CN106093934A - Multiple target location estimation method after through-wall radar imaging based on improvement dynamic programming - Google Patents

Abstract

The invention discloses multiple target location estimation method after a kind of through-wall radar imaging based on improvement dynamic programming；It includes that target area imaging, target image optimization, dynamic programming and sliding window smooth position are estimated.The present invention has fully taken into account target autgmentability in imaging results, improve the track rejection problem caused because of scattering coefficient difference, target echo decay, target occlusion etc. between target well, improve target detection probability, there is the advantage that target detection probability is high, position estimation accuracy is higher, can be applied directly in through-wall radar equipment.

Description

Multiple target location estimation method after through-wall radar imaging based on improvement dynamic programming

Technical field

The invention belongs to through-wall radar technical field, particularly relate to a kind of based on the through-wall radar imaging improving dynamic programming Rear multiple target location estimation method.

Background technology

Through-wall radar is to utilize ultra-wideband microwave the radar exploration technique, penetrates building by launching the electromagnetic wave of special frequency channel Thing body of wall, blindage etc., the scatter echo data receiving the targets such as human body after barrier realize the spy to concealed target acquisition of information Plant equipment.Due to its good characteristic, through-wall radar obtains widely in fields military, civil such as street fighting, anti-terrorism, disaster reliefs Pay close attention to.When actual application scenarios exists multiple target, due to the difference of target scattering coefficient, target travel characteristic not With and target occlusion problem, often there is larger difference in the amplitude between target in target imaging result.And now entering Then it is difficult to take into account target complete according to traditional fixed threshold detection when of row target detection, thus causes track rejection, Target location estimated accuracy is greatly lowered.Therefore, in actual multiple target location estimation application, each in taking into account detection scene Target, the key issue of multiple target location estimation after accurately the multiple targets of detection are through-wall radar imaging.

To the research in terms of through-wall radar multiple target location estimation, domestic and international research institution has pointed out many solutions. Such as, multiple low mobility smooth target in obtaining search coverage by improvement Kalman and joint probabilistic data association algorithm Flight path；The high maneuverability target utilizing interacting multiple model algorithm to change greatly kinetic characteristic carries out target location estimation.But Above method needs to carry out on the premise of target is the most by the most accurately detection, does not accounts for because of scattering coefficient between target The track rejection problem that the decay of difference, target echo, target occlusion etc. cause.From the point of view of the documents and materials published, at present Also not for due to after imaging target amplitude differ greatly the through-wall radar multiple target position causing target seriously to lose problem Method of estimation, therefore, after studying a kind of through-wall radar imaging based on improvement dynamic programming, multiple target location estimation method is being worn Wall radar target location estimation method has important actual application value.

Summary of the invention

The goal of the invention of the present invention is: in order to solve in prior art due to target amplitude differ greatly, echo attenutation journey It is a kind of based on the thunder through walls improving dynamic programming that degree difference, target occlusion etc. cause the problems such as track rejection, the present invention to propose Reach as rear multiple target location estimation method.

The technical scheme is that multiple target location estimation after a kind of through-wall radar imaging based on improvement dynamic programming Method, comprises the following steps:

A, the multiple mobile object utilizing through-wall radar to be treated after detecting region body of wall by MIMO array are detected, According to calculating resolution imaging region carried out discretization, and according to thickness of wall body, the dielectric constant in region to be detected and connect Each antenna imaging and focusing is postponed to compensate by receipts machine cable offset, then uses rear orientation projection's method to become imaging region Picture, the target area imaging array I after being compensated_k(X,Y)；

B, will step A compensate after target area imaging array reject graing lobe clutter, then by continuous some frame target figures As corresponding pixel points is multiplied, and carry out two-dimensional low pass ripple process, obtain the target image matrix D after imaging optimizes_k(X,Y)；

C, obtain continuous some frame imaging optimization according to step B after target image matrix be extended desired value function amass Tired, then the desired value function after accumulation is carried out target trajectory backtracking, and reject false track；

D, judge the current time testing result shape corresponding with the targetpath that corresponding target has been recovered of each target successively Whether state gap is in default range of error；If in default range of error, then last frame current time sliding window detected Testing result is as the state estimation result of current time correspondence target；If not in default range of error, then when utilizing current The state of the corresponding target of current time is estimated by the front some frame testing results carving sliding window detection.

