A method and device for multi-target tracking in a clutter environment with RCS information assistance

Abstract

The application discloses a RCS information auxiliary multi-target tracking method and device in a clutter environment, and the method comprises the following steps: constructing a Bayesian RCS information recursive estimation mechanism; introducing the Bayesian RCS information recursive estimation mechanism into a Poisson multi-Bernoulli mixed filter, augmenting a target state vector, and respectively deducing a corresponding prediction density in a gamma Gaussian mixed form for a Poisson part and a multi-Bernoulli mixed part; respectively deducing an updating density in a gamma Gaussian mixed form for the Poisson part and the multi-Bernoulli mixed part; combining the updating density in the gamma Gaussian mixed form for the Poisson part and the multi-Bernoulli mixed part to construct a cost matrix and an assignment matrix of a global hypothesis; after the cost matrix and the assignment matrix are obtained, generating k optimal global hypotheses and weights corresponding to the global hypotheses, selecting a global hypothesis with the maximum weight, and obtaining a multi-target motion state estimation. The application improves the multi-target state estimation precision and improves the tracking performance of the multi-target.

Description

Technical Field

[0001] This invention belongs to the field of intelligent sensing and processing technology, specifically relating to a multi-target tracking method and device assisted by RCS information in clutter environments. Background Technology

[0002] Currently, accurately and in real-time detecting all targets within a monitored area and predicting their trajectories is one of the key research directions in the field of radar data processing. The physical environment in which radar operates is extremely harsh, and the received data often contains a large amount of clutter or false alarms. Multi-target tracking in cluttered environments often faces challenges such as false alarms, missed detections, and potential overestimation. A large number of missed targets and false alarm measurements easily lead to problems such as a high probability of false track initiation, low track initiation accuracy, erroneous associations, and a "combinatorial explosion" of association numbers, posing a severe challenge to multi-target tracking. Summary of the Invention

[0003] To address this issue, the present invention provides a multi-target tracking method and apparatus assisted by RCS information in clutter environments, which solves the problem that traditional technologies cannot effectively distinguish between targets and clutter in clutter environments, resulting in low tracking accuracy for multiple targets.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a multi-target tracking method assisted by RCS information in a cluttered environment, comprising:

[0005] A Bayesian RCS information recursive estimation mechanism is constructed. The Bayesian RCS information recursive estimation mechanism uses a gamma distribution to represent the RCS correlation distribution, defines the probability density function of the inverse RCS parameter a as following a gamma distribution, and uses the conjugate of the gamma distribution and the Bayesian criterion to derive the prediction update expression of the target RCS posterior density.

[0006] The Bayesian RCS information recursive estimation mechanism is introduced into the Poisson-Dobernoli mixture filter, and the target state vector is augmented. The prediction part of the Gamma-Gaussian mixture implementation process of the Poisson-Dobernoli mixture filter assisted by RCS information is derived. The multi-target posterior density is expressed as a Gamma-Gaussian mixture form. The prediction density of the corresponding Gamma-Gaussian mixture form is derived for the Poisson part and the Dobernoli mixture part respectively.

[0007] The update part of the Poisson-Dobernoli mixture filter with RCS information assistance is derived in the process of gamma-Gaussian mixture. The posterior density of the multi-target is expressed as a gamma-Gaussian mixture. The update density of the corresponding gamma-Gaussian mixture is derived for the Poisson part and the Dobernoli mixture part respectively. The Poisson part represents the target that has not yet been detected, and the Dobernoli mixture part represents the missed target, the existing target and the newly generated target.

[0008] By combining the update density of the Gamma-Gaussian mixture form corresponding to the Poisson part and the multi-Bernoulli mixture part, the cost matrix and assignment matrix of the global hypothesis are constructed.

[0009] After obtaining the cost matrix and allocation matrix, the Murty algorithm is used to generate k optimal global hypotheses and their corresponding weights. The global hypothesis with the largest weight is selected to obtain the multi-target motion state estimate.

[0010] The existing global assumptions are trimmed and merged.

