Maneuvering target cluster state estimation method and device based on direction angle iteration identification

Abstract

The invention discloses a maneuvering target cluster state estimation method and device based on direction angle iteration identification. The method comprises the following steps: calculating a state one-step prediction quantity of a target cluster at a k moment according to a state estimation quantity of the target cluster at the k-1 moment; iteratively updating the direction angle of the target cluster based on the state one-step pre-measurement of the target cluster at the k moment; determining a state estimator of the target cluster at the moment k according to the final direction angle after iteration updating; according to the method, the change of the direction angle of the target cluster can be obtained in real time by estimating the direction angle in real time, so that when the target cluster maneuvers and rapidly changes along with the direction angle of the cluster, stable tracking of the target cluster can be realized, and the state estimation precision when the target cluster maneuvers is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of target tracking, and in particular relates to a method and device for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles. Background Art

[0002] With the continuous advancement of electronic warfare technology and drone technology, single drones are increasingly powerless in complex combat missions due to their own load limitations. With the help of the agile maneuverability and high efficiency of drones, combat forms represented by swarm drones have entered the public's field of vision. Their high density makes up for the shortcomings of single drone combat and has gradually become the mainstream combat method. However, due to the limitation of radar resolution, this combat method has brought considerable challenges to the tracking system, which has also prompted a series of extensive research on cluster targets such as swarm drones.

[0003] There are two main types of cluster target tracking: one is to expand the state of all targets in the cluster and then estimate the state of each target; the other is to treat all targets in the cluster as a whole and estimate the state of the cluster targets by estimating their center of mass state and contour state. However, most existing methods model the orientation of cluster targets as a uniform motion model, which will inevitably lead to the failure of tracking methods when the orientation of cluster targets changes rapidly. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles, so as to improve the state estimation accuracy of the target cluster during maneuvering.

[0005] The present invention adopts the following technical solution: a method for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles, comprising the following steps:

[0006] Calculate the one-step prediction of the state of the target cluster at time k based on the state estimate of the target cluster at time k-1;

[0007] Iteratively update the direction angle of the target cluster based on the one-step prediction of the state of the target cluster at time k;

[0008] The state estimation of the target cluster at time k is determined according to the final direction angle after iterative update.

[0009] Furthermore, iteratively updating the direction angle of the target cluster based on the one-step prediction of the state of the target cluster at time k includes:

[0010] Calculate the measurement error Δz based on the measurement vector and nonlinear measurement matrix of the target cluster at time k k ; Among them, the nonlinear measurement matrix is ​​calculated based on the state one-step prediction quantity and the direction angle;

[0011] Generate angular increments based on measurement errors

[0012] The bearing angle and bearing increment Merge into new directions;

[0013] When the difference between the new direction angle and the direction angle of the previous iterative update is less than or equal to the threshold or the number of iterative updates reaches the maximum number, the new direction angle is used as the final direction angle.

[0014] Furthermore, the method for generating the nonlinear measurement matrix includes:

[0015] Calculate the radial distance measurement matrix at time k based on the direction angle

[0016] According to the radial distance measurement matrix The one-step prediction of the state generates the nonlinear measurement matrix at time k.

[0017] Furthermore, the direction angle increment is generated according to the measurement error include:

[0018]

[0019] in, represents the direction angle increment of the t-th iteration at time k, represents the derivative of the nonlinear measurement matrix of the t-th iteration at time k with respect to the direction angle, represents the measurement error covariance matrix of the t-th iteration at time k, represents the measurement error of the tth iteration at time k.

[0020] Further, the state estimation includes state estimation under different sub-modes; wherein the sub-modes are different motion state models in the state estimation process;

[0021] Each submodule has its corresponding posterior probability.

[0022] Furthermore, determining the state estimation of the target cluster at time k according to the final direction angle after iterative update includes:

[0023] Estimate the estimated direction angle of the target cluster according to the final direction angle in each sub-mode;

[0024] A state estimate is determined based on the estimated direction angle.

[0025] Further, estimating the estimated direction angle of the target cluster according to the final direction angle in each sub-mode includes:

[0026] Calculate the posterior probability of each sub-mode according to the final direction angle and state one-step prediction quantity in each sub-mode;

[0027] The estimated direction angle is generated according to the final direction angle and the corresponding posterior probability in each sub-mode.

[0028] Furthermore, calculating the posterior probability of each sub-pattern includes:

[0029]

[0030] Among them, μ k|k,j represents the posterior probability corresponding to the jth sub-pattern at time k, Λ k,j represents the likelihood function value of the jth sub-pattern at time k, π ij represents the prior probability of transitioning from sub-mode i to sub-mode j, μ k-1|k-1,i represents the posterior probability corresponding to the i-th sub-pattern at time k-1, and N represents the number of sub-patterns.

