CN106249232A - Method for tracking target based on target travel situation information data association strategy - Google Patents

Abstract

The invention discloses method for tracking target based on target travel situation information data association strategy, including: step 1, create targetpath to be selected；Step 2, is tracked targetpath to be selected processing, and sets up rational official goals flight path according to target initial conditions；Step 3, utilizes Kalman filter model that tracking target and official goals flight path are filtered estimation and processes, obtain flight path state estimation.Calculate by ship collision prevention radar measurement model and follow the tracks of target and the least meeting distance point time TCPA of official goals, TCPA value is compared with target disengaging time thresholding, target time to approach thresholding and target thresholding overlapping time, obtains movement state information residing between target；Step 4: determine to follow the tracks of associating policy and the association algorithm of Targets Dots flight path according to movement state information residing between target, finally gives the incidence relation following the tracks of target with some mark.

Description

Method for tracking target based on target travel situation information data association strategy

Technical field

The invention belongs to radar target tracking technology, particularly relate to based on target travel situation information data association strategy Method for tracking target.

Background technology

Data association process determines that the measurement information and the process of target source corresponding relation that sensor receives, through Target is initial, target maintains and target terminating procedures.At present, data association algorithm conventional in engineering includes two big classes, first Class is to study up-to-date measurement set, and as Singer and Sea proposes optimum nearest neighbor algorithm, its ultimate principle selects and mesh Mark the nearest some mark of predicted position as relating dot.When target signal is higher or during target sparse, this algorithm keeps track effect Preferably.And low-resolution radar investigative range is wide, follow the tracks of destination number many.In addition radar resolution is low, azimuthal measurement low precision, mesh Mark easily presents intensive trend or target echo is overlapping, target overtakes, target crossover phenomenon.When target is in heavy dense targets district Or under target echo Overlay scenes, optimum nearest neighbor algorithm target following effect based on single positional information is poor.Equations of The Second Kind Data association algorithm is to study all measurement set before current time, the many hypothesis algorithm proposed such as Reid (MHT).In heavy dense targets region, many hypothesis algorithms utilize the metric data of follow-up multiple scan period, it is achieved stablizing of target Follow the tracks of.Donald B.Reid employing many hypothesis algorithm simulation simulates tracking effect (An during two target echo overlaps Algorithm for Tracking Multiple Target.IEEE Transctions on Automatic Control, Vol.AC-24, NO.6, December 1979.).Theoretically, many hypothesis algorithms use the posterior probability assuming flight path branch Solve compact district or the ambiguity of clutter district target data association.Masamichi Kojima proves that many hypothesis algorithms are one Plant algorithm (the A Study of Target Tracking that the Multiple Targets Data Association accuracy solved under complex environment is optimum Using Track-oriented Multiple Hypothesis Tracking.SICE, 1998.5:29-31.).But MHT calculates Method is produced assumes that flight path number of branches exponentially rises pass with false alarm rate, number of targets and handled multiple scan period numbers System, therefore the amount of calculation of algorithm and amount of storage are the biggest, be applied in real time in the radar target tracking engineering of sea still Acquire a certain degree of difficulty.

Summary of the invention

It is an object of the invention to utilize target travel situation information to identify target local environment, believe according to target environment Breath self adaptation uses different data association strategies and data association algorithm, improves Targets Dots track association accuracy, especially Be improve multiple target can meet, tracking ability in the complex scene such as intersection.

The technical solution realizing the object of the invention is: target based on target travel situation information data association strategy Tracking, step is as follows:

Comprise the steps:

Step 1, creates targetpath to be selected: the Searching point centered by following the tracks of target, in the r round regional extent as radius Mark information, search radius r is relevant with the radar antenna cycle.The radar antenna cycle is the longest, and search radius is the biggest.Generally may be used 1～7Km is arranged according to tracking position of object.If some mark belongs to this region, then this mark is utilized to create targetpath to be selected； The most do not create targetpath to be selected；

Step 2, was tracked targetpath to be selected processing: within the detections of radar cycle, if targetpath to be selected meets Targetpath initiates constraints, then transfer targetpath to be selected to official goals flight path, otherwise delete targetpath to be selected；

Step 3, utilizes Kalman filter model that tracking target and official goals flight path are filtered estimation and processes, obtain Flight path state estimation.When calculating, by ship collision prevention radar measurement model, the least meeting distance point following the tracks of target and official goals Between TCPA (time to closest point of approach), by close to TCPA value and target disengaging time thresholding, target Time threshold and target thresholding overlapping time compare, and obtain following the tracks of the motion shape that target and official goals are residing each other State information, such as separation, close or overlap condition；

Step 4: determine to follow the tracks of impact point according to the movement state information that tracking target and official goals are residing each other The associating policy of mark flight path and association algorithm, finally give the incidence relation following the tracks of target with some mark, thus realize target following.

Step 1 of the present invention is set up the targetpath step to be selected in the range of tracking target area as follows, it is assumed that follow the tracks of mesh The mark distance of A, orientation references are s_A, w_A.With (s_A,w_ACentered by), r is that radius determines regional extent.For falling into this region model Enclose interior some mark and set up targetpath to be selected.

