CN104199022A - Target modal estimation based near-space hypersonic velocity target tracking method - Google Patents

Abstract

The invention provides a target modal estimation based near-space hypersonic velocity target tracking method, which comprises the steps of: tracking a target by utilizing an interactive multi-model tracking algorithm, estimating a target movement mode dynamically and in real time, judging the target movement mode by counting target characteristics, and finally turning to corresponding single-mode matched tracking based on the estimated target movement mode. Competition among multiple models is avoided, and the problems of low tracking precision and the like caused by complex calculation and great model competition of an existing near-space hypersonic velocity tracking algorithm are solved. The target modal estimation based near-space hypersonic velocity target tracking method is low in calculation amount and high in tracking precision, and effectively improves the integral performance of the tracking system.

Description

A kind of hypersonic method for tracking target of near space of based target mode estimation

Technical field

The invention belongs to Radar Signal Processing Technology, particularly the tracking technique to the hypersonic typical motion target of near space.

Background technology

Near space span is the high spatial domain of 20～100Km overhead, and near space hypersonic aircraft speed can reach 4～20 Mach.At present, the hypersonic target following of near space is a focus in tracking field.The complicacy of near space maneuvering target motion and the polytrope of running environment, cause and be difficult to this type of target to set up accurate trace model, and the development of stealth technology has more strengthened the difficulty to the hypersonic target following of near space.

At present, based on interactive multi-model (IMM) structure track algorithm, be a kind of track algorithm of generally acknowledging the hypersonic target of the most effective near space.In document " Research of Method for Tracking High Speed and Highly Maneuvering Target; International Conference on ITS Telecommunications Proceedings; 1236-1239; 2006 ", track algorithm based on interactive multimode structure is proposed, this algorithm utilizes the motor-driven model of multiple differences alternately the hypersonic target of near space to be followed the tracks of, follow the tracks of and there is wider coverage compared to single mode, have larger motor-driven adaptability; But between this algorithm model, vie each other and cause tracking accuracy poor, and calculated amount is larger.

Summary of the invention

Technical matters to be solved by this invention is, provides a kind of tracking accuracy higher, and the less hypersonic method for tracking target of near space of calculated amount.

The present invention is that the technical scheme that solves the problems of the technologies described above employing is, a kind of hypersonic method for tracking target of near space of based target mode estimation, comprises the following steps:

Step 1, that the metric data that radar is collected carries out targetpath is initial;

Step 2, utilize interactive multimode to follow the tracks of IMM algorithm to estimate uniform motion model, uniformly accelerated motion model and the corresponding next frame dbjective state of the motion model that turns round, state covariance matrix and corresponding model transition probability; Described dbjective state comprises target location, velocity and acceleration; When determining the current residing mode of motion of target after transition probability, target velocity and acceleration that under continuous statistics L frame, uniform motion model, uniformly accelerated motion model and the motion model that turns round are corresponding, enter step 3, otherwise return to step 2;

If in L frame time, at the uniform velocity model transition probability u _cvkeep maximum, velocity variable △ v always _k≤ τ _v, acceleration change amount △ a _k≤ τ _a, judge that target is the uniform motion stage;

If in L frame time, even acceleration model transition probability u _cakeep maximum, velocity variable △ v always _k>=τ _v, acceleration change amount △ a _k≤ τ _a, judge that target is the uniform motion stage;

If in L frame time, even acceleration model transition probability u _ctkeep maximum, velocity variable △ v always _k>=τ _v, acceleration △ a _k>=τ _a, judge that target is in the jumping stage;

Wherein, τ _vrepresent velocity perturbation thresholding, τ _arepresent acceleration fluctuation thresholding;

Step 3, single mode matching are followed the tracks of:

3-1 determines after target travel mode, proceeds to single mode matching and follows the tracks of; Initialization single mode parameter, target original state;

3-2 utilizes single mode matching track algorithm to estimate next frame dbjective state;

3-3 calculates the normalization residual error square ε of current time k _vfor: v _xfor current measurement residuals, S _xrepresent current residual error covariance matrix, subscript x represents the motion model matching with target real motion, x ∈ CV, CA, CT}, CV represents uniform motion, CA represents uniformly accelerated motion, CT represents jumping, () ^trepresenting matrix transposition; As normalization residual error square ε _vwhile being less than or equal to thresholding τ, think that target state is constant, return to step 3-2, as normalization residual error square ε _vwhile being greater than thresholding τ, think that target state changes, and turns back to step 2.