Further, the target area imaging array after described step B will compensate in step A rejects graing lobe clutter, then will Continuous some frame target image corresponding pixel points are multiplied, and carry out two-dimensional low pass ripple process, obtain the target after imaging optimizes Image array D_k(X, Y), specifically includes following steps:

B1, according in step A compensate after target area imaging array, try to achieve the target location corresponding to max pixel value (x_u,y_u), and calculate the angle theta of this target and y-axis, obtain the preliminary graing lobe position (x of target_g,y_g)；Again with preliminary graing lobe position It is set to the Δ at center_gIn the range of position corresponding to search pixel maximum, obtain the actual graing lobe position (x ' of target_g,y′_g), and By the Δ centered by actual graing lobe position_gIn the range of element zero setting, obtain reject graing lobe clutter target imaging matrix I_k′ (X,Y)；

B2, by step B1 reject graing lobe clutter target imaging matrix in continuous some frame target image corresponding pixel points It is multiplied, obtains k moment target imaging matrix A to be filtered_k(X,Y)；

B3, two dimension is set according to theoretical resolution, image resolution ratio, antenna wavelength, antenna aperature and imaging region length The Range Profile cut-off frequency of low-pass filtering and orientation are to cut-off frequency, to target imaging matrix A to be filtered in step B2_k(X, Y) carry out two-dimensional low pass ripple process, obtain the target image matrix D after imaging optimizes_k(X,Y)。

Further, the target imaging matrix A that in described step B2, the k moment is to be filtered_k(X, Y) is embodied as:

A_k(X, Y)=I_k′_-2(X,Y).*I_k′_-1(X,Y).*I_k′(X,Y)

Wherein, I '_k(X,Y),I′_k-1(X,Y),I′_k-2(X, Y) is respectively k, the target imaging that k-1, the k-2 moment is to be filtered Matrix .* is point multiplication operation.

Further, the target image matrix D after imaging optimizes in described step B3_k(X, Y) is embodied as:

$D_k(X,Y)=A_k(X,Y)\⊗H(X,Y)$

Wherein, H (X, Y) two-dimensional low pass wave system system Jacobian matrix,Represent two-dimensional convolution symbol.

Further, the target image matrix D after described step C optimizes according to imaging in step B_k(X, Y) Extended target Value function accumulates, then the desired value function after accumulation is carried out target trajectory backtracking, and rejects false track, specifically includes following Step:

C1, obtain continuous some frame imaging optimization according to step B after target image matrix be extended desired value function Accumulation, an elementary state and respective value functional value in setting value Jacobian matrix, path matrix, target image matrix, right respectively This elementary state subsequent time speed carries out target expansion and obtains transfering variable matrix, utilizes maximum institute in transfering variable matrix Corresponding element index updated value Jacobian matrix and path matrix, then to target image matrix in x direction, y direction, x direction speed Degree, y direction speed four dimensions scan for the value function matrix after being accumulated and path matrix；

In C2, obtaining step C1, the value function matrix median function value after accumulation is more than the element rope presetting detection threshold Draw, carry out flight path backtracking by the path matrix in step C1, obtain targetpath sequence cluster；

The targetpath sequence that in value function matrix after accumulating in C3, obtaining step C1, max function value is corresponding, will The Euclidean distance meeting two elementary states in targetpath sequence cluster is deleted more than other flight path sequences of allowable position error, Repeat above operation until the flight path sequence number of targetpath bunch is 0, obtain pending flight path sequence cluster.

