reduced, and the micro-motion characteristic signal amplitude is significantly enhanced, which reflects the superiority of the method of the present invention.

[0214] Perform coherent multipath decoupling on the echo signal.

[0215] Firstly, CFAR detection is used to determine the distance between the stationary human target and the radar, while eliminating false detections caused by dynamic clutter interference.

[0216] Secondly, the Doppler phase compensation is performed on the matrix to quickly estimate the micro-motion speed and direction of the stationary person.

[0217] Secondly, an enhanced super-resolution algorithm that performs Toeplitz reconstruction of the covariance matrix is ​​used to solve the coherent multipath problem and achieve accurate angle estimation. The radar antenna is arranged as follows Figure 5 The algorithm flow chart is shown as Figure 6 As shown. Figure 7 As shown, a comparison diagram of the calculation results of the enhanced super-resolution algorithm (CD-MUSIC) and other algorithms during the implementation of the method of the present invention.

[0218] Adaptive clustering of target radar point clouds.

[0219] First, the coordinates of the target radar echo point cloud are converted from polar coordinates to Cartesian coordinates to obtain two-dimensional position coordinate information. Figure 8As shown, during the implementation of the method of the present invention, the distance and angle of the stationary target relative to the radar in the polar coordinate system are converted to the Cartesian coordinate system to obtain the radar point cloud in two-dimensional coordinates, as well as the results of comparison with other algorithms.

[0220] like Fig. 9 As shown, common building materials and their relative distances from the radar are targeted during the implementation of the method of the present invention.

[0221] Secondly, the relevance of the target point cloud is improved by improving the similarity metric.

[0222] Secondly, the key parameters of the clustering algorithm are adaptively determined by considering the scene spatial features;

[0223] Secondly, the DBSCAN algorithm is used for point cloud clustering.

[0224] Determine the target location.

[0225] The centroid of the clustered radar point cloud is determined to be the position of the stationary person in the non-line-of-sight target scene.

[0226] In this embodiment, the Texas Instruments AWR1642BOOST millimeter wave radar development board is selected. The radar hardware parameter indicators mainly include: frequency modulation range 77~81 GHz, maximum theoretical frequency modulation bandwidth B max Up to 4 GHz, with 2 transmit antennas and 4 receive antennas. Radar parameters include the number of transmit antennas N Tx is 2, the number of receiving antennas is N Rx is 4, the number of array elements L is 8, the frequency modulation bandwidth B is 3.07 GHz, and the starting frequency f c is 77.42 GHz, the frequency modulation slope K is 60.01 MHz / us, and the frequency modulation period T c 192 μs, the number of frequency modulation cycles per frame is N chirp is 128, the number of samples per frequency modulation cycle is N ADC is 128, frame period T f is 40 ms, ADC sampling rate F s is 2.5 MHz, the total number of frames M is 128, the carrier wavelength λ is 3.8 mm, and the speed of light c is 3×10 8 m / s, from which the distance resolution d of the FMCW millimeter wave radar is determined res is 4.8 cm, and the velocity resolution v res is 0.15 m / s, the angular resolution Δθ is 1°, the value of dim is 4, and the threshold coefficient δ is 0.3.

[0227] like Fig.10 As shown, Fig.10(a) Schematic diagram showing the distances of a stationary person relative to the radar when they are 0.8 m, 1.6 m, 2.4 m, 3.2 m and 4.0 m respectively. Fig.10 (b) shows the relative error of the method of the present invention for positioning a stationary person when the non-line-of-sight blocking materials are wood, glass, PE plastic board and gypsum board, respectively. For comparison, the positioning error in the line-of-sight condition (i.e. when there is no obstacle blocking) is also given.