[0011] As a preferred method for multi-target tracking in clutter environments assisted by RCS information, the Bayesian RCS information recursive estimation mechanism is constructed by defining the inverse RCS parameter 'a' as a random variable following a gamma distribution. It is assumed that the initial distribution parameters of 'a' are known, i.e., α0 = β0 = 0. Then, the probability density function of 'a' is:

[0012]

[0013] The evolutionary iterative formula for a is:

[0014]

[0015] In the formula, W j J(·) is an exponentially independent and identically distributed random variable with parameter 1 / d; J(·) is a Poisson distributed random variable with parameter a / d; α0 is the shape parameter; β0 is the c-scale parameter; Γ(·) is the ordinary gamma function.

[0016] Assuming the RCS observation of the target at time k-1 follows a gamma distribution, the expression is:

[0017]

[0018] In the formula, b is the shape parameter, a is the inverse RCS parameter that follows a gamma distribution as defined above, and the value of b is chosen so that the probability density function covers all Swerling models; σ t The RCS measurement value for the target.

[0019] As a preferred scheme for multi-target tracking methods in RCS-assisted clutter environments, p is set k-1 (a|σ t The distribution follows a gamma distribution, and the predicted distribution obtained according to Bayes' theorem is:

[0020]

[0021] In the formula, a is the inverse RCS parameter, σ t p is the RCS measurement value. k|k-1(a|a') is the state transition density function of the inverse RCS parameter a;

[0022] Using a moment matching strategy, the prediction distribution is approximated as a gamma distribution, i.e.:

[0023]

[0024]

[0025] In the formula, d is a parameter in the distribution that iteratively generates a. Let be the shape parameter of the gamma distribution at time k; Let α' be the scale parameter of the gamma distribution at time k; β' be the shape parameter of the gamma distribution at time k-1; and β' be the scale parameter of the gamma distribution at time k-1. Using Bayes' theorem, the updated distribution follows a gamma distribution, i.e.:

[0026]

[0027]

[0028] In the formula, τ is the detection threshold, which is used to calculate the inverse RCS parameter a and the minimum mean square error estimate of the local average RCS.

[0029] As a preferred method for multi-target tracking in cluttered environments assisted by RCS information, the expression for augmenting the target state vector is as follows:

[0030] x k =[p k,x ,v k,x ,p k,y ,v k,y ,a k ] T

[0031] In the formula, location information Represents position coordinates, v x,y Represents velocity; the measurement vector is represented as In the formula, This indicates the measurement value representing the target's location information; Represents the measured value of the target's RCS;

[0032] Assume the intensity function of the Poisson process at time k-1 is a gamma-Gaussian mixture, expressed as:

[0033]

[0034] The predicted Poisson process intensity is also in the form of a gamma-Gaussian mixture, expressed as:

[0035]

[0036] In the formula, J u,k|k-1 =J u,k-1 +v b , J u,k|k-1 J represents the number of Gaussian components predicted for the undetected target at time k. u,k-1 v represents the number of Gaussian components of the undetected target at time k-1. b Indicates the target number of new students. p represents the weight value predicted for the i-th Gaussian component of the undetected target at time k. s This represents the probability of the target's existence. This represents the weight value of the i-th Gaussian component of the target that was not detected at time k-1. This represents the shape parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the shape parameter of the i-th Gaussian component of the target that was not detected at time k-1. This represents the scale parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the scale parameter of the i-th Gaussian component of the target that was not detected at time k-1. This represents the mean of the predictions for the i-th Gaussian component of the undetected target at time k. Let represent the mean of the i-th Gaussian component of the undetected target at time k-1, and F represent the target's state transition matrix. This represents the covariance matrix predicted by the i-th Gaussian component of the undetected target at time k. Q represents the covariance matrix of the i-th Gaussian component of the undetected target at time k-1. k-1 This represents the process noise matrix at time k-1.