Step 2 of the present invention comprises the steps:

Step 2-1, it is assumed that at t₁Moment creates target starting point to be selected, from t₂In the moment, use optimum nearest neighbor algorithm Target to be selected is carried out a mark track association:

$d^2\left(z\right)={\[z-\hat{z}(k+1|k)\]}^{\′}{S}^{-1}(k+1)\[z-\hat{z}(k+1|k)\],$

Wherein, z represents radar detection value, S^-1Covariance, d is newly ceased for target to be selected filtering²Z () is newly to cease weighted norm,Represent the estimation to k+1 moment target prediction state of the k moment；

If d²(z)≤γ set up, then judge target association to be selected to available point mark, otherwise, it is determined that Targets Dots to be selected is lost Losing, γ is distance threshold value.If target association to be selected is to available point mark, then record target association to be selected some mark information, mesh to be selected Mark relating dot mark information includes a mark radial distance g₁, some mark orientation g₂, some mark energy g₃, some mark orientation extension g₄, some mark distance Extension g₅With a mark time g₆, described target association point mark information to be selected is formed relating dot mark characteristic vector G, is shown below:

G=[g₁ g₂ g₃ g₄ g₅ g₆]；

Step 2-2, according to Kalman filter formulation group, filter state and predicted state to target to be selected are estimated:

$\hat{X}\left(k\right|k)=\hat{X}\left(k\right|k-1)+K\left(k\right)\[Z\left(k\right)-H\left(k\right)\hat{X}\left(k\right|k-1)\]$

$\hat{X}\left(k\right|k-1)=\mathrm{\Φ}\left(k\right|k-1)\hat{X}(k-1|k-1)+U\left(k\right)u\left(k\right)$

K (k)=P (k | k-1) H'(k) [H (k) P (k | k-1) H'(k)+R_k]^-1

P (k | k-1)=Φ (k | k-1) P (k-1 | k-1) Φ ' (k | k-1)+Q (k)

P (k | k)=[I-K (k) H (k)] P (k | k-1)

Wherein, Z (k) represents k moment sensor detection information, and Φ (k | k-1) represent that the k-1 moment turns to the state in k moment Moving matrix, H (k) represents measurement matrix, and U (k) represents input control item matrix, and u (k) represents known input or control signal, R K () represents zero-mean, the covariance of White Gaussian measurement noise, Q (k) is zero-mean, the covariance of White Gaussian process noise,Represent the state vector in target k moment,Represent the target prediction in the k-1 moment to the k moment, K (k) table Showing k moment filtering gain, P (k | k) represents the covariance matrix in k moment, and P (k | k-1) represent prediction covariance matrix, I be singly Bit matrix, H'(k) it is the transposed matrix of measurement matrix, Φ ' (k | k-1) is the transposed matrix of state-transition matrix.

Step 2-3, repeats step 2-1～step 2-2, until t_nIn the moment, N is the detections of radar cycle, it is assumed that target to be selected Total M relating dot mark within N number of detections of radar cycle, relating dot mark sequence table is shown as [G₁,G₂...,G_M]；

Step 2-4, calculates target proximity association point trace to be selected to range rate according to equation below:

$v_k=\frac{g_{i1}-g_{j1}}{g_{i6}-g_{j6}},$

$a_k=\frac{v_{k+1}-v_k}{t_{k+1}-t_k},$

Wherein, g_i1Represent i-th radar scanning cycle target association to be selected some mark first feature, i.e. some trace to away from From, g_j1Represent first feature of jth radar scanning cycle target association to be selected some mark, g_i6Represent i-th radar scanning week Expect to select the 6th feature of target association point mark, i.e. some mark time, g_j6Represent jth radar scanning cycle target association to be selected 6th feature of some mark, i=j+1, j=1,2 ..., M-1.Such as g₁₁Represent the of first scan period target association point mark 1 feature, g₂₁Represent the 1st feature of second scan period target association point mark.t_k+1-t_kRepresent v_k+1With v_kBetween time Between poor, v_k, v_k+1Expression k, k+1 moment target radial distance speed, k=1,2 ..., n-2, a_kRepresent radial distance rate of change；

Step 2-5, according to the rate of azimuth change of equation below calculating target proximity association point mark to be selected:

Azi_k=g_i2-g_j2,

Azi_k'=Azi_k+1-Azi_k,

Wherein, g_i2And g_j2Represent orientation values (i.e. second spy of i-th radar scanning target cycle relating dot mark respectively Levy) and the orientation values of jth radar scanning target cycle relating dot mark, Azi_kRepresent jth radar scanning target cycle point mark Bearing variation value, Azi_k' represent rate of azimuth change,

Step 2-6, is calculated n target range rate of change value and Orientation differences respectively according to step 2-4 and step 2-5 Rate value, is calculated as follows the root-mean-square of radial distance rate of change and the root-mean-square RMSE of rate of azimuth change:

$RMSE=\sqrt{\frac{\underset{i=1}{\overset{n}{\Σ}}{X}_i-\stackrel{\‾}{X}}{n}},$

Wherein, X_iRepresent target range rate of change or rate of azimuth change,Be target radial range rate average or The average of person's rate of azimuth change；

Step 2-7, calculates detection probability p of target association to be selected some mark:

$p=\frac{m}{n};$

Step 2-8, if target radial range rate root-mean-square thresholding is R_dG, target bearing rate of change root-mean-square thresholding is R_aG, target detection probability thresholding is P_d.If the root-mean-square RMSE of the radial distance rate of change of target proximity association point mark to be selected_d、 The root-mean-square RMSE of rate of azimuth change_aIt is satisfied by following condition with detection probability p:

RMSE_d≤R_dG,

RMSE_d≤R_aG,

p≤P_d,

Then targetpath to be selected is transferred to official goals flight path, and maintains target following according to Kalman filter model, with Time record object Track In Track during relating dot mark characteristic vector.