[0228] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A non-line-of-sight stationary personnel positioning method, characterized in that: Specifically include: Obtain the intermediate frequency signal of the radar echo signal, and obtain the range-velocity spectrum matrix according to the intermediate frequency signal; Preprocessing the distance-velocity spectrum matrix to obtain a processed distance-velocity spectrum matrix; Performing constant false alarm rate detection on the processed distance-velocity spectrum matrix, detecting the distance unit index and velocity unit index of the stationary personnel target, obtaining the distance-velocity spectrum matrix of the stationary personnel target according to the distance unit index and velocity unit index of the stationary personnel target, and determining the distance and signal strength of the stationary personnel target in polar coordinates; Performing Doppler phase compensation on the range-velocity spectrum matrix of the stationary personnel target to obtain the compensated range-velocity spectrum matrix, and obtaining the micro-motion velocity of the stationary personnel target according to the compensated range-velocity spectrum matrix; The compensated range-velocity spectrum matrix is ​​reorganized according to the number of array elements, and coherent multipath decoupling is performed to obtain the angle of the stationary human target in polar coordinates; The distance and angle of the stationary human target in polar coordinates are converted into radar point clouds in Cartesian coordinates. According to the coordinates, micro-motion speed and signal strength of the radar point clouds, the radar point clouds in Cartesian coordinates are clustered to obtain the centroid of the clustered radar point clouds. The coordinates corresponding to the centroid of the clustered radar point cloud are used as the position of the stationary person target; The compensated range-velocity spectrum matrix is ​​reorganized according to the number of array elements, and coherent multipath decoupling is performed to obtain the angle of the stationary personnel target in polar coordinates, including: Step 1: Reorganize the compensated range-velocity spectrum matrix according to the number of array elements to form L×N chirp The matrix y of N chirp The number of Chirp signals transmitted in each frame period; Calculate the covariance matrix R of matrix y. The expression of covariance matrix R is as follows: R=E[yy H ]: Where E[-] represents the expected value, and the superscript H represents the conjugate transpose; Solve the eigenvalues ​​of the covariance matrix R and arrange the eigenvalues ​​in descending order as β1, β2, ..., β Q , β Q+1 , ..., β L-1 , β L , the eigenvectors corresponding to each eigenvalue are u1, u2, ..., u Q 、u Q+1 ,...,u L-1 、u L , where the subscript Q represents the number of coherent signals, and the subscript L represents the number of elements in the uniform antenna array; Step 2: Select the eigenvectors u1 and u2 corresponding to the first two largest eigenvalues ​​β1 and β2 to achieve decorrelation; let u1=[u 11 ,u 21 ,…,u (L-1)1 ,u L1 ] T ; The first vector u in the above formula 11 As a reference element, u 11 The correlation function r(k-1) between each element can be expressed as: Then we can get formula 1: Similarly, each element is related to u 11 The correlation function between them is expressed as r(-k+1): The second formula is: Combining equation 1 and equation 2, construct a new Toeplitz matrix Y1: Where r(k-1)=r * (-k+1); superscript * indicates conjugate transpose; For the Toeplitz matrix Y1, construct the corresponding unbiased estimation matrix YY1: YY1=Y1+JY1 * J; Where: J is an L×L commutative matrix, i.e. a matrix whose sub-diagonal elements are "1" and all other elements are "0"; Step 3: Select the first element u in u2 12 As a reference element, and through the same steps as step 2 above to obtain the unbiased estimation matrix YY2, that is: Step 4: Construct a new Toeplitz matrix Y based on the unbiased estimation matrices YY1 and YY2: Y=YY1+YY2; Get the eigenvalue decomposition of matrix Y: Where U represents the eigenvector matrix of matrix Y, U = [u1,…,u L ]; U H represents the conjugate transposed matrix of the eigenvector matrix U; Σ represents the eigenvalue diagonal matrix of the matrix Y, Σ=diag{β1,β2...,β Q ,β Q+1 ...β L ,β L+1 }; U ss The eigenvector matrix representing the signal subspace, U ss =[u1,u2...,u Q-1 ,u Q ]; The conjugate transposed matrix of the eigenvector matrix representing the signal subspace; Σ ss represents the eigenvalue matrix of the signal subspace, Σ ss =diag{β1,β2,...,β Q-1 ,β Q }; U nn Represents the eigenvector matrix of the noise subspace, U nn =[u Q+1 ,...,u L-1 ,u L ]; The conjugate transposed matrix of the eigenvector matrix representing the noise subspace; Σ nn represents the eigenvalue matrix of the signal-noise subspace, Σ nn =diag{β Q+1 ,...,β L }; According to the eigenvalue decomposition results, the DOA of the coherent incident signal is estimated, and the spatial spectrum function is obtained as follows: Where a(θ) is the array manifold matrix of the coherent signal, a(θ) = [a(θ1), a(θ2), ..., a(θ Q )],a(θ i ) represents the angle θ of the i-th coherent signal i The corresponding array steering vector; The peak value of the spatial spectrum function is searched, and the position of the spectrum peak is the angle of the stationary person target in polar coordinates.