[0031] Further, determining the state estimation amount according to the estimated direction angle includes:

[0032] Calculate the estimated state quantity in each sub-mode according to the estimated direction angle and update the posterior probability of each sub-mode;

[0033] Decompose the estimated state quantity under each sub-mode to generate the estimated state quantity of the centroid of the target cluster and the estimated state quantity of the radial distance set under the sub-mode;

[0034] The estimated state quantities of the target cluster centroid and the estimated state quantities of the radial distance set under different sub-modes are interacted respectively to obtain new estimated state quantities of the target cluster centroid and the estimated state quantities of the radial distance set;

[0035] The estimated state quantity of the new target cluster centroid and the estimated state quantity of the radial distance set are combined to obtain the estimated state quantity.

[0036] Another technical solution of the present invention: a device for estimating the state of a cluster of maneuvering targets based on iterative identification of directional angles, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above method is implemented when the processor executes the computer program.

[0037] The beneficial effect of the present invention is that the present invention can obtain the change of the target cluster's directional angle in real time by estimating the directional angle in real time, so that when the target cluster maneuvers and the cluster directional angle changes rapidly, the target cluster can be robustly tracked, thereby improving the state estimation accuracy of the target cluster when maneuvering. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of target cluster maneuvering in an embodiment of the present invention;

[0039] Figure 2 Schematic diagram of reference radial distance of target cluster in an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of target cluster measurement generation in an embodiment of the present invention;

[0041] Figure 4 Target cluster tracking result diagrams of different methods in embodiments of the present invention;

[0042] Figure 5 is a graph of the root mean square errors (RMSEs) of the cluster centroid x-axis of different methods in the embodiments of the present invention;

[0043] Figure 6 Graphs showing the root mean square errors (RMSEs) of the cluster centroid y axis for different methods in the embodiments of the present invention;

[0044] Figure 7 is a graph of cluster direction angle root mean square errors (RMSEs) of different methods in an embodiment of the present invention;

[0045] Figure 8 It is a graph of the intersection over union (IOU) of cluster profile estimation ranges of different methods in the embodiments of the present invention. DETAILED DESCRIPTION

[0046] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] The modeling of target motion within a cluster mainly considers that the individual motion of the target is not only related to its own state, but also affected by other targets. Typical methods include the continuous time differential equation method based on point target force analysis, Markov random field model, evolutionary network model, etc. However, these methods always consider the overall evolution of the group structure and the multi-target motion within the group separately. In fact, the target cluster is task-driven, and the targets within the group collaborate in the back-and-forth switching of the leading or following roles. The emergence of the leading-following model overcomes the shortcomings of the above methods. The dimension expansion method can obtain detailed position / velocity / acceleration estimates of the targets within the group. However, as the number of cluster targets increases, the high-dimensional state obtained by this dimension expansion method often leads to numerical instability of matrix operations and huge computational complexity.

[0048] With the development of drone cluster control technology, more and more point targets can be gathered into a group target. At this time, the effect of the dimension expansion method is greatly reduced, so a tracking method similar to the extended target came into being, that is, the second type of method, which can simultaneously estimate the centroid and contour of the target cluster. The modeling method according to the contour can be roughly divided into two categories, one is the modeling method of regular shape, and the other is the modeling method of irregular shape.

[0049] The modeling of regular shapes is mainly carried out using the random matrix method, which models the target's contour as an ellipse and considers the measurement points as being caused by the extension of the target. On this basis, many new methods have been developed, such as the estimation of maneuvering target clusters based on moment matching theory, and the use of variational methods to track target clusters when their orientation changes.

[0050] The modeling of irregular shapes mainly includes the multi-ellipse method, random hypersurface method and Gaussian process method. The multi-ellipse method is based on the theory of single ellipse, considering the target shape as a combination of multiple ellipses and taking the constraints between multiple ellipses into consideration, so as to achieve the overall tracking of cluster targets.

[0051] However, this method of describing the contour is not precise and is not applicable in situations where high-precision tracking is required. Therefore, methods based on the star convex model (drawing a line from the target center of mass to the edge of the target contour, the shape of the line still within the target range is called the star convex shape, and the line length is called the radial distance) have also been developed. The most representative ones are random hypersurfaces and Gaussian process methods. The random hypersurface describes the radial moment with the coefficients of the Fourier expansion and estimates it as a state. This method is not intuitive, while the Gaussian process learns the radial distance as a function of the polar angle θ in the local coordinate system. Relatively speaking, this method is more intuitive, and the Gaussian process's natural strong fitting ability for nonlinear regression and the online recursion ability for the posterior distribution allow it to obtain a good characterization of the cluster contour. However, existing Gaussian process methods rarely consider the situation of target maneuvers, especially the situation where the cluster orientation changes rapidly when the target maneuvers.