[0037] As a preferred scheme for multi-target tracking methods in cluttered environments assisted by RCS information, for the multi-Bernoulli mixture, if the Bernoulli density of the i-th target under the h-th global assumption at time k-1 is a gamma-Gaussian form, that is:

[0038]

[0039] The corresponding prediction density is:

[0040]

[0041]

[0042]

[0043] In the formula, Let represent the weight value of the predicted i-th target under the h-th global hypothesis at time k. Let represent the weight value of the i-th objective under the h-th global assumption at time k-1. This represents the probability of the existence of the i-th target prediction under the h-th global hypothesis at time k. This represents the probability of the existence of the i-th target under the h-th global hypothesis at time k-1. This represents the shape parameter predicted for the i-th target under the h-th global assumption at time k. This represents the shape parameter of the i-th target under the h-th global assumption at time k-1. Let represent the scale parameter for the prediction of the i-th target under the h-th global hypothesis at time k. Let represent the scale parameter of the i-th target under the h-th global assumption at time k-1. Let represent the mean of the predictions for the i-th target under the h-th global hypothesis at time k. Let represent the mean of the i-th objective under the h-th global assumption at time k-1. Let represent the covariance matrix of the prediction of the i-th target under the h-th global assumption at time k. Let represent the covariance matrix of the i-th target under the h-th global assumption at time k-1.

[0044] As a preferred method for multi-target tracking in cluttered environments with RCS information assistance, the update density in the corresponding gamma-Gaussian mixture form is derived from the Poisson part. The updated intensity function is expressed as follows:

[0045]

[0046] In the formula, J u,k =J u,k|k-1 , J u,k J represents the number of Gaussian components of the undetected target at time k. u,k|k-1 This represents the number of Gaussian components predicted for undetected targets at time k-1. This represents the updated weight value of the i-th Gaussian component of the target that was not detected at time k. Let represent the weight value predicted for the i-th Gaussian component of the undetected target at time k, and τ represent the threshold value of the inverse RCS parameter. This represents the shape parameter updated by the i-th Gaussian component of the undetected target at time k. This represents the shape parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the scale parameter updated for the i-th Gaussian component of the undetected target at time k. This represents the scale parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the mean of the update of the i-th Gaussian component of the undetected target at time k. This represents the mean of the predictions for the i-th Gaussian component of the undetected target at time k. This represents the covariance matrix updated by the i-th Gaussian component of the undetected target at time k. Let represent the covariance matrix of the prediction of the i-th Gaussian component of the undetected target at time k.

[0047] As a preferred scheme for multi-target tracking methods in cluttered environments assisted by RCS information, the update density of the corresponding gamma-Gaussian mixture form is derived for the multi-Bernoulli mixture part, considering the update of the Bernoulli terms for three types of targets:

[0048] (I) Update the missed detection status of the i-th target under the h-th global hypothesis:

[0049]

[0050]

[0051]

[0052] In the formula, Let represent the updated weight value of the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the probability of the i-th target being updated under the h-th global hypothesis in the case of a missed detection at time k. This represents the shape parameter updated for the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the scale parameter updated for the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the mean update value of the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the covariance matrix updated for the i-th target under the h-th global hypothesis in the case of missed detection at time k;

[0053] (II) Update the past potential existence of the i-th target under the h-th global assumption of measurement z:

[0054]

[0055]

[0056]

[0057] In the formula, This represents the updated weight value of the i-th target under the h-th global hypothesis, assuming that time k exists. This represents the probability of the i-th target being updated under the h-th global hypothesis given that the target exists at time k. This represents the shape parameter updated for the i-th target under the h-th global assumption, assuming the existence of time k. This represents the scale parameter updated for the i-th target under the h-th global hypothesis, assuming the existence of time k. Let H represent the mean update of the i-th target under the h-th global hypothesis given that time k exists, where H represents the target's measurement matrix, K represents the target's Kalman gain, and I represents the identity matrix. Let represent the covariance matrix updated by the i-th target under the h-th global assumption given the existence of time k;

[0058] (III) Target update for the first detection of measurement z:

[0059]

[0060]

[0061] In the formula, e k This represents the target's weight value during the first detection at time k. This represents the weight value of the i-th Gaussian component in the case of the first detection at time k. λ represents the weight value in the case of the first detection at time k. c Indicates clutter intensity. This represents the probability of the target's existence under the first detection at time k. This represents the shape parameter of the i-th Gaussian component under the first detection at time k. Let represent the scale parameter of the i-th Gaussian component in the case of the first detection at time k. Let represent the mean of the i-th Gaussian component under the first detection at time k. R represents the covariance matrix of the i-th Gaussian component under the first detection at time k. k This represents the measurement noise matrix of the target at time k.