2. A non-line-of-sight stationary personnel positioning method according to claim 1, characterized in that: The intermediate frequency signal acquisition method of the radar echo signal specifically includes: The linear frequency modulation signal emitted by the radar to the target is obtained. The expression of the linear frequency modulation signal is as follows: Among them, S Tx (t) represents the linear frequency modulation signal, t represents time, A T represents the amplitude of the transmitted signal, f c is the starting frequency of the transmitted Chirp signal, B is the frequency modulation bandwidth of the transmitted Chirp signal, T c is the duration of transmitting a single Chirp signal, φ(t) is the phase noise of the transmitted signal, and π is the circumference of a circle; The echo signal of the target is obtained, and the expression of the echo signal is as follows: Among them, S Rx (t) represents the echo signal, α represents the attenuation coefficient of the amplitude during signal propagation, τ represents the delay of the received signal relative to the transmitted signal, and φ(t-τ) represents the phase noise of the echo signal; The linear frequency modulation signal is coherently mixed with the echo signal, and an intermediate frequency signal is output through a low-pass filter. The expression of the intermediate frequency signal is as follows: Wherein, S(t) represents the intermediate frequency signal.

3. A non-line-of-sight stationary personnel positioning method according to claim 2, characterized in that: The radar adopts a frequency modulated continuous wave millimeter wave radar.

4. The non-line-of-sight stationary personnel positioning method according to claim 1, characterized in that: The obtaining of the distance-velocity spectrum matrix according to the intermediate frequency signal specifically includes: The intermediate frequency signal is processed by discretization and frame division to obtain M frequency modulation period sequence signal matrix S i , the frequency modulation period sequence signal matrix S i The expression is as follows: S i =αA T (n,m)sin{2π[f b (n)+φ b (m)]n / F s }; Where n represents the fast time index within a single frequency modulation cycle, n = 0, ..., N ADC -1, where N ADC Indicates the number of sampling points in a single frequency modulation cycle; m represents the slow time index in a single frame, m = 0,...,N chirp -1, where N chirp The number of Chirp signals transmitted in each frame period; F s represents the ADC sampling rate of the intermediate frequency signal, (n,m) represents the dimension corresponding to the fast time index n and the slow time index m in the signal matrix Si, and f b (n) represents the frequency of the intermediate frequency signal corresponding to the fast time index, φ b (m) represents the phase of the intermediate frequency signal corresponding to the slow time index, i = 1,...,M; α represents the attenuation coefficient of the amplitude during signal propagation; A T Indicates the amplitude of the transmitted signal; For each frequency modulation cycle sequence signal matrix S i Perform frequency domain transformation in the fast time index direction of the representative column to obtain the two-dimensional distance spectrum matrix SR i , where the fast time index direction is divided into N adc distance intervals, the distance interval is the distance resolution R res ; For the two-dimensional distance spectrum matrix SR i Perform frequency domain transformation in the slow time index direction of the representative row to obtain the range-velocity spectrum matrix SRD i , where the slow time index direction is divided into N chirp speed intervals, which is the speed resolution.

5. The non-line-of-sight stationary personnel positioning method according to claim 1, characterized in that: The preprocessing of the distance-velocity spectrum matrix to obtain the processed distance-velocity spectrum matrix specifically includes: Remove each range-velocity spectrum matrix SRD i After the environmental static clutter is removed, the DC component interference is removed to obtain the matrix after the interference is eliminated; The matrix after interference elimination is coherently accumulated to obtain the processed range-velocity spectrum matrix.