Step 3 of the present invention comprises the steps:

Step 3-1, according to the method described in step 2-1～step 2-7, utilize Kalman filter model to follow the tracks of target and Official goals flight path is filtered estimation and processes, and obtains flight path state distance, orientation, course and speed of a ship or plane estimated value.Utilize boats and ships Radar for collision avoidance measurement model, tracking target and the distance of official goals, orientation, course, the speed of a ship or plane, calculate and follow the tracks of target and formal mesh Target least meeting distance point time TCPA: set up plane coordinate system XOY, Y-axis is direct north, if following the tracks of target S₁It is positioned at seat Mark initial point O and with course d₁Speed of a ship or plane V₁Uniform motion, follows the tracks of the official goals S detected around target₂With course d₂Speed of a ship or plane V₂Even Speed motion, T₁Moment, official goals S₂Relatively follow the tracks of target S₁Position be A point, relative distance is R, and relative bearing is θ, AA' Being two target virtual courses, AA' is φ with the angle of Y-axis_r, the relative speed of a ship or plane is V_r, use equation below to calculate φ_rAnd V_r:

$V_r=\sqrt{V_2^2+V_1^1-2V_2{V}_{1}cos(d_1-d_2)},$

${\mathrm{\φ}}_r=arctan\frac{V_2\cos d_2{V}_1\cos d_1}{V_{2}sin{d}_2-V_1\sin d_1},$

After t, calculate official goals S₂Relatively follow the tracks of target S₁Displacement:

$\mathrm{\Δ}X=X_{t0}+\underset{0}{\∫}(V_2{\mathrm{sind}}_2-V_1\sin d_1)dt,$

$\mathrm{\Δ}Y=Y_{t0}+\underset{0}{\∫}(V_2{\mathrm{cosd}}_2-V_1\cos d_1)dt,$

Official goals S is calculated by Δ X and Δ Y₂Relatively follow the tracks of target S₁Distance R (t) after t and orientation θ (t):

$R\left(t\right)=\sqrt{{\mathrm{\ΔX}}^2+{\mathrm{\ΔY}}^2},$

$\mathrm{\θ}\left(t\right)=arctan\frac{\mathrm{\Δ}X}{\mathrm{\Δ}Y},$

Wherein, X_t0And Y_t0Represent t official goals S respectively₂Original position abscissa and t official goals S₂'s Original position vertical coordinate, Δ X and Δ Y represent target S respectively₂In t X-direction motion position and Y-direction motion position Put, in practical engineering application, by t radar to target S₂Detection, positional information R (t) and θ (t) can be obtained.Pass through Equation below calculates follows the tracks of target and the least meeting distance point time TCPA of official goals:

TCPA=R (t) | cos (φ_r-θ(t))|/V_r,

Step 3-2, if GT₁For target echo overlapping time, GT₂For target echo time to approach.If TCPA absolute value is less than GT₁, then judge to follow the tracks of target with official goals as meeting state；If TCPA absolute value is more than GT₁Less than GT₂, then judge to follow the tracks of mesh Mark and official goals are proximity state；If TCPA absolute value is more than GT₂, then judge to follow the tracks of target with official goals as separating shape State.

Step 4 of the present invention comprises the steps:

Step 4-1, if tracking target and target are in released state, then uses optimum nearest neighbor algorithm to carry out following the tracks of target Data association；

Step 4-2, if tracking target and official goals are in close or overlap condition, it is judged that follow the tracks of target and formal mesh Whether mark is associated with common point mark.If following the tracks of target and official goals being respectively associated different some marks, then by normal tracking Reason；If following the tracks of target and official goals relating dot mark being common point mark, then judge that following condition is set up the most simultaneously:

$abs(E_d-\mathrm{\α}\×({\stackrel{\‾}{E}}_A+{\stackrel{\‾}{E}}_B\left)\right)\≤G_E,$

$abs(l_d-\mathrm{\β}\×(\stackrel{-}{l_A}+\stackrel{-}{l_B}\left)\right)\≤G_l,$

Wherein,WithRepresent average energy value and the energy of official goals relating dot mark following the tracks of target association point mark respectively Amount average,WithRepresent respectively and follow the tracks of the echo bearing extension average of target association point mark and returning of official goals relating dot mark Ripple orientation extension average, E_dFor radar observable common point mark energy value, l_dFor radar observable common point mark orientation Expanding value.α, β are proportionality coefficient, and span is respectively 0.7≤α≤1 and 0.7≤β≤1.G_EFor energy error thresholding, G_lFor Orientation extension error threshold, if condition is set up simultaneously, then some mark is to follow the tracks of target and official goals echo overlap point mark, puts mark Being not assigned to follow the tracks of target and official goals, target carries out extrapolation process respectively；If condition is false, then some mark derive from Track target or official goals, be calculated as follows Δ_AAnd Δ_B:

${\mathrm{\Δ}}_A=abs(E_d-\stackrel{-}{E_A}),$

${\mathrm{\Δ}}_B=abs(E_d-\stackrel{-}{E_B}),$

Δ_ARepresent the energy of common point mark and follow the tracks of the similarity of target energy average, Δ_BRepresent the energy of common point mark With the similarity of official goals average energy value, if Δ_A≤Δ_B, some mark is for following the tracks of target association point mark, if Δ_A≥Δ_B, some mark is Official goals relating dot mark.

In step 4-1 of the present invention, when following the tracks of target and official goals is in released state, optimum nearest neighbor algorithm is used to enter Row data association.If following the tracks of target association point mark energy feature collection to be combined into { E₁,E₂,...,E_n, echo bearing extensive features sets For { l₁,l₂,...,l_n, n is the count value following the tracks of target association point mark.Official goals relating dot mark energy feature collection is combined into {E'₁,E₂',...,E_m', echo bearing extensive features sets is { l₁',l₂',...,l_m'}.M is official goals relating dot mark Count value.Point mark energy feature and orientation extension feature can use the some mark extractive technique of maturation to realize, and i.e. utilize target echo Seriality in orientation, distance carries out Plot coherence, obtains target bearing extension feature.Energy feature is by accumulating participation Energy corresponding to azran carry out summation and obtain.