Further, described step C1 is turned by introducing target spreading factor and Extended target shape mask matrix Shifting matrix of variables is embodied as:

$\mathrm{Temp}=D_{i_2+1}(x_a+{\mathrm{\Δ}}_1+v_x^{\′}-q:x_a+{\mathrm{\Δ}}_1+v_x^{\′}+q,y_b+{\mathrm{\Δ}}_2+v_y^{\′}-q:y_b+{\mathrm{\Δ}}_2+v_y^{\′}+q)$

J(Δ₁,Δ₂)=Sum (Temp.*Mock)

Wherein, Temp matrix represents by target image matrixMiddle xth_a+Δ₁+v_x'-q row is to x_a+Δ₁+v_x′+ Q row, y_b+Δ₂+v′_y-q arranges y_b+Δ₂+v′_yThe matrix that size is (2q+1) * (2q+1) of+q column element composition, x_aRepresent The position in target x direction, y_bRepresent the position in target y direction, Δ₁,Δ₂Represent x direction and the machine of y direction kinetic characteristic respectively Reason, v_x′,v′_yFor target subsequent time x direction and y direction speed, q is target spreading factor, and Mock represents Extended target Shape mask matrix, all elements in matrix is added by Sum () expression one by one, J (Δ₁,Δ₂) represent transfering variable matrix.

The invention has the beneficial effects as follows: the present invention carries out fast short-term training according to detection scenario parameters and empirical value to search coverage Picture, obtains the imaging results of better quality, then by rejecting graing lobe, multiframe is even taken advantage of, two-dimensional low pass ripple processes and can tie imaging Fruit is further optimized, and facilitates subsequent dynamic planning to carry out objective accumulation；Fully take into account target in imaging results Autgmentability, improve the target caused because of scattering coefficient difference, target echo decay, target occlusion etc. between target well Loss problem, improves target detection probability, has the advantage that target detection probability is high, position estimation accuracy is higher, can be direct It is applied in through-wall radar equipment.

Accompanying drawing explanation

Fig. 1 be the present invention based on improve dynamic programming through-wall radar imaging after multiple target location estimation method flow process show It is intended to.

Fig. 2 is through-wall radar array layout structural representation in the embodiment of the present invention.

Fig. 3 is target imaging result schematic diagram in the embodiment of the present invention.

Fig. 4 is imaging optimum results schematic diagram in the embodiment of the present invention.

Fig. 5 is estimated result schematic diagram in target location in the embodiment of the present invention.

Detailed description of the invention

In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, right The present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, not For limiting the present invention.

As it is shown in figure 1, be multiple target location estimation side after the through-wall radar imaging based on improvement dynamic programming of the present invention Method schematic flow sheet.Multiple target location estimation method after a kind of through-wall radar imaging based on improvement dynamic programming, including following Step:

A, the multiple mobile object utilizing through-wall radar to be treated after detecting region body of wall by MIMO array are detected, According to calculating resolution imaging region carried out discretization, and according to thickness of wall body, the dielectric constant in region to be detected and connect Each antenna imaging and focusing is postponed to compensate by receipts machine cable offset, then uses rear orientation projection's method to become imaging region Picture, the target area imaging array I after being compensated_k(X,Y)；

B, will step A compensate after target area imaging array reject graing lobe clutter, then by continuous some frame target figures As corresponding pixel points is multiplied, and carry out two-dimensional low pass ripple process, obtain the target image matrix D after imaging optimizes_k(X,Y)；

C, obtain continuous some frame imaging optimization according to step B after target image matrix be extended desired value function amass Tired, then the desired value function after accumulation is carried out target trajectory backtracking, and reject false track；

D, judge the current time testing result shape corresponding with the targetpath that corresponding target has been recovered of each target successively Whether state gap is in default range of error；If in default range of error, then last frame current time sliding window detected Testing result is as the state estimation result of current time correspondence target；If not in default range of error, then when utilizing current The state of the corresponding target of current time is estimated by the front some frame testing results carving sliding window detection.

In step, after the present invention utilizes through-wall radar to use two eight receipts arrays to treat one layer of region of detection body of wall Multiple mobile object detects, and imaging region is (x₁,x₂)×(y₁,y₂) rectangular area；Re-use calculating resolutionDiscrete Changing imaging region and initialisation image matrix is I (X, Y), wherein Δ r is theoretical resolution, and can obtain matrix size is M*N；Root again According to detecting thickness of wall body d, the DIELECTRIC CONSTANT ε of scene and connecing receiver length of cable, obtain each group of antenna by empirical value and become Image focu delay compensation value [τ₁′₁,τ₁′₂,τ₁′₃,...,τ₁′₈,τ₂′₁,τ₂′₂,τ₂′₃,...,τ₂′₈].Jth reception antenna connects The i-th received launches echo-signal Y that antenna is launched_ij(t) be:

$Y_{ij}\left(t\right)=s(t-{\mathrm{\τ}}_{ij,x_h,y_h})$

Wherein, s (t) is launched stepped frequency signal,For target echo propagation delay, it is represented by:

${\mathrm{\τ}}_{ij,x_h,y_h}=r_{ij(x_h,y_h)}/c$

Wherein,It is pixel (x_h,y_h) launch the distance of antenna with pixel (x to i-th_h,y_h) to jth The distance sum of reception antenna, c is the light velocity.