[0062] As the preferred method for multi-target tracking in cluttered environments assisted by RCS information, the cost matrix of the global assumption is constructed as follows:

[0063]

[0064]

[0065] In the formula, η represents the weight of different types of objectives; F o F represents the weight matrix of previously detected targets. p This represents the weight matrix of the first detected target.

[0066] The allocation matrix S is an m×(m+n) matrix containing either 0 or 1 entries. o ) type matrix.

[0067] As a preferred method for multi-target tracking in cluttered environments with RCS information assistance, after obtaining the cost matrix and allocation matrix, the Murty algorithm is used to generate k optimal global hypotheses and their corresponding weights, and the global hypothesis with the largest weight is selected. The expression is:

[0068]

[0069] In the formula, This represents the total weight value under the h-th global hypothesis at time k. This represents the weight value of the i-th objective under the h-th global assumption at time k.

[0070] Based on the obtained optimal global assumptions, the corresponding multi-objective state estimates are extracted.

[0071] As a preferred scheme for multi-target tracking methods in clutter environments assisted by RCS information, the existing global assumptions are trimmed and merged during the trimming and merging process. Global assumptions with weights below a set threshold are trimmed, and similar global assumptions are merged.

[0072] The present invention also provides an RCS-assisted multi-target tracking device in a clutter environment, employing the aforementioned RCS-assisted multi-target tracking method in a clutter environment, comprising:

[0073] The recursive estimation mechanism construction module is used to construct a Bayesian RCS information recursive estimation mechanism. The Bayesian RCS information recursive estimation mechanism uses a gamma distribution to represent the RCS correlation distribution, defines the probability density function of the inverse RCS parameter a as following a gamma distribution, and uses the conjugate of the gamma distribution and the Bayesian criterion to derive the prediction update expression of the target RCS posterior density.

[0074] The predictive analysis module is used to introduce the Bayesian RCS information recursive estimation mechanism into the Poisson-Dobernoli mixture filter, augment the target state vector, derive the prediction part of the Gamma-Gaussian mixture implementation process of the Poisson-Dobernoli mixture filter assisted by RCS information, express the multi-target posterior density in the form of Gamma-Gaussian mixture, and derive the corresponding prediction density of the Gamma-Gaussian mixture form for the Poisson part and the Dobernoli mixture part respectively.

[0075] The update analysis module is used to derive the update part in the gamma-Gaussian mixture implementation process of the Poisson-Dobernouri mixture filter assisted by RCS information. It expresses the posterior density of multiple targets as a gamma-Gaussian mixture form, and derives the update density of the corresponding gamma-Gaussian mixture form for the Poisson part and the Dobernouri mixture part respectively. The Poisson part represents targets that have not yet been detected, and the Dobernouri mixture part represents missed targets, existing targets, and newly generated targets.

[0076] The global hypothesis processing module is used to combine the update density of the gamma-Gaussian mixture form corresponding to the Poisson part and the multi-Bernoulli mixture part to construct the cost matrix and assignment matrix of the global hypothesis.

[0077] The motion state estimation module is used to generate k optimal global hypotheses and their corresponding weights using the Murty algorithm after obtaining the cost matrix and allocation matrix. The global hypothesis with the largest weight is selected to obtain the motion state estimation of the multi-target target.

[0078] The global hypothesis post-processing module is used to trim and merge existing global hypotheses.