The average energy value following the tracks of target association point mark is calculated by equation belowEnergy with official goals relating dot mark Average

$\stackrel{-}{E_A}=\underset{t=1}{\overset{n}{\Σ}}{E}_t,$

$\stackrel{-}{E_B}=\underset{t=1}{\overset{m}{\Σ}}{E_m}^{\′},$

The echo bearing extension average following the tracks of target association point mark is calculated by equation belowWith official goals relating dot The echo bearing extension average of mark

$\stackrel{-}{l_A}=\underset{t=1}{\overset{n}{\Σ}}{l}_t,$

$\stackrel{-}{l_B}=\underset{t=1}{\overset{m}{\Σ}}{l_t}^{\′}.$

Beneficial effect: the present invention compared with prior art, its remarkable advantage: during (1) target data association, not only examine Consider to the state estimation following the tracks of target itself, it is also contemplated that follow the tracks of the environmental information around target, establish tracking target and week Enclose the state of motion relation between other target, it is achieved that the self adaptation switching of multiple data association algorithm.(2) use between target Situation information judge target be in can meet, intersection etc. is complicated when following the tracks of scene, binding site mark diverse characteristics information assistance data Association algorithm, reduces and associates, based on single positional information, the associated errors rate caused.(3) compared with many hypothesis algorithms, based on The amount of calculation of data association strategy process of target travel situation information, amount of storage are less, are suitable for engineer applied.

Accompanying drawing explanation

Being the present invention with detailed description of the invention below in conjunction with the accompanying drawings and further illustrate, the present invention's is above-mentioned And/or otherwise advantage will become apparent.

Fig. 1 is for creating object delineation to be selected.

Fig. 2 is that target to be selected turns official goals process chart.

Fig. 3 is ship collision prevention radar measurement model.

Fig. 4 is that target travel situation judges process chart.

Fig. 5 is that TCPA state of motion judges explanatory diagram.

Fig. 6 is that multiple target overtakes tracking schematic diagram.

Fig. 7 is to follow the tracks of target situation change curve during multiple target overtakes tracking.

Fig. 8 is that multiple target intersection follows the tracks of schematic diagram.

Fig. 9 is to follow the tracks of target situation change curve during multiple target intersection is followed the tracks of.

Detailed description of the invention

In conjunction with Fig. 1, the first step, illustrates that targetpath to be selected creates process.Centered by following the tracks of target A, r is the circle of radius Searching point mark information in regional extent.Wherein, search radius r is relevant with the radar antenna cycle.The radar antenna cycle is the longest, searches Rope radius is the biggest.Generally can arrange 1～7Km according to tracking position of object.If some mark falls into this region, then create Targetpath to be selected；Otherwise, targetpath to be selected is not created.(in radar detection object procedure, be vulnerable to clutter, target etc. because of Element impact, can there are multiple somes marks in following the tracks of object wave door in synchronization.In Fig. 1, the meaning in two t1 moment is when identical In carving, in object wave door, occur in that two some marks, two some mark time consistencies simultaneously.In like manner, also there are two points in the t3 moment Mark.)

Second step, is tracked processing to the targetpath to be selected in the range of tracking target a-quadrant.In some detection cycles In, if goal satisfaction targetpath initial conditions to be selected, then target to be selected is transferred to official goals flight path B, otherwise deletes to be selected Targetpath, described in detail below.

(2-1): t₁Moment has created target starting point to be selected.From t₂In the moment, utilize optimum nearest neighbor algorithm formula pair Target to be selected carries out a mark track association.Wherein, z represents radar detection value, S^-1Covariance, d is newly ceased for target to be selected filtering² Z () is for newly ceasing weighted norm.

$d^2\left(z\right)={\[z-\hat{z}(k+1|k)\]}^{\′}{S}^{-1}(k+1)\[z-\hat{z}(k+1|k)\],$

If d²Z ()≤γ sets up, target association the most to be selected is to available point mark, and otherwise, Targets Dots to be selected is lost.γ be away from From threshold value.If target association to be selected is to some mark, then record target association to be selected some mark information.Target association point mark information to be selected Mainly it is made up of six parts: some mark radial distance g₁, some mark orientation g₂, some mark energy g₃, some mark orientation extension g₄, some mark distance Extension g₅, put mark time g₆, form relating dot mark characteristic vector G, as shown in formula.

G=[g₁ g₂ g₃ g₄ g₅ g₆]

(2-2): after target data association to be selected is disposed, according to Kalman filter formulation group, to target to be selected Filter state, predicted state are estimated.

$\hat{X}\left(k\right|k)=\hat{X}\left(k\right|k-1)+K\left(k\right)\[Z\left(k\right)-H\left(k\right)\hat{X}\left(k\right|k-1)\]$

$\hat{X}\left(k\right|k-1)=\mathrm{\Φ}\left(k\right|k-1)\hat{X}(k-1|k-1)+U\left(k\right)u\left(k\right)$

K (k)=P (k | k-1) H'(k) [H (k) P (k | k-1) H'(k)+R_k]^-1

P (k | k-1)=Φ (k | k-1) P (k-1 | k-1) Φ ' (k | k-1)+Q (k)

P (k | k)=[I-K (k) H (k)] P (k | k-1)

(2-3): repeat (2-1)～(2-2) step until t_NIn the moment, N is the detection cycle.In view of radar false alarm probability and Ensureing target effective relating dot mark sample number to be selected, N typically takes 5～7.

(2-4): combine the process that Fig. 2 illustrates that object judgement to be selected is official goals.Assume that target to be selected is visited at N number of radar In the survey cycle, relating dot mark sequence is represented by [G₁,G₂...,G_M], M is target association to be selected some mark number, G_iFor a mark feature Vector.