Imaging region is used rear orientation projection's method to carry out imaging by the present invention again, is expressed as:

$I_k(x_h,y_h)=\underset{i=1}{\overset{2}{\Σ}}\underset{j=1}{\overset{8}{\Σ}}{Y}_{ij}(t+{\mathrm{\τ}}_{ij}^{\′}){|}_{t=0}$

Wherein, Σ is summation sign, | for assignment, t is time variable, x_h∈ (1, M), y_h∈(1,N).By to one-tenth K moment target imaging matrix I is i.e. can get as pixels all in region carry out aforesaid operations_k(X,Y)。

As in figure 2 it is shown, be through-wall radar array layout structural representation in the embodiment of the present invention, the present invention uses one two Sending out the Step Frequency through-wall radar of eight receipts configurations, Through-Wall Radar System 10 meters from wall is placed on centre position, launches 1GHz-2GHz's Stepped frequency continuous wave signal, stepped intervals is 2MHz, and two movement human targets of body of wall opposite side are implemented through-wall detection；Root Estimating that thickness of wall body is 0.25m according to prior information, body of wall dielectric constant is about 9.Search coverage be (-4m～4m) * (10m～ 20m), image resolution ratio is 0.075²m²/ pixel, after discretization image matrix, image array dimension is 107*201, passes through Fast imaging algorithm, carries out imaging to imaging region, as it is shown on figure 3, be target imaging result schematic diagram in the embodiment of the present invention.

In stepb, the target area imaging array after compensating in step A rejects graing lobe clutter, then will be the most some Frame target image corresponding pixel points is multiplied, and carries out two-dimensional low pass ripple process, obtains the target image matrix after imaging optimizes D_k(X, Y), specifically includes following steps:

B1, according in step A compensate after target area imaging array, try to achieve the target location corresponding to max pixel value (x_t,y_t), and calculate the angle theta of this target and y-axis, obtain the preliminary graing lobe position (x of target_g,y_g)；Again with preliminary graing lobe position It is set to the Δ at center_gIn the range of position corresponding to search pixel maximum, obtain the actual graing lobe position (x ' of target_g,y′_g), and By the Δ centered by actual graing lobe position_gIn the range of element zero setting, obtain reject graing lobe clutter target imaging matrix I_k′ (X,Y)；

B2, by step B1 reject graing lobe clutter target imaging matrix in continuous some frame target image corresponding pixel points It is multiplied, obtains k moment target imaging matrix A to be filtered_k(X,Y)；

B3, two dimension is set according to theoretical resolution, image resolution ratio, antenna wavelength, antenna aperature and imaging region length The Range Profile cut-off frequency of low-pass filtering and orientation are to cut-off frequency, to target imaging matrix A to be filtered in step B2_k(X, Y) carry out two-dimensional low pass ripple process, obtain the target image matrix D after imaging optimizes_k(X,Y)。

In step bl is determined., according to the target area imaging array after compensating in step A, try to achieve corresponding to max pixel value Target location (x_u,y_u), and calculate the angle theta of this target and y-axis, it is expressed as:

θ=atan (x_u/y_u)

The preliminary graing lobe position (x of target is can get according to its angle theta size_g,y_g), it is expressed as:

Again with preliminary graing lobe position (x_g,y_gΔ centered by)_gIn the range of, i.e. at matrix I (x_g-Δ_g:x_g+Δ_g,y_g-Δ_g: y_g+Δ_gThe position that in), search pixel maximum is corresponding, obtains the actual graing lobe position (x ' of target_g,y′_g), and will be with actual grid Flap position (x '_g,y′_gΔ centered by)_gIn the range of pixel value zero setting, wherein I (x_g-Δ_g:x_g+Δ_g,y_g-Δ_g:y_g+Δ_g) table Show by xth in I (X, Y) matrix_g-Δ_gRow arrives x_g+Δ_gOK, y_g-Δ_gRow arrive y_g+Δ_gThe size of column element composition is (2 Δs_g+ 1)*(2Δ_g+ 1) submatrix, Δ_gIt is the fault-tolerant factor in graing lobe position, thus obtains rejecting the target imaging matrix of graing lobe clutter I_k′(X,Y)。

In step B2, by the continuous 3 frame target images in the target imaging matrix rejecting graing lobe clutter of step B1 Corresponding pixel points is multiplied, and is expressed as:

A_k(X, Y)=I_k′_-2(X,Y).*I_k′_-1(X,Y).*I_k′(X,Y)

Wherein, I '_k(X,Y),I′_k-1(X,Y),I′_k-2(X, Y) is respectively k, the target imaging that k-1, the k-2 moment is to be filtered Matrix .* is point multiplication operation；By pixels all in imaging region carry out aforesaid operations, i.e. to can get the k moment to be filtered Target imaging matrix A_k(X,Y)。

In step B3, according to theoretical resolution, image resolution ratio, antenna wavelength, antenna aperature, imaging region length etc. The Range Profile cut-off frequency w of two-dimensional low pass ripple is set_rcWith orientation to cut-off frequency w_crc, thus obtain two-dimensional low pass ripple FIR The system function matrix of wave filter is H (X, Y), to target imaging matrix A to be filtered in step B2_kIt is low that (X, Y) carries out two dimension Pass filter processes, and suppresses high frequency spurs, obtains the target image matrix D after imaging optimizes_k(X, Y), is expressed as:

$D_k(X,Y)=A_k(X,Y)\⊗H(X,Y)$

Wherein,Represent two-dimensional convolution symbol.

Preferably, the present invention is by fault-tolerant for graing lobe position factor Δ_gBeing set to 5, wherein two-dimensional low-pass filter orientation is to resolution Rate w_crc=0.036*2 π, range resolution w_c=0.05*2 π, through rejecting graing lobe, successive frame is multiplied, two-dimensional low pass ripple Process, obtain the target image matrix D after imaging optimizes_k(X, Y), as shown in Figure 4, ties for imaging optimization in the embodiment of the present invention Really schematic diagram.

In step C, the target image matrix after obtaining continuous some frame imaging optimization according to step B is extended target Value function accumulates, then the desired value function after accumulation is carried out target trajectory backtracking, and rejects false track, specifically includes following Step:

C1, obtain continuous some frame imaging optimization according to step B after target image matrix be extended desired value function Accumulation, an elementary state and respective value functional value in setting value Jacobian matrix, path matrix, target image matrix, right respectively This elementary state subsequent time speed carries out target expansion and obtains transfering variable matrix, utilizes maximum institute in transfering variable matrix Corresponding element index updated value Jacobian matrix and path matrix, then to target image matrix in x direction, y direction, x direction speed Degree, y direction speed four dimensions scan for the value function matrix after being accumulated and path matrix；

In C2, obtaining step C1, the value function matrix median function value after accumulation is more than the element rope presetting detection threshold Draw, carry out flight path backtracking by the path matrix in step C1, obtain targetpath sequence cluster；

The targetpath sequence that in value function matrix after accumulating in C3, obtaining step C1, max function value is corresponding, will Targetpath sequence cluster meetsOther flight path sequences delete, repeat above operation until The flight path sequence number of targetpath bunch is 0, obtains pending flight path sequence cluster.