The present invention utilizes Interactive Multiple-Model track algorithm tracking target, and real-time dynamic estimation target travel mode, judges target travel mode by statistics target property; Finally, proceed to corresponding single mode matching follow the tracks of according to estimated target travel mode, avoided the competition between multi-model, solved the hypersonic track algorithm of existing near space and had calculation of complex, model competition causes greatly the problems such as tracking accuracy is low.

The invention has the beneficial effects as follows to there is calculated amount little high with tracking accuracy, effectively promoted the overall performance of tracker.

Brief description of the drawings

The track algorithm process flow diagram of Fig. 1 based on Interactive Multiple-Model and the estimation of target travel modal dynamic.

The new track algorithm of Fig. 2 and interactive multimode track algorithm tracking error curve figure.

The new track algorithm of Fig. 3 and interactive multimode track algorithm are followed the tracks of probability of success curve under different detection probabilities.

The new track algorithm of Fig. 4 and interactive multimode track algorithm are followed the tracks of probability of success curve under different sampling rates.

Embodiment

Adopt the method for Computer Simulation to verify effect of the present invention, by MATLAB-R2010b, implementation step as shown in Figure 1:

Step 1, radar data conversion

Tracked target enters ground radar scanning region, and it is metric data that radar collects data doppler information (Doppler) f with target _d, r represent target range, θ represent target azimuth, represent the target angle of pitch.Utilize correction to measure zero deflection and change method MUCMKF by polar coordinate system metric data be transformed under rectangular coordinate system, obtain the metric data after conversion wherein represent respectively to measure target location under rectangular coordinate system after conversion;

$x_m^u={\mathrm{\λ}}_{\mathrm{\θ}}{\mathrm{\λ}}_{\mathrm{\ϵ}}{\mathrm{\γ}}_m\cos {\mathrm{\θ}}_m\cos {\mathrm{\ϵ}}_m$

$y_m^u={\mathrm{\λ}}_{\mathrm{\θ}}{\mathrm{\λ}}_{\mathrm{\ϵ}}{\mathrm{\γ}}_m\sin {\mathrm{\θ}}_m\cos {\mathrm{\ϵ}}_m$

$z_m^u={\mathrm{\λ}}_{\mathrm{\ϵ}}{\mathrm{\γ}}_m\sin {\mathrm{\ϵ}}_m$

Wherein, compensating factor represent respectively the mean square deviation of position angle and luffing angle, θ _m, ε _mrepresent respectively position angle and the angle of pitch.

Step 2, targetpath are initial

2.1 store the metric data after 3 frame conversions, k moment data Z (k)={ z ₁(k), z ₂(k), z ₃(k) ... z _m1(k) }, the data set that obtains is Z (k+1)={ z the k+1 moment ₁(k+1), z ₂(k+1), z ₃(k+1) ... z _m2(k+1) }, the data set that obtains is Z (k+2)={ z the k+2 moment ₁(k+2), z ₂(k+2), z ₃(k+2) ... z _m3(k+2) }, associated continuous three frame data of traversal, utilize highly constrained h _min<h<h _maxwith constraint of velocity v _min<v<v _maxwith Doppler information f _dreject part clutter point mark.Wherein, z _i(k) represent i the measurement of k moment, m1, m2, m3 represents respectively the measurement number in 3 moment, h _min, h _maxrepresent respectively minimum and the highest flying height, v _min, v _maxfor minimum and maximum flying speed.