(2-4-1): calculate target consecutive points trace to be selected to range rate by formula.

$v_k=\frac{g_{i1}-g_{j1}}{g_{i6}-g_{j6}}$

$a_k=\frac{v_{k+1}-v_k}{t_{k+1}-t_k}$

G in formula_ijThe jth feature of expression i-th radar scanning target cycle relating dot mark, i=j+1, j=1,2 ..., m-1。v_k, v_k+1Expression k, k+1 moment target radial distance speed, k=1,2 ..., n-2, t_k+1-t_kFor v_k+1With v_kBetween time Between poor, a_kFor target radial range rate value.

(2-4-2): calculate the rate of azimuth change of target consecutive points mark to be selected by formula.

Azi_k=g_i2-g_j2, i=j+1

Azi_k'=Azi_k+1-Azi_k

g_ijRepresent the jth feature of i-th radar scanning target cycle relating dot mark, Azi_k' for rate of azimuth change.

(2-4-3): calculate target radial range rate to be selected and the root-mean-square value of rate of azimuth change.

$RMSE=\sqrt{\frac{\underset{i=1}{\overset{n}{\Σ}}{X}_i-\stackrel{\‾}{X}}{n}}$

X in formula_iRepresent target radial range rate value or rate of azimuth change value,It it is target radial range rate Average or the average of rate of azimuth change,

(2-4-4): calculate the detection probability of Targets Dots to be selected.

$p=\frac{M}{N}$

If target radial range rate root-mean-square to be selected, rate of azimuth change root-mean-square and some mark detection probability be satisfied by with Lower targetpath initiates constraints:

RMSE_d≤R_dG

RMSE_d≤R_aG

p≤P_d

Target the most to be selected transfers official goals flight path to.And maintain target following, simultaneously record according to Kalman filter model Relating dot mark characteristic vector in object tracking process.

3rd step: combine Fig. 3 explanation, utilizes ship collision prevention radar measurement model to calculate and follows the tracks of target A and the minimum of target B Meeting distance point time (TCPA) method.In plane coordinate system XOY, Y-axis is direct north.If following the tracks of target S₁It is positioned at coordinate former Point, with course d₁Speed of a ship or plane V₁Uniform motion.Follow the tracks of target S detected around target₂With course d₂Speed of a ship or plane V₂Uniform motion.T₁Time Carve, S₂S relatively₁Position be A point, relative distance is R, relative bearing θ.AA' is two target virtual courses, with the angle of Y-axis is φ_r, the relative speed of a ship or plane is V_r.φ can be calculated by geometrical relationship_rAnd V_r,

$V_r=\sqrt{V_2^2+V_1^1-2V_2{V}_{1}cos(d_1-d_2)}$

${\mathrm{\φ}}_r=arctan\frac{V_2\cos d_2{V}_1\cos d_1}{V_{2}sin{d}_2-V_1\sin d_1}$

After t, calculate target S₂Relative target S₁Displacement,

$\mathrm{\Δ}X=X_{t0}+\underset{0}{\∫}(V_2{\mathrm{sind}}_2-V_1\sin d_1)dt,$

$\mathrm{\Δ}Y=Y_{t0}+\underset{0}{\∫}(V_2{\mathrm{cosd}}_2-V_1\cos d_1)dt$

S is derived by Δ X, Δ Y₂S relatively₁Distance and bearing,

$R\left(t\right)=\sqrt{{\mathrm{\ΔX}}^2+{\mathrm{\ΔY}}^2}$

$\mathrm{\θ}\left(t\right)=arctan\frac{\mathrm{\Δ}X}{\mathrm{\Δ}Y}$

In formula, X_t0And Y_t0For t official goals S₂Original position, Δ X and Δ Y represents target S respectively₂In t Interior X-direction and the position of Y-direction motion.In practical engineering application, can be by t radar to target S₂Detection obtain position Confidence breath R (t) and θ (t).By relative distance and orientation, calculate TCPA by formula,

TCPA=R (t) | cos (φ_r-θ(t))|/V_r

Illustrate to judge tracking target A and state of motion flow chart between target B about according to TCPA value in conjunction with Fig. 4. GT₁For target echo overlapping time, GT₂For target echo time to approach.Target situation judge process is illustrated in conjunction with Fig. 5.If GT₁=20, GT₂=40.If TCPA absolute value is less than GT₁, then follow the tracks of target A and be judged as meeting state with target B；If TCPA is exhausted To value more than GT₁Less than GT₂, then follow the tracks of target A and be judged as proximity state with target B；If TCPA absolute value is more than GT₂, then follow the tracks of Target A and target B are judged as released state.Ordinary circumstance, arranges GT according to radar period₁And GT₂.The present invention is at engineer applied Middle GT₁It is set to 1 minute, GT₂It is set to 5 minutes.Illustrate that multiple target follows the tracks of target situation during overtaking tracking in conjunction with Fig. 6 bent The change of line chart, Fig. 7 is the situation curve variation diagram overtaking target 0009 batch.When following the tracks of the TCPA of target 0009 batch more than GT₁ Time, following the tracks of target 0009 situation curve is released state.Within radar the 10th to 35 scan period, follow the tracks of target 0009 batch with just Formula target echo does not separates, and follows the tracks of target TCPA value less than GT for 0009 batch₁Time, situation curve holding can meet state.When radar exists During the 36th scan period, follow the tracks of target and complete to overtake process.0009 batch of tracking target and official goals situation curve are the most more New is released state.Illustrating that multiple target intersection follows the tracks of the change of target situation curve figure during following the tracks of in conjunction with Fig. 8, Fig. 9 is for handing over The situation curve variation diagram of fork target 0015 batch.When following the tracks of the TCPA of target 0015 batch more than GT₁Time, follow the tracks of 0015 batch of state of target Power curve is released state.When follow the tracks of target 0015 batch and 0018 batch of echo of official goals close to time, TCPA value result of calculation Jie Enter GT₁And GT₂Between, follow the tracks of 0015 batch of situation curve of target and be displayed in proximity to state.In radar the 57th to 77 scan period, with Track target 0015 and 0018 batch of echo overlap of official goals, TCPA value result of calculation is less than GT₁, follow the tracks of 0015 batch of situation song of target Line is for can meet state.Along with radar scanning and target travel, follow the tracks of target and official goals echo is gradually disengaged, follow the tracks of target The situation curve of 0015 shows by meeting, and becomes close, eventually becomes released state.