In step C1, obtain the target image square needed to carry out after the continuous 6 frame imagings optimizations of batch processing according to step B Battle arraySetting value Jacobian matrix is I_k,max(X,Y,V_x,V_y), path matrix isIts Being 4 dimension matrixes, 4 dimensions represent position, target x direction, position, target y direction, target x direction speed and target y direction respectively Speed；In target setting image array, some elementary state isx_aRepresent target x direction Position, y_bRepresent the position in target y direction, v_xaRepresent the speed in target x direction, v_ybRepresent the speed in y direction, its respective value letter Numerical value isAssume that this elementary state subsequent time speed is for (v '_x,v′_y), then have:

$Temp=D_{i_2+1}(x_a+{\mathrm{\Δ}}_1+v_x^{\′}-q:x_a+{\mathrm{\Δ}}_1+v_x^{\′}+q,y_b+{\mathrm{\Δ}}_2+v_y^{\′}-q:y_b+{\mathrm{\Δ}}_2+v_y^{\′}+q)$

J(Δ₁,Δ₂)=Sum (Temp.*Mock)

Wherein, Temp matrix represent byXth in matrix_a+Δ₁+v_x'-q row is to x_a+Δ₁+v_x'+q row, y_b +Δ₂+v′_y-q arranges y_b+Δ₂+v′_y+ q column element composition the matrix that size be (2q+1) * (2q+1), q be target extend because of Son, Mock represents Extended target shape mask matrix, and relevant with goal-selling shape .* representing matrix corresponding element is multiplied, Sum All elements in matrix is added by () expression one by one, Δ₁,Δ₂Represent x direction and the motor-driven factor of y direction kinetic characteristic, J (Δ₁,Δ₂) matrix representative transfering variable matrix.

Make (Δ_x,Δ_y) it is J (Δ₁,Δ₂) element index corresponding to maximum in matrix, then have its correspondence at D_i2+1 The element index of (X, Y) is x '_a=x_a+Δ_x+v_x′,y_b'=y_b+Δ_y+v′_y；Thus dbjective state is updated toUpdated value Jacobian matrix with path matrix is:

$I_{k,max}\left(S_{i_2+1,x_a,y_b}\right)=I_{k,max}\left(S_{i_2,x_a,y_b}\right)+J({\mathrm{\Δ}}_x,{\mathrm{\Δ}}_y)$

${\mathrm{Track}}_{i_2}\left(S_{i_2+1,x_a,y_b}\right)=S_{i_2,x_a,y_b}$

Again to target image matrix in x direction, y direction, x direction speed, y direction speed four dimensions scan for obtaining Value function matrix I after accumulation_k,max(X,Y,V_x,V_y) and path matrix

Preferably, it is contemplated that the kinetic characteristic of target, x, y direction speed search scope is set and is-5m/s to 5m/s, interval It is 1m/s.X direction and the motor-driven factor Δ of y direction kinetic characteristic₁,Δ₂In the range of-4 to 4, it is spaced apart 1.Consider that target is in imaging Result shows as approximate ellipsoidal, therefore sets Extended target shape mask matrixAnd mesh Mark spreading factor q=2.

Value function matrix I in step C2, after accumulating in obtaining step C1_k,max(X,Y,V_x,V_y) median function value is big In default detection threshold V_tElement indexPass through path matrixCarry out 5 flight path backtrackings i.e.Obtain 1 targetpath sequence comprising 6 frame dbjective statesWherein num is flight path sequence bar number；Repeat above operation then can preliminary to obtain To targetpath sequence cluster C={E₁,E₂,...,E_num}.Preferably, detection threshold V is preset_t=1500.

In step C3, target setting is the most overlapping, the value function matrix I after accumulating in obtaining step C1_k,max(X,Y,V_x, V_yThe targetpath sequence that in), max function value is correspondingBy other boats in targetpath sequence cluster Mark sequence table is shown asIf flight path sequenceMeet

$dis(S_{i_2,x_a,y_b},S_{i_2,x_a,y_b}^{\′})>R_{max}$

And flight path sequenceIn meet the frame number of condition more than 5, then by flight path sequenceDelete from flight path sequence cluster C, R_maxFor allowable position error, dis () represents the Euclidean distance of two elementary states；Again by the remaining boat meeting condition Mark sequence mark is U₁,U₂,...,U_n, removal flight path sequence cluster C also joins pending flight path sequence cluster T.Repeat above behaviour Make, until flight path bunch C flight path sequence number is 0, to obtain pending flight path sequence cluster T={U₁,U₂,...U_n, wherein n is flight path Sequence bar number, is also target number.Preferably, allowable position error R_max=5.