[0079] This invention has the following advantages: By constructing a Bayesian RCS information recursive estimation mechanism, which uses a gamma distribution to represent the RCS correlation distribution, defines the probability density function of the inverse RCS parameter a as following a gamma distribution, and uses the conjugate of the gamma distribution and the Bayesian criterion to derive the prediction update expression of the target RCS posterior density; the Bayesian RCS information recursive estimation mechanism is introduced into a Poisson-Dobernoli mixture filter, and the target state vector is augmented to derive the prediction part of the gamma-Gaussian mixture implementation process of the Poisson-Dobernoli mixture filter assisted by RCS information, expressing the multi-target posterior density as a gamma-Gaussian mixture form, and deriving the corresponding gamma-Gaussian mixture form prediction density for the Poisson part and the Dobernoli mixture part respectively; the RCS is derived. In the update part of the information-assisted Poisson-Dobernoli mixture filter's gamma-Gaussian mixture implementation, the multi-target posterior density is expressed as a gamma-Gaussian mixture form. The update density of the corresponding gamma-Gaussian mixture form is derived for both the Poisson and Dobernoli mixture parts. The Poisson part represents targets not yet detected, while the Dobernoli mixture part represents missed targets, existing targets, and newly detected targets. Combining the update densities of the corresponding gamma-Gaussian mixture forms for the Poisson and Dobernoli mixture parts, a cost matrix and allocation matrix for the global hypothesis are constructed. After obtaining the cost and allocation matrices, the Murty algorithm is used to generate k optimal global hypotheses and their corresponding weights. The global hypothesis with the largest weight is selected to obtain the multi-target motion state estimate. Existing global hypotheses are then pruned and merged. This invention, with the assistance of RCS information, can effectively reduce multi-target OSPA and GOSPA errors, improve multi-target estimation accuracy, and thus enhance multi-target tracking performance. Attached Figure Description

[0080] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0081] Figure 1 This is a schematic diagram of the multi-target tracking method in a clutter environment assisted by RCS information provided in an embodiment of the present invention;

[0082] Figure 2 A diagram showing the actual motion trajectory of a target in a cluttered scene;

[0083] Figure 3 This is a diagram showing the multi-target state estimation results of the RCS information-assisted multi-target tracking method provided by this invention in a cluttered environment.

[0084] Figure 4 These are experimental results on target potential estimation and OSPA error in clutter scenarios;

[0085] Figure 5 These are experimental results of target GOSPA, position error, missed detection error, and false alarm error in cluttered scenarios.

[0086] Figure 6 The results are experimental results of the root mean square error of the tracks of surviving targets in cluttered environments;

[0087] Figure 7 These are experimental results of target track position error, track missed detection error, h-track false alarm error, and target swapping error in clutter scenarios.

[0088] Figure 8 This is a schematic diagram of the architecture of a multi-target tracking device in a clutter environment assisted by RCS information, provided in an embodiment of the present invention. Detailed Implementation

[0089] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0090] In related technologies, existing data association methods, such as probabilistic data interconnection methods, suffer from reduced accuracy when the number of measurements is large. While multi-hypothesis tracking methods can alleviate data association errors, the number of hypotheses grows exponentially in cluttered environments, significantly reducing computational efficiency and resulting in poor real-time performance. In recent years, multi-target tracking algorithms based on random finite sets have emerged, eliminating the data association step and offering high computational efficiency. Therefore, it is worthwhile to fully explore multi-target tracking algorithms based on random finite sets.

[0091] As a standard output of modern radar systems, target RCS information can serve as an additional information source, enhancing the ability to distinguish targets from clutter. Existing RCS-assisted multi-target tracking methods can be categorized into four types based on whether the average RCS (SNR, amplitude) is known and whether the average RCS changes. Most RCS-assisted multi-target tracking methods are based on the assumption that the average RCS is known and constant. However, in practice, the average RCS information of a target is often unknown and time-varying. Therefore, an RCS-assisted strategy can be introduced into the multi-target tracking filter to extract the different characteristics of the target and clutter, construct the distribution function of the target and clutter RCS, and derive and implement the RCS-assisted multi-target tracking filter and its method within the framework of random finite sets. The following is a detailed description of the embodiments of this invention.