4th step, determines to follow the tracks of associating policy and the association algorithm of target A point mark flight path according to target travel situation information. If following the tracks of target A to be in released state with target B about, then target A is followed the tracks of in explanation and target B does not haves and can meet or hand over Fork scene.Follow the tracks of target A and carry out data association by optimum nearest neighbor algorithm.If follow the tracks of target A and target B about be in close to or During person's overlap condition, in fact it could happen that two kinds of some mark association situations.The first, the some mark that each auto correlation of target is different；The second, mesh The point mark that mark association is identical.For the first situation, it is not necessary to consider some mark feature.The second situation need to consider a mark feature.False If following the tracks of target A point mark energy feature collection to be combined into { E₁,E₂,...,E_n, orientation extensive features sets is { l₁,l₂,...,l_n, n For following the tracks of the count value of target association point mark.Target B point mark energy feature collection is combined into { E'₁,E₂',...,E_m', orientation extension spy Collection is combined into { l₁',l₂',...,l_m', m is the count value following the tracks of target association point mark.

(4-1): calculate target A, the average energy value of target B relating dot mark respectively by formula,

${\stackrel{\‾}{E}}_A=\underset{t=1}{\overset{n}{\Σ}}{E}_t$

${\stackrel{\‾}{E}}_B=\underset{t=1}{\overset{m}{\Σ}}{E}_t$

(4-2): calculate the echo bearing extension average of target A, target B relating dot mark respectively by formula.

${\stackrel{\‾}{l}}_A=\underset{t=1}{\overset{n}{\Σ}}{l}_t$

${\stackrel{\‾}{l}}_B=\underset{t=1}{\overset{m}{\Σ}}{l}_t$

(4-3): by common point mark ENERGY E_d, orientation extension l_dRespectively with tracking target A, target B average energy value Orientation extension averageRow compares.If meeting following condition, illustrate that common point mark is two target echo overlap point marks, then public Concurrent mark is not assigned to target A, target B, two target extrapolation process.Otherwise, process by the 4th step.G in formula_EFor energy threshold, G_l For energy threshold.Ordinary circumstance, target energy and orientation extension according to radar detection are configured G_EAnd G_l。

$abs(E_d-\mathrm{\α}\×({\stackrel{\‾}{E}}_A+{\stackrel{\‾}{E}}_B\left)\right)\≤G_E$

$abs(l_d-\mathrm{\β}\×(\stackrel{-}{l_A}+\stackrel{-}{l_B}\left)\right)\≤G_l$

(4-4): calculate following equalities.

${\mathrm{\Δ}}_A=abs(E_d-\stackrel{-}{E_A})$

${\mathrm{\Δ}}_B=abs(E_d-\stackrel{-}{E_B})$

If Δ_A≤Δ_B, some mark is for following the tracks of target A relating dot mark.If Δ_A≥Δ_B, some mark is target B relating dot mark.Polynary Characteristic correlating method is disposed.

Although the present invention is illustrated with regard to preferred implementation and has been described, it is understood by those skilled in the art that Without departing from scope defined by the claims of the present invention, the present invention can be carried out variations and modifications.

Claims (5)

1. method for tracking target based on target travel situation information data association strategy, it is characterised in that comprise the steps:

Step 1, creates targetpath to be selected: the Searching point mark centered by following the tracks of target, in the r round regional extent as radius is believed Breath, if some mark belongs to this region, then utilizes this mark to create targetpath to be selected；The most do not create targetpath to be selected；

Step 2, is tracked targetpath to be selected processing: within the detections of radar cycle, if targetpath to be selected meets target Track initiation constraints, then transfer targetpath to be selected to official goals flight path, otherwise delete targetpath to be selected；

Step 3, utilizes Kalman filter model that tracking target and official goals flight path are filtered estimation and processes, keep away by boats and ships Touch radar measurement model and calculate tracking target and the least meeting distance point time TCPA of official goals, TCPA value is divided with target Compare from time threshold, target time to approach thresholding and target thresholding overlapping time, obtain following the tracks of target and official goals Movement state information residing each other；

Step 4: determine to follow the tracks of Targets Dots boat according to the movement state information that tracking target and official goals are residing each other The associating policy of mark and association algorithm, finally give the incidence relation following the tracks of target with some mark, thus realize target following.