In step D, the target boat that the current time testing result of each target has been recovered with corresponding target is judged successively Whether mark corresponding states gap is in default range of error；For m-th target, current time testing result isThe targetpath that corresponding target has been recovered isIf the detection of present frame sliding window Before front 5 frame dbjective states are corresponding with confirming targetpath, the state difference of 5 frames is away from range of error, the most satisfied:

$\underset{i_2=k-5}{\overset{k-1}{\Σ}}\mathrm{\δ}\left(dis\right(S_{i_2,m},S_{i_2,m}^{\′})-R_{max})\≤5$

Wherein,R_maxFor allowable position error；Then with the last frame inspection of current sliding window detection Survey result as the current time state estimation result to target, it may be assumed that

S_k′_,m=S_k,m；

If the state difference of 5 frames is not away from before the front 5 frame dbjective states of present frame sliding window detection are corresponding with confirming targetpath In range of error, then utilize the testing result of front 3 frames that current sliding window detects that the dbjective state of target current time is carried out Estimate, be expressed as:

S_k′_,m=S_k-1,m+S_k-2,m-S_k-3,m

All targetpath sequences are repeated above operation, i.e. can get final goal location estimation value.As it is shown in figure 5, For estimated result schematic diagram in target location in the embodiment of the present invention.

Those of ordinary skill in the art it will be appreciated that embodiment described here be to aid in reader understanding this Bright principle, it should be understood that protection scope of the present invention is not limited to such special statement and embodiment.This area It is each that those of ordinary skill can make various other without departing from essence of the present invention according to these technology disclosed by the invention enlightenment Planting concrete deformation and combination, these deform and combine the most within the scope of the present invention.

Claims (6)

1. multiple target location estimation method after a through-wall radar imaging based on improvement dynamic programming, it is characterised in that include Following steps:

A, the multiple mobile object utilizing through-wall radar to be treated after detecting region body of wall by MIMO array are detected, according to Calculating resolution carries out discretization to imaging region, and according to thickness of wall body, dielectric constant and the receiver in region to be detected Each antenna imaging and focusing is postponed to compensate by cable offset, then uses rear orientation projection's method to carry out imaging imaging region, Target area imaging array I after being compensated_k(X,Y)；

B, will step A compensate after target area imaging array reject graing lobe clutter, then by continuous some frame target images pair Answer pixel to be multiplied, and carry out two-dimensional low pass ripple process, obtain the target image matrix D after imaging optimizes_k(X,Y)；

C, obtain continuous some frame imaging optimization according to step B after target image matrix be extended desired value function accumulation, Again the desired value function after accumulation is carried out target trajectory backtracking, and reject false track；

D, judge that the targetpath corresponding states that the current time testing result of each target has been recovered with corresponding target is poor successively Away from whether in default range of error；If in default range of error, then last frame detection current time sliding window detected Result is as the state estimation result of current time correspondence target；If not in default range of error, then utilize current time sliding The state of the corresponding target of current time is estimated by front some frame testing results of window detection.

2. multiple target location estimation method after through-wall radar imaging based on improvement dynamic programming as claimed in claim 1, its Being characterised by, the target area imaging array after described step B will compensate in step A rejects graing lobe clutter, then will be the most some Frame target image corresponding pixel points is multiplied, and carries out two-dimensional low pass ripple process, obtains the target image matrix after imaging optimizes D_k(X, Y), specifically includes following steps:

B1, according in step A compensate after target area imaging array, try to achieve the target location (x corresponding to max pixel value_u, y_u), and calculate the angle theta of this target and y-axis, obtain the preliminary graing lobe position (x of target_g,y_g)；With preliminary graing lobe position it is again The Δ at center_gIn the range of position corresponding to search pixel maximum, obtain the actual graing lobe position (x ' of target_g,y′_g), and will be with Δ centered by actual graing lobe position_gIn the range of element zero setting, obtain reject graing lobe clutter target imaging matrix I '_k(X, Y)；

B2, by step B1 reject graing lobe clutter target imaging matrix in continuous some frame target image corresponding pixel points phases Take advantage of, obtain k moment target imaging matrix A to be filtered_k(X,Y)；