[0092] Example 1

[0093] See Figure 1 This invention provides a multi-target tracking method assisted by RCS information in a cluttered environment, comprising the following steps:

[0094] S1. Construct a Bayesian RCS information recursive estimation mechanism. The Bayesian RCS information recursive estimation mechanism uses a gamma distribution to represent the RCS correlation distribution, defines the probability density function of the inverse RCS parameter a as following a gamma distribution, and uses the conjugate of the gamma distribution and the Bayesian criterion to derive the prediction update expression of the target RCS posterior density.

[0095] S2. The Bayesian RCS information recursive estimation mechanism is introduced into the Poisson-Dobernoli mixture filter, and the target state vector is augmented. The prediction part of the Gamma-Gaussian mixture implementation process of the Poisson-Dobernoli mixture filter assisted by RCS information is derived. The multi-target posterior density is expressed as a Gamma-Gaussian mixture form. The prediction density of the corresponding Gamma-Gaussian mixture form is derived for the Poisson part and the Dobernoli mixture part respectively.

[0096] S3. Derive the update part of the Gamma-Gaussian mixture implementation process of the Poisson-Dobernouri mixture filter assisted by RCS information, express the multi-target posterior density in the form of Gamma-Gaussian mixture, and derive the update density of the corresponding Gamma-Gaussian mixture form for the Poisson part and the Dobernouri mixture part respectively. The Poisson part represents the targets that have not yet been detected, and the Dobernouri mixture part represents the missed targets, existing targets and newly generated targets.

[0097] S4. Combining the update density of the Gamma-Gaussian mixture form corresponding to the Poisson part and the multi-Bernoulli mixture part, construct the cost matrix and allocation matrix of the global hypothesis;

[0098] S5. After obtaining the cost matrix and allocation matrix, use the Murty algorithm to generate k optimal global hypotheses and their corresponding weights. Select the global hypothesis with the largest weight to obtain the multi-target motion state estimate.

[0099] S6. Perform pruning and merging operations on existing global assumptions.

[0100] In this embodiment, during step S1, in the process of constructing the Bayesian RCS information recursive estimation mechanism, the inverse RCS parameter a is defined as a random variable following a gamma distribution. It is assumed that the initial distribution parameters of a are known, i.e., α0 = β0 = 0; then the probability density function of a is:

[0101]

[0102] The evolutionary iterative formula for a is:

[0103]

[0104] In the formula, W j J(·) is an exponentially independent and identically distributed random variable with parameter 1 / d; J(·) is a Poisson distributed random variable with parameter a / d; α0 is the shape parameter; β0 is the scale parameter c; and Γ(·) is the ordinary gamma function.

[0105] Assuming the RCS observation of the target at time k-1 follows a gamma distribution, the expression is:

[0106]

[0107] In the formula, b is the shape parameter, a is the inverse RCS parameter that follows a gamma distribution as defined above, and the value of b is chosen so that the probability density function covers all Swerling models; σ t The RCS measurement value for the target.

[0108] Based on the above, a recursive estimation mechanism for target RCS information can be established. Assume p k-1 (a|σ t If the distribution follows a gamma distribution, then the predicted distribution can be obtained according to Bayes' theorem:

[0109]

[0110] In the formula, a is the inverse RCS parameter, σ t p is the RCS measurement value. k|k-1 (a|a') is the state transition density function of the inverse RCS parameter a;

[0111] Using a moment matching strategy, the prediction distribution is approximated as a gamma distribution, i.e.:

[0112]

[0113]

[0114] In the formula, d is a parameter in the distribution that iteratively generates a. Let be the shape parameter of the gamma distribution at time k; Let α' be the scale parameter of the gamma distribution at time k; β' be the shape parameter of the gamma distribution at time k-1; and β' be the scale parameter of the gamma distribution at time k-1. Similarly, using Bayes' theorem, the updated distribution follows a gamma distribution, i.e.:

[0115]

[0116]

[0117] In the formula, τ is the detection threshold, which is used to calculate the inverse RCS parameter a and the minimum mean square error estimate of the local average RCS.