[0052] In many practical applications such as cluster target tracking in game confrontation environments, when targets perform some escape / attack tasks by maneuvering, the target orientation often changes rapidly. Existing dimensionality expansion methods are limited by radar resolution and computer computing power and cannot directly achieve tracking of high-density cluster targets. Modeling and tracking methods represented by random matrices lack an accurate description of the target contour, while methods represented by Gaussian processes have not fully considered the situation of rapid changes in target orientation, and there is little research on the situation of target maneuvers.

[0053] Therefore, based on the actual needs of mobile cluster target tracking in a game confrontation environment, the present invention proposes a mobile cluster target estimation method based on iterative reweighted least squares under a Gaussian process, aiming at the situation where the existing tracking methods are unable to handle the detailed description of the cluster contour under the rapid change of the target orientation. This method can obtain the target orientation estimation while achieving the detailed characterization of the mobile target contour.

[0054] The present invention discloses a method for estimating the state of a mobile target cluster based on iterative identification of azimuth angles, comprising the following steps: calculating a one-step prediction value of the state of a target cluster at moment k according to a state estimation value of the target cluster at moment k-1; iteratively updating the azimuth angle of the target cluster based on the one-step prediction value of the state of the target cluster at moment k; and determining a state estimation value of the target cluster at moment k according to a final azimuth angle after iterative updating.

[0055] The present invention can obtain the change of the target cluster's directional angle in real time by estimating the directional angle in real time, so that when the target cluster maneuvers and the cluster directional angle changes rapidly, the target cluster can be robustly tracked, thereby improving the state estimation accuracy of the target cluster when maneuvering.

[0056] The present invention designs a mobile target cluster estimation method based on iterative reweighted least squares under Gaussian process, which belongs to the field of information fusion and target tracking technology. In many practical applications, due to the density of target clusters and the limitation of radar resolution, the tracking radar cannot always obtain all the measurement information of the target cluster. At the same time, the tracked target cluster attempts to increase the tracking error of the radar or evade the detection of the radar in a maneuvering form. At this time, how to achieve robust and accurate tracking of the mobile target cluster under incomplete measurement when the target cluster is maneuvering is an important problem.

[0057] Consider the target cluster as follows Figure 1 The target cluster moves in the maneuvering form described above, and the contour of the target cluster can be modeled as a collection of cluster radial distances (referred to as reference radial distances) at a series of reference points. Therefore, the estimation of the cluster contour is converted into the estimation of the reference radial distance. It should be noted that the reference radial distance is based on the cluster centroid, and the estimation accuracy of the two is closely related to the accuracy of the cluster azimuth angle. Most existing methods do not focus on the estimation of the cluster azimuth angle. In fact, the maneuvering of the target cluster often leads to drastic changes in the cluster azimuth angle. Therefore, the present invention focuses on the robust tracking of the target cluster when the target maneuvers and the cluster azimuth angle changes rapidly.

[0058] Considering the similarity of the target motion within the cluster, the maneuver of the target cluster can be uniformly characterized by the maneuver of the cluster mass center. Therefore, the maneuver of the cluster mass center can be modeled by a jump Markov process. The motion vector of the target cluster mass center in the two-dimensional rectangular coordinate system o-ξη is: where [ξ k η k ] T represents the centroid position of the target cluster at time k, Represents the corresponding speed. On this basis, the center of mass maneuvering motion is modeled as follows using the jump Markov process:

[0059]

[0060] in, Represents the status information of the centroid of the target cluster at time k+1, Indicates that the parameter of the mode set at time k is Θ k The state transfer matrix at the time θ, the lower right corner subscript k represents the k moment, and the upper right corner subscript c represents the cluster centroid. k represents the jump Markov parameter under the finite pattern set {1,…,N}, N represents the number of pattern sets (i.e., the number of sub-patterns), and its model transition probability p ij =P{Θ k+1 =j|Θ k =i}, the initial mode probability μ i,0 =p{Θ0=i}. In the i-th mode, the system matrix in the corresponding state evolution equation is F k,i Here, F k,i It can represent target motion nominal models such as uniform linear motion, uniform left turn motion, uniform right turn motion, etc. represents the modeling error at time k, which can be described as a zero-mean Gaussian white noise sequence with a covariance of

[0061] The target cluster contour morphology is modeled using a set of reference radial distances, such as Figure 2 As shown, the reference radial distance is a function of the polar angle θ, that is, r = f(θ r ), where r=(r1,r2,...,r M ) T , M is the number of reference radial distances, r represents the reference radial distance set, θ r Represents the polar angle set. The estimation of the reference radial distance can be completed with the help of Gaussian Process (GP), where GP is a non-parametric method that defines the possible function distribution of a set of points. The distribution is mainly defined by the mean and covariance function:

[0062]