B3, according to theoretical resolution, image resolution ratio, antenna wavelength, antenna aperature and imaging region length arrange two dimension low pass The Range Profile cut-off frequency of filtering and orientation are to cut-off frequency, to target imaging matrix A to be filtered in step B2_k(X, Y) enters Row two-dimensional low pass ripple processes, and obtains the target image matrix D after imaging optimizes_k(X,Y)。

3. multiple target location estimation method after through-wall radar imaging based on improvement dynamic programming as claimed in claim 2, its It is characterised by, the target imaging matrix A that in described step B2, the k moment is to be filtered_k(X, Y) is embodied as:

A_k(X, Y)=I '_k-2(X,Y).*I′_k-1(X,Y).*I′_k(X,Y)

Wherein, I '_k(X,Y),I′_k-1(X,Y),I′_k-2(X, Y) is respectively k, the target imaging square that k-1, the k-2 moment is to be filtered Battle array .* is point multiplication operation.

4. multiple target location estimation method after through-wall radar imaging based on improvement dynamic programming as claimed in claim 3, its It is characterised by, the target image matrix D after imaging optimizes in described step B3_k(X, Y) is embodied as:

$D_k(X,Y)=A_k(X,Y)\⊗H(X,Y)$

Wherein, H (X, Y) two-dimensional low pass wave system system Jacobian matrix,Represent two-dimensional convolution symbol.

5. multiple target location estimation method after through-wall radar imaging based on improvement dynamic programming as claimed in claim 4, its Being characterised by, the target image matrix after described step C obtains continuous some frame imaging optimization according to step B is extended target Value function accumulates, then the desired value function after accumulation is carried out target trajectory backtracking, and rejects false track, specifically includes following Step:

C1, obtain continuous some frame imaging optimization according to step B after target image matrix be extended desired value function accumulation, An elementary state and respective value functional value in setting value Jacobian matrix, path matrix, target image matrix respectively, to this element State subsequent time speed carries out target expansion, is turned by introducing target spreading factor and Extended target shape mask matrix Move matrix of variables, utilize in transfering variable matrix the element index updated value Jacobian matrix corresponding to maximum and path matrix, Again to target image matrix after x direction, y direction, x direction speed, y direction speed four dimensions scan for being accumulated Value function matrix and path matrix；

In C2, obtaining step C1, the value function matrix median function value after accumulation is more than the element index of default detection threshold, leads to Cross the path matrix in step C1 and carry out flight path backtracking, obtain targetpath sequence cluster；

The targetpath sequence that in value function matrix after accumulating in C3, obtaining step C1, max function value is corresponding, by target The Euclidean distance meeting two elementary states in flight path sequence cluster is deleted more than other flight path sequences of allowable position error, repeats More than operation is until the flight path sequence number of targetpath bunch is 0, obtains pending flight path sequence cluster.

6. multiple target location estimation method after through-wall radar imaging based on improvement dynamic programming as claimed in claim 5, its It is characterised by, described step C1 obtains transfering variable square by introducing target spreading factor and Extended target shape mask matrix Battle array is embodied as:

$Temp=D_{i_2+1}(x_a+{\mathrm{\Δ}}_1+v_x^{\′}-q:x_a+{\mathrm{\Δ}}_1+v_x^{\′}+q,y_b+{\mathrm{\Δ}}_2+v_y^{\′}-q:y_b+{\mathrm{\Δ}}_2+v_y^{\′}+q)$

J(Δ₁,Δ₂)=Sum (Temp.*Mock)

Wherein, Temp matrix represents by target image matrixMiddle xth_a+Δ₁+v′_x-q row is to x_a+Δ₁+v′_x+ q row, Y_b+Δ₂+v′_y-q arranges y_b+Δ₂+v′_yThe matrix that size is (2q+1) * (2q+1) of+q column element composition, x_aRepresent target x The position in direction, y_bRepresent the position in target y direction, Δ₁,Δ₂Represent respectively x direction and y direction kinetic characteristic motor-driven because of Son, v '_x,v′_yFor target subsequent time x direction and y direction speed, q is target spreading factor, and Mock represents Extended target shape Mask matrix, all elements in matrix is added by Sum () expression one by one, J (Δ₁,Δ₂) represent transfering variable matrix.