Method the most according to claim 1, it is characterised in that step 2 comprises the steps:

Step 2-1, it is assumed that at t₁Moment creates target starting point to be selected, from t₂In the moment, optimum nearest neighbor algorithm is used to treat Target is selected to carry out a mark track association:

$d^2\left(z\right)={\[z-\hat{z}(k+1|k)\]}^{\′}{S}^{-1}(k+1)\[z-\hat{z}(k+1|k)\],$

Wherein, z represents radar detection value, S^-1Covariance, d is newly ceased for target to be selected filtering²Z () is newly to cease weighted norm,Represent the estimation to k+1 moment target prediction state of the k moment；

If d²(z)≤γ set up, then judge target association to be selected to available point mark, otherwise, it is determined that Targets Dots to be selected loss, γ For distance threshold value, if target association to be selected is to available point mark, then record target association to be selected some mark information, target association to be selected Point mark information includes a mark radial distance g₁, some mark orientation g₂, some mark energy g₃, some mark orientation extension g₄, some mark extended distance g₅ With a mark time g₆, described target association point mark information to be selected is formed relating dot mark characteristic vector G, is shown below:

G=[g₁ g₂ g₃ g₄ g₅ g₆]；

Step 2-2, according to Kalman filter formulation, filter state and predicted state to target to be selected are estimated:

$\hat{X}\left(k\right|k)=\hat{X}\left(k\right|k-1)+K\left(k\right)\[Z\left(k\right)-H\left(k\right)\hat{X}\left(k\right|k-1)\]$

$\hat{X}\left(k\right|k-1)=\mathrm{\Φ}\left(k\right|k-1)\hat{X}(k-1|k-1)+U\left(k\right)u\left(k\right)$

K (k)=P (k | k-1) H'(k) [H (k) P (k | k-1) H'(k)+R (k)]^-1

P (k | k-1)=Φ (k | k-1) P (k-1 | k-1) Φ ' (k | k-1)+Q (k)

P (k | k)=[I-K (k) H (k)] P (k | k-1)

Wherein, the co-ordinate position information of Z (k) expression k moment sensor detection target, Φ (k | k-1) represent when the k-1 moment is to k Carving target state transfer matrix, H (k) represents target measurement matrix, and U (k) represents input control item matrix, and u (k) has represented Knowing input or control signal, R (k) represents zero-mean, the covariance of White Gaussian measurement noise, and Q (k) is zero-mean, white The covariance of Gaussian process noise,Represent the state estimation vector in target k moment,Represent the k-1 moment pair The dbjective state prediction in k moment, K (k) represents k moment filtering gain, the covariance matrix in P (k | k) expression k moment, P (k | k- 1) representing prediction covariance matrix, I is unit matrix, H'(k) it is the transposed matrix of target measurement matrix, Φ ' (k | k-1) it is shape The transposed matrix of state transfer matrix；

Step 2-3, repeats step 2-1～step 2-2, until t_NIn the moment, N is the detections of radar cycle, it is assumed that target to be selected is N number of Total M relating dot mark in the detections of radar cycle, relating dot mark sequence table is shown as [G₁,G₂...G_M]；

Step 2-4, calculates target proximity association point trace to be selected to range rate according to equation below:

$v_k=\frac{g_{i1}-g_{j1}}{g_{i6}-g_{j6}},$

$a_k=\frac{v_{k+1}-v_k}{t_{k+1}-t_k},$

Wherein, g_i1Represent first feature of i-th radar scanning cycle target association to be selected some mark, i.e. some mark radial distance, g_j1Represent first feature of jth radar scanning cycle target association to be selected some mark, g_i6Represent the i-th radar scanning cycle 6th feature of target association point mark to be selected, i.e. some mark time, g_j6Represent jth radar scanning cycle target association to be selected point 6th feature of mark, i=j+1, j=1,2 ..., M-1, t_k+1-t_kRepresent v_k+1With v_kBetween time difference, v_kRepresent the k moment Target radial distance speed, k=1,2 ..., n-2, a_kRepresent target radial range rate；

Step 2-5, according to the rate of azimuth change of equation below calculating target proximity association point mark to be selected:

Azi_k=g_i2-g_j2,

Azi_k'=Azi_k+1-Azi_k,

Wherein, g_i2And g_j2Represent orientation values and the jth radar scanning week of i-th radar scanning target cycle relating dot mark respectively The orientation values of phase target association point mark, Azi_kRepresent jth radar scanning target cycle point mark bearing variation value, Azi_k' expression side Position rate of change.

Step 2-6, is calculated n target range rate of change value and rate of azimuth change respectively according to step 2-4 and step 2-5 Value, is calculated as follows the root-mean-square of radial distance rate of change and the root-mean-square RMSE of rate of azimuth change:

$RMSE=\sqrt{\frac{\underset{i=1}{\overset{n}{\Σ}}{X}_i-\stackrel{\‾}{X}}{n}},$

Wherein, X_iRepresent target range rate of change or rate of azimuth change,It is target radial range rate average or orientation The average of rate of change；

Step 2-7, calculates detection probability p of target association to be selected some mark:

$p=\frac{M}{N};$

Step 2-8, if target radial range rate root-mean-square thresholding is R_dG, target bearing rate of change root-mean-square thresholding is R_aG, Target detection probability thresholding is P_dIf, the root-mean-square RMSE of the radial distance rate of change of target proximity association point mark to be selected_d, orientation The root-mean-square RMSE of rate of change_aIt is satisfied by following condition with detection probability p:

RMSE_d≤R_dG,

RMSE_d≤R_aG,

p≤P_d,

Then targetpath to be selected is transferred to official goals flight path, and maintains target following according to Kalman filter model, remember simultaneously Relating dot mark characteristic vector during record targetpath tracking.