[0063] Where, m(u) = E[f(u)], κ(u,u′) = E[(f(u)-m(u))(f(u′)-m(u′))], u is the input of the function. From the definition of the reference radial distance, we can know that the input of the function is u = θ, and u′ represents the specified polar angle. Here, κ(:,:) is the covariance function, also called the kernel function. Let:

[0064]

[0065] σ f , σ rand l are their hyperparameters. Since Gaussian regression is a batch process and is not suitable for tasks with strong real-time requirements such as target tracking, the present invention uses recursive Gaussian regression and its state space form to complete the modeling of the reference radial distance, as follows:

[0066]

[0067] In the formula, the state vector represents the state transfer matrix, represents the modeling error, which can be described as a zero-mean Gaussian white noise sequence with a covariance of represents the measurement matrix, represents the radial distance noise at time k, Subject to mean zero and covariance The specific form is as follows:

[0068]

[0069]

[0070] In the formula, α represents the forgetting factor, T is the sampling interval, and R e Represents the measurement noise covariance matrix of the radial distance, θ k Measure r at time k k The corresponding radial distance angle is Figure 1 As shown, In this formula, It can be seen that in the present invention, the radial distance measurement matrix at time k is calculated based on the direction angle:

[0071] In the present invention, the local coordinate system is the target coordinate system, in which the center of mass is the origin, and the directions of the two coordinate axes can be selected as needed. K is the GP covariance matrix, where the GP covariance matrix corresponding to the a-dimensional vector s and the b-dimensional vector q is as follows:

[0072]

[0073] Note that Figure 2 As shown, the state space model of the cluster contour is built on the target body, that is, the angle θ in the model is relative to the target coordinate system, and the accuracy of the final target cluster estimation is not only closely related to the estimation accuracy of the cluster centroid and contour, but also the estimation accuracy of the cluster direction angle plays a vital role, especially the accurate estimation when the target cluster direction angle changes dramatically. In this regard, the present invention no longer performs traditional modeling estimation on the cluster direction angle, but uses it as a parameter for real-time estimation.

[0074] In fact, considering the real-time requirements of the algorithm and the structure of the actual measurement data, the present invention combines the state space model of the cluster centroid and the cluster contour to construct a joint state space model to achieve simultaneous estimation of the cluster centroid and the contour. Figure 1 , the measurement can be modeled as a surface model and a contour model. For simplicity, only the contour model is considered. Therefore, the joint state space model has the following structure:

[0075]

[0076] In the formula, c k =[ξ k η k ] T , d k,l represents the unit direction vector from the cluster centroid to the measurement point, R z is the measurement noise covariance. The upper right corner L represents the local coordinate system, G represents the global coordinate system, and the lower right corner l represents the lth data, such as z k,l represents the lth measurement data at time k, Indicates z k,l The angle in the local coordinate system, Indicates z k,l The angle in the global coordinate system.

[0077] From formula (12), we can see that the nonlinear measurement matrix at time k is based on the radial distance measurement matrix and the expanded state vector.

[0078] The unit direction vector is calculated as follows:

[0079]

[0080] also, and Here, the cluster direction angle is unknown and not modeled.

[0081] Taking into account the nonlinearity of the above measurement equation and the pattern uncertainty it contains, the recursive estimation of the above state space model can be carried out by using an interactive multi-model assisted by a nonlinear filter, such as an interactive multi-model assisted by an extended Kalman filter (IMM-EKF). However, since the expanded state vector contains both the center of mass and the reference radial distance set, the traditional interactive input / output method is no longer applicable to the interaction of the reference radial distance. This is because the cluster azimuth angles estimated in different sub-modes (i.e., modes corresponding to different jump Markov parameters) have different accuracies. This radial distance estimation based on inaccurate cluster azimuth angles will have a certain impact on the estimation accuracy of the reference radial distance, thereby affecting the accuracy of the radial distance after the interaction.

[0082] Therefore, the present invention proposes a new interactive multi-model structure, which first performs a weighted summation on the estimated values ​​of the cluster direction angles under different sub-modes, and then estimates and interacts the expanded-dimensional state vectors under different sub-modes based on this. In addition, considering the particularity of the radial value function, the present invention introduces the interaction of two-dimensional Fourier transform-assisted radial distances to improve its estimation accuracy and robustness. Finally, considering the time complexity of the algorithm, it is proposed to use iterative reweighted least squares method to estimate the cluster direction angles under each sub-mode, which improves the estimation accuracy while sharing the derivation process of the measurement matrix in the EKF, reducing the time complexity.

[0083] Next, the application of iterative reweighted least squares in maneuvering target cluster tracking and the new structure of the proposed interactive multi-model are introduced in detail.