Continuous 3 frame data of remainder are projected to two planes (x-z and y-z plane) by 2.2, again two panel datas are transformed to parameter space, ρ=xcos θ+ysin θ, wherein, (θ, ρ) be the coordinate in parameter space, (x, y) is observation data under rectangular coordinate system; Recycling is revised Hough Hough conversion point counting is other two plane metric data are carried out to track initiation, contrast the initial flight path of associated two planes, and then definite targetpath is initial.In order to suppress as far as possible false track, preferably use here and revise Hough Hough conversion.To near space targetpath, initial meeting causes false track to increase to other track initiation method of this area, is unfavorable for target following, such as directly track initiation, correction logic method.

Step 3, utilize interactive multimode to follow the tracks of IMM algorithm keeps track target

After 3.1 targetpaths are initial, first initialization dbjective state $X={\left[\begin{array}{ccccccccc}x& \stackrel{\&CenterDot;}{x}& \stackrel{\&CenterDot;\&CenterDot;}{x}& y& \stackrel{\&CenterDot;}{y}& \stackrel{\&CenterDot;\&CenterDot;}{y}& z& \stackrel{\&CenterDot;}{z}& \stackrel{\&CenterDot;\&CenterDot;}{z}\end{array}\right]}^{\&prime;}.$ Each model transition probability u=[u _cvu _cau _ct] and each model state covariance matrix P _cv, P _ca, P _ct.The measuring value of supposing front 3 moment is respectively Z (1), Z (2), Z (3), and Z (i)=[x (i), y (i), z (i)] ^t, observe noise covariance matrix R;

Wherein, R directly tries to achieve by measuring transfer equation;

System state initialization:

$X=\left[\begin{array}{c}x\left(3\right)\\ (x\left(3\right)-x\left(2\right))/T\\ ((x\left(3\right)-x\left(2\right))/T-(x\left(2\right)-x\left(1\right))/T)/T\\ y\left(3\right)\\ (y\left(3\right)-y\left(2\right))/T\\ ((y\left(3\right)-y\left(2\right))/T-(y\left(2\right)-\left(1\right))/T)/T\\ z\left(3\right)\\ (z\left(3\right)-z\left(2\right))/T\\ ((z\left(3\right)-z\left(2\right))/T-(z\left(2\right)-z\left(1\right)))\end{array}\right]$

Wherein, ${\left[\begin{array}{ccccccccc}x& \stackrel{\&CenterDot;}{x}& \stackrel{\&CenterDot;\&CenterDot;}{x}& y& \stackrel{\&CenterDot;}{y}& \stackrel{\&CenterDot;\&CenterDot;}{y}& z& \stackrel{\&CenterDot;}{z}& \stackrel{\&CenterDot;\&CenterDot;}{z}\end{array}\right]}^{\&prime;}$ Represent respectively position, speed and acceleration in x-axis, y-axis and z-axis; [u _cv, u _ca, u _ct], [P _cv, P _ca, P _ct] be respectively at the uniform velocity model, even acceleration model and the corresponding model probability of turn model and state covariance matrix, and u _cv+ u _ca+ u _ct=1;

3.2 utilize Interactive Multiple-Model track algorithm, and prediction uniform motion CV model, uniformly accelerated motion CA model and the corresponding next frame dbjective state of motion CT model of turning round are respectively corresponding state covariance matrix is

3.3 utilize predicted value to estimate the corresponding dbjective state of each model dbjective state covariance matrix and upgrade the transition probability of each model

Each model state upgrades:

Each model state covariance is upgraded:

Estimate the transition probability of each model:

$C_j=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{p}_{\mathrm{ij}}{u}_i(k-1)$

${\mathrm{\&Lambda;}}^x=\frac{1}{\sqrt{2\mathrm{\&pi;}\left|S_x\right|}}{e}^{-\frac{1}{2}{v}_x^T{S}_x^{-1}{v}_x}$