Method the most according to claim 2, it is characterised in that step 3 comprises the steps:

Step 3-1, when utilizing ship collision prevention radar measurement model to calculate the least meeting distance point following the tracks of target and official goals Between TCPA: setting up plane coordinate system XOY, Y-axis is direct north, if follow the tracks of target S₁It is positioned at zero O and with course d₁The speed of a ship or plane V₁Uniform motion, follows the tracks of the official goals S detected around target₂With course d₂Speed of a ship or plane V₂Uniform motion, T₁Moment, official goals S₂Relatively follow the tracks of target S₁Position be A point, relative distance is R, relative bearing be θ, AA' be two target virtual courses, AA' and Y The angle of axle is φ_r, the relative speed of a ship or plane is V_r, use equation below to calculate φ_rAnd V_r:

$V_r=\sqrt{V_2^2+V_1^1-2V_2{V}_{1}cos(d_1-d_2)},$

${\mathrm{\φ}}_r=\arctan \frac{V_2{\mathrm{cosd}}_2{V}_1{\mathrm{cosd}}_1}{V_2{\mathrm{sind}}_2-{\mathrm{Vsind}}_1},$

After t, calculate official goals S₂Relatively follow the tracks of target S₁Displacement:

$\mathrm{\Δ}X=X_{t0}+\underset{0}{\∫}(V_2{\mathrm{sind}}_2-V_1{\mathrm{sind}}_1)dt,$

$\mathrm{\Δ}Y=Y_{t0}+\underset{0}{\∫}(V_2{\mathrm{cosd}}_2-V_1{\mathrm{cosd}}_1)dt,$

Official goals S is calculated by Δ X and Δ Y₂Relatively follow the tracks of target S₁Distance R (t) after t and orientation θ (t):

$R\left(t\right)=\sqrt{{\mathrm{\ΔX}}^2+{\mathrm{\ΔY}}^2},$

$\mathrm{\θ}\left(t\right)=\arctan \frac{\mathrm{\Δ}X}{\mathrm{\Δ}Y},$

Wherein, X_t0And Y_t0Represent t official goals S respectively₂Original position abscissa and t official goals S₂Initial Position vertical coordinate, Δ X and Δ Y represent target S respectively₂In t, the position of X-direction motion and the position of Y-direction motion, logical Cross equation below to calculate and follow the tracks of target and the least meeting distance point time TCPA of official goals:

TCPA=R (t) | cos (φ_r-θ(t))|/V_r；

Step 3-2, if GT₁For target echo overlapping time, GT₂For target echo time to approach.If TCPA absolute value is less than GT₁, Then judge to follow the tracks of target with official goals as meeting state；If TCPA absolute value is more than GT₁Less than GT₂, then judge follow the tracks of target with Official goals is proximity state；If TCPA absolute value is more than GT₂, then judge to follow the tracks of target and official goals as released state.

Method the most according to claim 3, it is characterised in that step 4 comprises the steps:

Step 4-1, if tracking target and target are in released state, then uses optimum nearest neighbor algorithm to carry out data to following the tracks of target Association；

Step 4-2, if tracking target and official goals are in close or overlap condition, then needs to judge to follow the tracks of target and formal mesh Whether mark is associated with common point mark.If following the tracks of target and official goals being respectively associated different some marks, then by normal tracking Reason；If following the tracks of target and official goals relating dot mark being common point mark, then judge that following condition is set up the most simultaneously:

$abs(E_d-\mathrm{\α}\×(\stackrel{-}{E_A}+\stackrel{-}{E_B}\left)\right)\≤G_E,$

$abs(l_d-\mathrm{\β}\×(\stackrel{-}{l_A}+\stackrel{-}{l_B}\left)\right)\≤G_l$

Wherein,WithThe average energy value of expression tracking target association point mark and the energy of official goals relating dot mark are equal respectively Value,WithRepresent echo bearing extension average and the echo side of official goals relating dot mark following the tracks of target association point mark respectively Bits Expanding average, E_dFor radar observable common point mark energy value, l_dExtend for radar observable common point mark orientation Value, α and β is proportionality coefficient, G_EFor energy error thresholding, G_lError threshold is extended for orientation.If condition is set up simultaneously, then point Mark is to follow the tracks of target and official goals echo overlap point mark, and some mark is not assigned to follow the tracks of target and official goals, and target is entered respectively Row extrapolation process；If condition is false, then some mark derives from tracking target or official goals, is calculated as follows Δ_A And Δ_B:

${\mathrm{\Δ}}_A=abs(E_d-\stackrel{-}{E_A}),$

${\mathrm{\Δ}}_B=abs(E_d-\stackrel{-}{E_B}),$

Δ_ARepresent the energy of common point mark and follow the tracks of the similarity of target energy average, Δ_BRepresent common point mark energy with just The similarity of formula target energy average, if Δ_A≤Δ_B, some mark is for following the tracks of target association point mark, if Δ_A＞ Δ_B, some mark is formal Target association point mark.

Method the most according to claim 4, it is characterised in that in step 4-1, is in point when following the tracks of target and official goals When state, optimum nearest neighbor algorithm is used to carry out data association, if following the tracks of target association point mark energy feature collection to be combined into { E₁, E₂,...,E_n, echo bearing extensive features sets is { l₁,l₂,...,l_n, n is the count value following the tracks of target association point mark, just Formula target association point mark energy feature collection is combined into { E'₁,E₂',...,E_m', echo bearing extensive features sets is { l₁', l₂',...,l_m', m is the count value of official goals relating dot mark, utilizes target echo seriality in orientation, distance to carry out Plot coherence, obtains target bearing extension feature, and energy feature is by carrying out energy corresponding for the azran participating in accumulation Summation obtains,

The average energy value following the tracks of target association point mark is calculated by equation belowAverage energy value with official goals relating dot mark

$\stackrel{-}{E_A}=\underset{t=1}{\overset{n}{\Σ}}{E}_t,$

$\stackrel{-}{E_B}=\underset{t=1}{\overset{m}{\Σ}}{E_t}^{\′},$

The echo bearing extension average following the tracks of target association point mark is calculated by equation belowWith returning of official goals relating dot mark Ripple orientation extension average

$\stackrel{-}{l_A}=\underset{t=1}{\overset{n}{\Σ}}{l}_t,$

$\stackrel{-}{l_B}=\underset{t=1}{\overset{m}{\Σ}}{l_t}^{\′}.$