[0084] The direction angle of the target cluster is iteratively updated based on the one-step prediction of the state of the target cluster at time k, including: calculating the measurement error Δz based on the measurement vector and nonlinear measurement matrix of the target cluster at time k k ; Among them, the nonlinear measurement matrix is ​​calculated based on the state one-step prediction and the direction angle; the direction angle increment is generated based on the measurement error The bearing angle and bearing increment The new direction angle is merged into a new direction angle; when the difference between the new direction angle and the direction angle of the previous iterative update is less than or equal to the threshold or the number of iterative updates reaches the maximum number, the new direction angle is used as the final direction angle.

[0085] In the i-th sub-mode, the predicted values ​​of the state vector and covariance at time k are as follows:

[0086]

[0087] Based on the above nonlinear measurement equation, the optimization goal of iterative reweighted least squares is:

[0088]

[0089] in, represents the estimated value of the cluster direction angle at time k, z k =[z k,1 z k,2 … z k,y ] T , z k represents the measurement vector at time k, z k,y represents the position of the yth measurement point, h k =[h k,1 h k,2 … h k,y ] T ,h k represents the measurement matrix at time k, R k =blkdiag(R k,1 ,R k,2 ,...,R k,y ), R k represents the error covariance matrix at time k, y is the number of measurements at time k, and the result of the tth iteration of the given direction angle in the i-th sub-mode, Then the equivalent measurement equation has the following form:

[0090]

[0091] in, Note that in the above derivation process, the state vector x k,i is also unknown. Here, let x k,i =x k|k-1,i .

[0092] Therefore, Equation (17) is about the direction angle increment The linear equation of , according to the result of least squares:

[0093]

[0094] In the formula, R k According to the initial designation Calculated.

[0095] The direction angle increment is generated based on the measurement error by Calculated, represents the direction angle variable of the tth iteration at time k, represents the derivative of the nonlinear measurement matrix with respect to the direction angle at the t-th iteration at time k, represents the measurement error covariance matrix of the t-th iteration at time k, represents the measurement error of the tth iteration at time k.

[0096] Therefore, the result of the tth iteration in the i-th mode is:

[0097]

[0098] The above iterative process continues until the termination condition is met, such as:

[0099]

[0100] therefore,

[0101] The new interactive multi-model structure is introduced below, and the specific process is as follows.

[0102] In the present invention, the state estimation includes state estimation under different sub-modes; wherein the sub-modes are different motion state models in the state estimation process; and each sub-mode has its corresponding posterior probability.

[0103] 1) Interaction of extended dimensional states.

[0104] make and is the state estimate and corresponding covariance matrix of sub-mode j at time k-1, then the interactive input of the j-th sub-mode at time k is as follows:

[0105] Centroid Interaction:

[0106]

[0107] represents the interactive input result of the centroid of the target cluster in the jth sub-mode at time k-1,

[0108] represents the estimated result of the centroid of the target cluster under the i-th sub-mode at time k-1, μ k-1|k-1 (i|j) represents the normalized probability of the i-th sub-pattern transferring to the j-th sub-pattern at time k-1.

[0109]

[0110] represents the covariance interaction input result of the target cluster centroid in the jth sub-mode at time k-1, represents the cosquare estimation result of the target cluster centroid in the i-th sub-mode at time k-1, (·) represents

[0111] In the formula, π ij represents the prior probability of transitioning from sub-mode i to sub-mode j, μ k-1|k-1,i represents the posterior probability of the i-th sub-pattern at time k-1, is the normalization constant.

[0112] Radial distance interaction:

[0113]

[0114] In the formula, the upper right corner is r * Represents the coordinate point corresponding to the reference radial distance, Represents the result of interactive input, represents the estimated value; FFT2 and IFFT2 represent Fourier transform and its inverse transform, EKF j Represents the filter function under the jth sub-mode. It represents the set of coordinate values ​​of the radial distance from the end point in the jth sub-mode at time k-1; represents the estimated radial distance set under the jth submode at time k-1; The estimated coordinates of the center of mass of the jth sub-mode at time k-1 respectively. It represents the interactive input result set of radial distance from the endpoint in the jth sub-mode at time k-1; Represents the interactive input of the cluster radial distance set at the jth sub-mode at time k-1.

[0115] 2) Sub-model posterior probability update.

[0116] According to the above interactive input results, the preliminary estimation results of the j-th sub-mode at time k are as follows:

[0117]

[0118] The right side of the first equation in formula (27) The expanded state vector is obtained by concatenating the results of formulas (21) and (25). It is the concatenation of the results obtained from formula (22) and (26), z k represents the measurement vector at time k, z k|k-1,j represents the measured predicted value in the jth sub-mode at time k. k|k-1,j It represents the predicted value (i.e., one-step prediction) of the expanded state vector under the jth sub-mode obtained at time k by using the parameters at time k-1. I is a unit pair whose diagonal elements are all 1 and other elements are all 0. The direction angle of the jth sub-mode at time k Obtained by iterative reweighted iterative least squares, H k h k The Jacobian matrix of the one-step-forecast value of the state vector.