[0118] In this embodiment, in step S2, the expression for augmenting the target state vector is:

[0119] x k =[p k,x ,v k,x ,p k,y ,v k,y ,a k ] T

[0120] In the formula, location information Represents position coordinates, v x,y Represents velocity; the measurement vector is represented as In the formula, This indicates the measurement value representing the target's location information; Represents the measured value of the target's RCS;

[0121] Assume the intensity function of the Poisson process at time k-1 is a gamma-Gaussian mixture, expressed as:

[0122]

[0123] The predicted Poisson process intensity is also in the form of a gamma-Gaussian mixture, expressed as:

[0124]

[0125] In the formula, J u,k|k-1 =J u,k-1 +|v b |, J u,k|k-1J represents the number of Gaussian components predicted for the undetected target at time k. u,k-1 v represents the number of Gaussian components of the undetected target at time k-1. b Indicates the target number of new students. p represents the weight value predicted for the i-th Gaussian component of the undetected target at time k. s This represents the probability of the target's existence. This represents the weight value of the i-th Gaussian component of the target that was not detected at time k-1. This represents the shape parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the shape parameter of the i-th Gaussian component of the target that was not detected at time k-1. This represents the scale parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the scale parameter of the i-th Gaussian component of the target that was not detected at time k-1. This represents the mean of the predictions for the i-th Gaussian component of the undetected target at time k. Let represent the mean of the i-th Gaussian component of the undetected target at time k-1, and F represent the target's state transition matrix. This represents the covariance matrix predicted by the i-th Gaussian component of the undetected target at time k. Q represents the covariance matrix of the i-th Gaussian component of the undetected target at time k-1. k-1 This represents the process noise matrix at time k-1.

[0126] In this embodiment, for the multi-Bernoulli mixture, if the Bernoulli density of the i-th target under the h-th global assumption at time k-1 is a gamma-Gaussian form, that is:

[0127]

[0128] The corresponding prediction density is:

[0129]

[0130]

[0131]

[0132] In the formula, Let represent the weight value of the predicted i-th target under the h-th global hypothesis at time k. Let represent the weight value of the i-th objective under the h-th global assumption at time k-1. This represents the probability of the existence of the i-th target prediction under the h-th global hypothesis at time k. This represents the probability of the existence of the i-th target under the h-th global hypothesis at time k-1. This represents the shape parameter predicted for the i-th target under the h-th global assumption at time k. This represents the shape parameter of the i-th target under the h-th global assumption at time k-1. Let represent the scale parameter for the prediction of the i-th target under the h-th global hypothesis at time k. Let represent the scale parameter of the i-th target under the h-th global assumption at time k-1. Let represent the mean of the predictions for the i-th target under the h-th global hypothesis at time k. Let represent the mean of the i-th objective under the h-th global assumption at time k-1. Let represent the covariance matrix of the prediction of the i-th target under the h-th global assumption at time k. Let represent the covariance matrix of the i-th target under the h-th global assumption at time k-1.

[0133] In this embodiment, in step S3, the updated density of the corresponding gamma-Gaussian mixture form is derived from the Poisson part, and the updated intensity function is expressed as:

[0134]

[0135] In the formula, J u,k =J u,k|k-1 , J u,k J represents the number of Gaussian components of the undetected target at time k. u,k|k-1 This represents the number of Gaussian components predicted for undetected targets at time k-1. This represents the updated weight value of the i-th Gaussian component of the target that was not detected at time k. Let represent the weight value predicted for the i-th Gaussian component of the undetected target at time k, and τ represent the threshold value of the inverse RCS parameter. This represents the shape parameter updated by the i-th Gaussian component of the undetected target at time k. This represents the shape parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the scale parameter updated for the i-th Gaussian component of the undetected target at time k. This represents the scale parameter predicted by the i-th Gaussian component of the undetected target at time k. This represents the mean of the update of the i-th Gaussian component of the undetected target at time k. This represents the mean of the predictions for the i-th Gaussian component of the undetected target at time k. This represents the covariance matrix updated by the i-th Gaussian component of the undetected target at time k. Let represent the covariance matrix of the prediction of the i-th Gaussian component of the undetected target at time k.