$c=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&Lambda;}}_j{C}_j$

${\hat{u}}_x=\frac{1}{c}{\mathrm{\&Lambda;}}^x{C}_x$

Wherein, N represents interaction models number, u _i(k-1) transition probability of expression previous frame model i, p is known model transition probability matrix, K _i, v _iand S _irepresent respectively gain, the residual sum residual error covariance matrix of the Kalman filter that model i is corresponding, and subscript i ∈ { CV, CA, CT};

3.4 according to model transition probability $u=\left[\begin{array}{ccc}{\hat{u}}_{\mathrm{cv}}& {\hat{u}}_{\mathrm{ca}}& {\hat{u}}_{\mathrm{ct}}\end{array}\right]$ With each model estimating target state mutual output estimation dbjective state X.

$X={\hat{X}}_{\mathrm{cv}}\&CenterDot;{\hat{u}}_{\mathrm{cv}}+{\hat{X}}_{\mathrm{ca}}\&CenterDot;{\hat{u}}_{\mathrm{ca}}+{\hat{X}}_{\mathrm{ct}}\&CenterDot;{\hat{u}}_{\mathrm{ct}}$

Step 4, target travel mode estimation

In utilizing interactive multimode track algorithm tracking target, suppose that the model transition probability in k moment and dbjective state are respectively u (k)=[u _cv(k) u _ca(k) u _ct(k)] and ${\hat{X}}_k={\left[\begin{array}{ccccccccc}x& \stackrel{\&CenterDot;}{x}& \stackrel{\&CenterDot;\&CenterDot;}{x}& y& \stackrel{\&CenterDot;}{y}& \stackrel{\&CenterDot;\&CenterDot;}{y}& z& \stackrel{\&CenterDot;}{z}& \stackrel{\&CenterDot;\&CenterDot;}{z}\end{array}\right]}^{\&prime;},$ Speed and acceleration are respectively v _k, a _k; Add up continuous L frame model transition probability u, target velocity and acceleration information, the statistics to target travel characteristic in L frame time, judges the current residing mode of motion of target:

Target travel modal identification method:

If in L frame time, at the uniform velocity model probability u _cvkeep maximum, velocity variable △ v always _k≤ τ _v, acceleration change amount △ a _k≤ τ _a, judge that target is the uniform motion stage.

If in L frame time, even acceleration model probability u _cakeep maximum, velocity variable △ v always _k>=τ _v, acceleration change amount △ a _k≤ τ _a, judge that target is the uniform motion stage.

If in L frame time, even acceleration model probability u _ctkeep maximum, velocity variable △ v always _k>=τ _v, acceleration △ a _k>=τ _a, judge that target is in the jumping stage.

Wherein, τ _vrepresent velocity perturbation thresholding, τ _arepresent acceleration fluctuation thresholding.

Step 5, single mode matching track algorithm

5.1 determine after target travel mode, proceed to single mode matching and follow the tracks of; Initialization single mode parameter, target original state $X={\left[\begin{array}{ccccccccc}x& \stackrel{\&CenterDot;}{x}& \stackrel{\&CenterDot;\&CenterDot;}{x}& y& \stackrel{\&CenterDot;}{y}& \stackrel{\&CenterDot;\&CenterDot;}{y}& z& \stackrel{\&CenterDot;}{z}& \stackrel{\&CenterDot;\&CenterDot;}{z}\end{array}\right]}^{\&prime;}$ With goal displacement probability matrix P _o.

5.2 utilize single mode matching track algorithm target of prediction next frame state with corresponding error covariance matrix

Wherein, subscript x represents the motion model matching with target real motion, i.e. x ∈ { CV, CA, CT}.CV represents uniform motion; CA represents uniformly accelerated motion; The CT motion that represents to turn round.

5.3 utilize predicted value estimating target state and corresponding state covariance matrix:

Wherein, K _xfor Kalman filter gain, v _xfor measurement residuals, S _xrepresent residual error covariance matrix.