[0119] Assuming that the measurement points on the target cluster contour are independent of each other, the posterior probability μ of the jth sub-pattern is k|k,j It can be obtained as follows:

[0120]

[0121] Λ k,j =Λ k,j,1 Λ k,j,2 …Λ k,j,y (29)

[0122]

[0123] Here, μ k|k,j represents the posterior probability corresponding to the jth sub-pattern at time k, the l in the lower right corner represents the lth measurement, Λ k,j,l represents the likelihood function value of the lth measurement under the jth sub-mode at time k, S k,j =blkdiag(S k,j,1 ,S k,j,2 ,…,S k,j,y ), Λ k,j represents the likelihood function value of the jth sub-pattern at time k, μ k-1|k-1,i represents the posterior probability corresponding to the i-th sub-mode at time k-1, and N represents the number of sub-modes. It is calculated by the final direction angle and state one-step prediction amount under each sub-mode, that is, the posterior probability of each sub-mode can be calculated based on the final direction angle and state one-step prediction amount under each sub-mode.

[0124] 3) Cluster direction angle estimation fusion.

[0125] The estimated direction angle of the target cluster is estimated based on the final direction angle in each sub-mode. First, the estimated direction angle is generated based on the final direction angle in each sub-mode and the corresponding posterior probability. The estimated direction angle of the target cluster at time k is updated as follows:

[0126]

[0127] 4) Sub-filter dimension expansion state update.

[0128] After the estimated direction angle is obtained, the expanded dimension state estimator can be determined based on it. That is, determining the state estimator based on the estimated direction angle includes: calculating the estimated state quantity under each sub-mode based on the estimated direction angle and updating the posterior probability of each sub-mode; decomposing the estimated state quantity under each sub-mode to generate the estimated state quantity of the target cluster centroid and the radial distance set state estimator under the sub-mode; interacting the estimated state quantity of the target cluster centroid and the radial distance set state estimator under different sub-modes respectively to obtain the new estimated state quantity of the target cluster centroid and the radial distance set estimated state quantity; combining the new estimated state quantity of the target cluster centroid and the radial distance set estimated state quantity to obtain the estimated state quantity.

[0129] make i = 1, 2, ..., N, and the extended state vector and its covariance matrix are re-encoded according to the filter equation system described in equation (27) and the posterior probability μ of the subpattern k|k,i to update.

[0130] 5) Extended dimensionality state estimation fusion.

[0131] The cluster centroid state and cluster contour state are decomposed according to the updated expanded-dimensional state vector, and weighted fusion is performed on them based on the posterior probabilities of the corresponding sub-patterns.

[0132] Centroid output:

[0133]

[0134] Cluster reference radial distance output:

[0135]

[0136] Therefore, the state of the target cluster at time k is estimated to be The cluster direction angle is

[0137] The pseudo code flow of the above method as an algorithm can be seen in Table 1.

[0138] Table 1 Pseudocode flow of FLMGP

[0139]

[0140]

[0141] The method of the present invention has been verified to be effective as follows.

[0142] To verify the effectiveness of the present invention, 500 Monte Carlo simulation experiments were conducted in the following typical scenario. Assuming that the target performs an “S”-shaped maneuver in the o-ξη plane, the target state vector The filtering algorithm model set contains three sub-modes, namely N = 3, namely uniform linear motion (CV) + contour evolution model, uniform left turn motion (LCT) + contour evolution model, uniform right turn motion (RCT) + contour evolution model, and the corresponding dynamic evolution models are as follows:

[0143]

[0144]

[0145] In the formula,

[0146] When the turning rate ω>0, it is a left turn; when the turning rate ω<0, it is a right turn; where T=1, T represents the sampling time interval, σ q =0.01, ω = ±9rad / s (i.e. +9 or -9). Initial state of center of mass (100 / 50 represents the horizontal and vertical coordinates, 20 and 0 represent the speed in the horizontal and vertical coordinate axis directions), the initial state of the profile is determined according to the trend of the reference radial distance function, here, the number of reference radial distances M = 25. The minimum distance between targets in the cluster (the minimum distance between adjacent targets; the formation remains unchanged during flight) is 2.9m, the maximum distance is 3.9m, there are 50 target points in the cluster, and the cluster direction angle is The speed varies in a "sawtooth" manner between ±πrad.

[0147] The radar located at the origin of the o-ξη plane detects the target and randomly obtains the position observations of 20 target points in the group at each sampling time. The radar measurement noise covariance R = σ 2 I2,σ 2 =9, where the target motion simulation is as follows Figure 3 ,The figure shows the changing trends of the measurement points, centroid and direction angle of the target cluster.