6. A non-line-of-sight stationary personnel positioning method according to claim 5, characterized in that: The removal of each range-velocity spectrum matrix SRD i After the environmental static clutter is removed, the DC component interference is removed to obtain the matrix after the interference is eliminated; The interference-eliminated matrix is ​​coherently accumulated to obtain the processed range-velocity spectrum matrix, which specifically includes: The vector mean elimination method is used to remove each range-velocity spectrum matrix SRD i The static clutter in the environment is removed by using the zero channel zeroing method for each range-velocity spectrum matrix SRD i The DC component interference in the matrix is ​​eliminated to obtain the matrix after the interference is eliminated; Superimpose M interference-eliminated matrices to obtain the processed range-velocity spectrum matrix x.

7. The non-line-of-sight stationary personnel positioning method according to claim 1, characterized in that: The method of performing constant false alarm rate detection on the processed distance-velocity spectrum matrix, detecting the distance unit index and the velocity unit index of the stationary personnel target, obtaining the distance-velocity spectrum matrix of the stationary personnel target according to the distance unit index and the velocity unit index of the stationary personnel target, and determining the distance and signal strength of the stationary personnel target in polar coordinates specifically includes: Select a set of reference cells from the cells to be tested in the processed range-velocity spectrum matrix x; estimating the background noise level of the processed range-velocity spectrum matrix x using the selected reference cells; Calculate an adaptive detection threshold based on the estimated background noise level and a preset false alarm rate; The signal strength of the unit to be tested is compared with the calculated adaptive detection threshold. If the signal amplitude of the unit to be tested exceeds the adaptive detection threshold, it is determined to be a stationary human target; otherwise, it is determined to be noise or clutter; According to the stationary personnel target, the distance unit index and the speed unit index of the stationary personnel target are obtained, and according to the distance unit index and the speed unit index of the stationary personnel target, the distance-speed spectrum matrix of the stationary personnel target is obtained, and the distance and signal strength of the stationary personnel target in polar coordinates are determined.

8. The non-line-of-sight stationary personnel positioning method according to claim 1, characterized in that: The method of converting the distance and angle of the stationary human target in polar coordinates into a radar point cloud in a Cartesian coordinate system, clustering the radar point cloud in the Cartesian coordinate system according to the coordinates, micro-motion speed, and signal strength of the radar point cloud to obtain the centroid of the clustered radar point cloud specifically includes: The distance and angle of the stationary human target in polar coordinates are converted into a radar point cloud in a Cartesian coordinate system. The coordinate value of the radar point cloud in the Cartesian coordinate system is (x, y), and the expression is as follows: Where d represents the distance in polar coordinates, and θ represents the angle in polar coordinates; The horizontal coordinate x, the vertical coordinate y, the micro-motion speed v, and the signal strength Strength of the stationary human target are normalized to obtain the normalized horizontal coordinate x, the vertical coordinate y, the micro-motion speed v, and the signal strength Strength; The Euclidean distance between two points in the radar echo point cloud is calculated using the normalized horizontal coordinate x, vertical coordinate y, micro-motion speed v, and signal strength Strength; Calculate the minimum number of points minPts within the neighborhood radius. The expression is as follows: minPts=2dim-1; Among them, dim is the dimension of the input data; Calculate the neighborhood radius Eqs, the expression is as follows: Eqs = Δθ × d × δ; Among them, Δθ is the angular resolution of the radar, d is the distance from the point cloud target to the radar, and δ is the custom threshold coefficient; The Euclidean distance between two points in the radar echo point cloud, the minimum number of points within the neighborhood radius minPts, and the neighborhood radius Eqs are input into the DBSCAN algorithm to cluster the radar echo point cloud of stationary personnel and obtain the centroid of the clustered radar point cloud.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, a non-line-of-sight stationary personnel positioning method as claimed in any one of claims 1 to 8 is implemented.

10. A computer device, characterized in that: include: A memory for storing instructions; The processor is used to execute the instructions so that the computer device performs the operations of the non-line-of-sight stationary personnel positioning method as described in any one of claims 1 to 8.

Images