The residual sum residual error covariance of 5.4 pairs of each estimated states of moment of target is assessed, and k moment normalization residual error square is:

${\mathrm{\&epsiv;}}_v={v_x}^T\left(k\right)S_x^{-1}\left(k\right)v_x\left(k\right)$

Wherein, ε _vit is n that obedience has degree of freedom _zχ ²distribute, n _zrepresent to measure dimension.

Work as ε _vwhile exceeding thresholding τ, think that target state changes, and turns back to step 3.

Simulation result as shown in Figure 2, Figure 3, Figure 4, track algorithm of the present invention is compared with existing IMM algorithm keeps track, and the less and tracking accuracy of calculated amount is higher, follows the tracks of the probability of success higher under different detection probabilities, under different sampling rates, follow the tracks of the probability of success higher, make to follow the tracks of entirety and obtain performance boost.

Claims (2)

1. the hypersonic method for tracking target of the near space of based target mode estimation, is characterized in that, comprises the following steps:

Step 1, that the metric data that radar is collected carries out targetpath is initial;

Step 2, utilize interactive multimode to follow the tracks of IMM algorithm to estimate model transition probability and the dbjective state of uniform motion model, uniformly accelerated motion model and the corresponding next frame of motion model that turns round; Described dbjective state comprises target location, velocity and acceleration; When determining the current residing mode of motion of target after transition probability, target velocity and acceleration that under continuous statistics L frame, uniform motion model, uniformly accelerated motion model and the motion model that turns round are corresponding, enter step 3, otherwise return to step 2;

If in L frame time, at the uniform velocity model transition probability u _cvkeep maximum, velocity variable △ v always _k≤ τ _v, acceleration change amount △ a _k≤ τ _a, judge that target is the uniform motion stage;

If in L frame time, even acceleration model transition probability u _cakeep maximum, velocity variable △ v always _k>=τ _v, acceleration change amount △ a _k≤ τ _a, judge that target is the uniform motion stage;

If in L frame time, turn model transition probability u _ctkeep maximum, velocity variable △ v always _k>=τ _v, acceleration △ a _k>=τ _a, judge that target is in the jumping stage;

Wherein, τ _vrepresent velocity perturbation thresholding, τ _arepresent acceleration fluctuation thresholding;

Step 3, single mode matching are followed the tracks of:

3-1 determines after target travel mode, proceeds to single mode matching and follows the tracks of; Initialization single mode parameter, target original state;

3-2 utilizes single mode matching track algorithm prediction next frame dbjective state;

3-3 calculates the normalization residual error square ε of current time k _vfor: v _xfor current measurement residuals, S _xrepresent current residual error covariance matrix, subscript x represents the motion model matching with target real motion, x ∈ CV, CA, CT}, CV represents uniform motion, CA represents uniformly accelerated motion, CT represents jumping, () ^trepresenting matrix transposition; As normalization residual error square ε _vwhile being less than or equal to thresholding τ, think that target state is constant, return to step 3-2, as normalization residual error square ε _vwhile being greater than thresholding τ, think that target state changes, and turns back to step 2.

2. the hypersonic method for tracking target of a kind of near space of based target mode estimation as claimed in claim 1, it is characterized in that, utilize interactive multimode to follow the tracks of IMM algorithm and estimate that the concrete grammar of the dbjective state of uniform motion model, uniformly accelerated motion model and the corresponding next frame of motion model that turns round is: utilize interactive multimode to follow the tracks of IMM algorithm and obtain target prediction state value under next frame uniform motion model, uniformly accelerated motion model and the motion model that turns round and corresponding state covariance matrix then upgrade the dbjective state that each model is corresponding with corresponding covariance matrix estimate again each model transition probability matrix last export target estimated state X:

Single mode matching track algorithm estimates that the concrete grammar of next frame dbjective state is: utilize single mode matching track algorithm to obtain next frame target prediction state value and state covariance matrix upgrade again dbjective state and state covariance matrix k _x, v _xand S _xrepresent respectively gain, the residual sum residual error covariance matrix of the Kalman filter that model x is corresponding, and subscript x ∈ { CV, CA, CT}.