[0148] The proposed interactive multi-model filtering algorithm (FLMGP) assisted by Fourier transform and iterative reweighted least squares, the interactive multi-model filtering algorithm (LSMGP) assisted by iterative reweighted least squares, the filtering algorithm based on interactive multi-models under azimuth uniform velocity modeling (CVMGP), the filtering algorithm based on interactive multi-models under azimuth maneuver modeling (MM2GP), and the filtering algorithm based on centroid and azimuth uniform velocity modeling (CV2GP) were compared in experiments. Among them, azimuth maneuver modeling refers to the use of an interactive multi-model algorithm containing two sub-modes, uniform velocity and uniform acceleration, to estimate the azimuth. The initial values ​​of the states of each algorithm are consistent, and the initial state of the cluster azimuth is The corresponding angular velocity and angular acceleration are also 0. In addition, the hyperparameter σ of the Gaussian process r =2,σ f =2,

[0149] In a single Monte Carlo simulation, the tracking results of all algorithms are as follows Figure 4 In 500 Monte Carlo simulations, the root mean square errors (RMSEs) of the cluster centroid positions and cluster centroid direction angles of all algorithm filtering outputs are shown in Figure 5 , Figure 6 and Figure 7 As shown, Figure 5 and Figure 6 are the root mean square errors of the horizontal and vertical coordinates of the centroid, Figure 7is the root mean square error of the direction angle, and the intersection over union (IOU) of the corresponding cluster contour range and the true contour range is as follows: Figure 8 Here:

[0150]

[0151] in, Represents the state x in the mth Monte Carlo simulation k The j-th element of represents the corresponding estimated value, is the true value, is the estimated area of ​​the cluster target contour range at time k, S k is the actual area of ​​the cluster target contour range at time k, U represents the total number of Monte Carlo simulations, and K represents the total sampling step length of the system operation.

[0152] From the results, we can see that even in the initial stage of tracking, when all the comparison algorithms are under the condition of model adaptation, that is, the target cluster moves at a constant speed and with a constant direction angle, the proposed FLMGP algorithm can still achieve the best tracking accuracy. As the target cluster maneuvers, the tracking accuracy of the comparison algorithms (except LSMGP) drops significantly. Although LSMGP can also track normally, the peak error of FLMGP is always smaller than that of LSMGP. In summary, the filtering accuracy of FLMGP is the best among all the comparison algorithms, and compared with LSMGP, it can be found that the introduction of Fourier transform improves the accuracy and robustness of the algorithm to a certain extent.

[0153] In summary, the present invention defines a new interactive multi-model algorithm framework to address the problem that the traditional interactive multi-model algorithm ignores the influence of the cluster direction angle on the cluster centroid and reference radial distance estimation accuracy when performing the interaction of the target cluster dimensional expansion state vector, which reduces the influence of the cluster direction angle estimation accuracy to a certain extent, so as to obtain more accurate state interaction input / output. In view of the particularity of the target cluster reference radial distance, it is proposed to use Fourier transform to assist the interaction of the reference radial distance state vector, and obtain its coordinate points through inverse Fourier transform as pseudo-measurements and input them into the corresponding filter to obtain the interaction of the reference radial distance state vector covariance. In view of the problem that the estimation accuracy of the target cluster direction angle is not high under the assistance of the model, it is proposed to use the iterative reweighted least squares method to use the direction angle as a parameter for real-time estimation, which further improves the estimation accuracy of the target cluster.

[0154] In many practical applications such as tracking of maneuvering target clusters in game confrontation environments, when targets perform some attack / escape tasks through maneuvers, the target cluster's azimuth often changes rapidly. As the existing methods can only achieve a rough characterization of the cluster outline when the target cluster's azimuth changes slowly, a maneuvering cluster target estimation method based on iterative reweighted least squares under Gaussian process is proposed.

[0155] Therefore, in the method of the present invention, firstly, based on the target cluster motion example, the state space model of the maneuvering target cluster without cluster azimuth modeling is constructed. For the special extended dimension vector of cluster centroid + reference radial distance, a new interactive multi-model algorithm is defined to reduce the influence of low-precision cluster azimuth on state estimation and state interaction. At the same time, it is proposed to use Fourier transform to assist the interaction of cluster reference radial distance to further improve the accuracy of state estimation. Finally, the iterative reweighted least squares method is used to estimate the target cluster azimuth in real time, and the tracking characterization of high-precision maneuvering target cluster is realized. The simulation of typical scenarios of maneuvering target cluster tracking shows that the proposed method can effectively handle the situation of rapid changes in cluster azimuth angle, and can obtain higher estimation accuracy than the interactive multi-model filtering algorithm (LSMGP) assisted by iterative reweighted least squares, the filtering algorithm (CVMGP) based on interactive multi-model under azimuth angle uniform velocity modeling, the filtering algorithm (MM2GP) based on interactive multi-model under azimuth angle maneuver modeling, and the filtering algorithm (CV2GP) based on centroid and azimuth angle uniform velocity modeling.

[0156] The present invention also discloses a device for estimating the state of a mobile target cluster based on iterative identification of azimuth angles, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The method is implemented when the processor executes the computer program.

[0157] The present invention also discloses an embodiment that provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0158] The present invention also provides a computer program product. When the computer program product runs on a data storage device, the data storage device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0159] If the integrated unit module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device capable of carrying the computer program code to a storage device, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electrical carrier signal, a telecommunication signal, and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0160] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0161] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0162] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules or units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0163] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Claims

1. A method for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles, characterized in that: The following steps are involved: Calculate the one-step prediction of the state of the target cluster at time k based on the state estimate of the target cluster at time k-1; Iteratively updating the direction angle of the target cluster based on the one-step prediction of the state of the target cluster at time k; The state estimation of the target cluster at time k is determined according to the final direction angle after iterative update.

2. A method for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles as claimed in claim 1, characterized in that: Iteratively updating the direction angle of the target cluster based on the one-step prediction of the state of the target cluster at time k includes: Calculate the measurement error Δz based on the measurement vector and nonlinear measurement matrix of the target cluster at time k k ; Wherein, the nonlinear measurement matrix is ​​calculated based on the state one-step prediction quantity and the direction angle; Generate a direction angle increment based on the measurement error The direction angle and the direction angle increment Merge into new directions; When the difference between the new direction angle and the direction angle of the previous iterative update is less than or equal to a threshold or the number of iterative updates reaches a maximum number, the new direction angle is used as the final direction angle.

3. The method for estimating the state of a maneuvering target cluster based on iterative identification of direction angles according to claim 2, characterized in that: The method for generating the nonlinear measurement matrix includes: Calculate the radial distance measurement matrix at time k based on the direction angle According to the radial distance measurement matrix The one-step prediction of the state generates a nonlinear measurement matrix at time k.

4. The method for estimating the state of a maneuvering target cluster based on iterative identification of direction angles according to claim 2, characterized in that: Generate a direction angle increment based on the measurement error include: in, represents the direction angle increment of the t-th iteration at time k, represents the derivative of the nonlinear measurement matrix at the t-th iteration at time k with respect to the direction angle, represents the measurement error covariance matrix of the t-th iteration at time k, represents the measurement error of the tth iteration at time k.

5. A method for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles according to any one of claims 2 to 4, characterized in that: The state estimation includes state estimation under different sub-modes; wherein the sub-modes are different motion state models in the state estimation process; Each of the submodules has its corresponding posterior probability.

6. The method for estimating the state of a maneuvering target cluster based on iterative identification of direction angles according to claim 5, characterized in that: The state estimation of the target cluster at time k is determined according to the final direction angle after iterative update, including: estimating an estimated direction angle of the target cluster according to the final direction angle in each of the sub-modes; The state estimation amount is determined according to the estimated direction angle.

7. The method for estimating the state of a maneuvering target cluster based on iterative identification of direction angles according to claim 6, characterized in that: Estimating the estimated direction angle of the target cluster according to the final direction angle in each of the sub-modes includes: Calculating the posterior probability of each sub-mode according to the final direction angle and the state one-step prediction quantity under each sub-mode; The estimated direction angle is generated according to the final direction angle in each of the sub-modes and the corresponding posterior probability.

8. The method for estimating the state of a maneuvering target cluster based on iterative identification of direction angles according to claim 7, characterized in that: Calculating the posterior probability of each of the sub-patterns includes: Among them, μ k|k,j represents the posterior probability corresponding to the jth sub-pattern at time k, Λ k,j represents the likelihood function value of the jth sub-pattern at time k, π ij represents the prior probability of transitioning from sub-mode i to sub-mode j, μ k-1|k-1,i represents the posterior probability corresponding to the i-th sub-pattern at time k-1, and N represents the number of sub-patterns.

9. A method for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles as claimed in claim 6, 7 or 8, characterized in that: Determining the state estimation amount according to the estimated direction angle comprises: Calculating the estimated state quantity in each sub-mode according to the estimated direction angle and updating the posterior probability of each sub-mode; Decompose the estimated state quantity under each sub-mode to generate the estimated state quantity of the centroid of the target cluster and the estimated state quantity of the radial distance set under the sub-mode; The estimated state quantities of the target cluster centroid and the estimated state quantities of the radial distance set under different sub-modes are interacted respectively to obtain new estimated state quantities of the target cluster centroid and the estimated state quantities of the radial distance set; The estimated state quantity of the new target cluster centroid and the estimated state quantity of the radial distance set are combined to obtain the estimated state quantity.

10. A device for estimating the state of a maneuvering target cluster based on iterative identification of azimuth angles, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

Images