[0136] Specifically, the update density of the corresponding gamma-Gaussian mixture form is derived for the multi-Bernoulli mixture part, considering the update of the Bernoulli term for three types of objectives:

[0137] (I) Update the missed detection status of the i-th target under the h-th global hypothesis:

[0138]

[0139]

[0140]

[0141] In the formula, Let represent the updated weight value of the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the probability of the i-th target being updated under the h-th global hypothesis in the case of a missed detection at time k. This represents the shape parameter updated for the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the scale parameter updated for the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the mean update value of the i-th target under the h-th global hypothesis in the case of missed detection at time k. Let represent the covariance matrix updated for the i-th target under the h-th global hypothesis in the case of missed detection at time k;

[0142] (II) Update the past potential existence of the i-th target under the h-th global assumption of measurement z:

[0143]

[0144]

[0145]

[0146] In the formula, This represents the updated weight value of the i-th target under the h-th global hypothesis, assuming that time k exists. This represents the probability of the i-th target being updated under the h-th global hypothesis given that the target exists at time k. This represents the shape parameter updated for the i-th target under the h-th global assumption, assuming the existence of time k. This represents the scale parameter updated for the i-th target under the h-th global hypothesis, assuming the existence of time k. Let H represent the mean update of the i-th target under the h-th global hypothesis given that time k exists, where H represents the target's measurement matrix, K represents the target's Kalman gain, and I represents the identity matrix. Let represent the covariance matrix updated by the i-th target under the h-th global assumption given the existence of time k;

[0147] (III) Target update for the first detection of measurement z:

[0148]

[0149]

[0150] In the formula, e k This represents the target's weight value during the first detection at time k. This represents the weight value of the i-th Gaussian component in the case of the first detection at time k. λ represents the weight value in the case of the first detection at time k. c Indicates clutter intensity. This represents the probability of the target's existence under the first detection at time k. This represents the shape parameter of the i-th Gaussian component under the first detection at time k. Let represent the scale parameter of the i-th Gaussian component in the case of the first detection at time k. Let represent the mean of the i-th Gaussian component under the first detection at time k. R represents the covariance matrix of the i-th Gaussian component under the first detection at time k. k This represents the measurement noise matrix of the target at time k.

[0151] In this embodiment, in step S4, based on the update densities of different types of targets derived in the above steps, the cost matrix of the global hypothesis is constructed as follows:

[0152]

[0153]

[0154] In the formula, η represents the weight of different types of objectives; F o F represents the weight matrix of previously detected targets. p This represents the weight matrix of the first detected target.

[0155] The allocation matrix S is an m×(m+n) matrix containing either 0 or 1 entries. o ) type matrix.

[0156] In this embodiment, after obtaining the cost matrix and allocation matrix in step S5, the Murty algorithm is used to minimize tr(S T C) Generate k optimal global hypotheses and their corresponding weights, and select the global hypothesis with the largest weight. The expression is:

[0157]

[0158] In the formula, This represents the total weight value under the h-th global hypothesis at time k. This represents the weight value of the i-th objective under the h-th global assumption at time k;

[0159] Based on the obtained optimal global assumptions, the corresponding multi-objective state estimates are extracted.

[0160] In this embodiment, during step S6, when performing the trimming and merging operation on existing global hypotheses, global hypotheses with weights below a set threshold are trimmed, and similar global hypotheses are merged.

[0161] The technical effects of the present invention will be further described below through experiments:

[0162] Consider a multi-target tracking scenario with cluttered backgrounds and intersecting targets. Assume the monitoring area is [-1000m, 1000m] × [-1000m, 1000m], the target motion model is near-uniform motion, and the state transition matrix and process noise are defined as follows:

[0163]

[0164] Where, σ Q =4, T=1.

[0165] The observation matrix and observation noise are as follows:

[0166]

[0167] Where σ R =5.

[0168] The Optimal Sub-Pattern Assignment (OSPA) criterion can be used to measure multi-target tracking performance, as it can evaluate both potential estimation error and target position error. Furthermore, the recently proposed Generalized Optimal Sub-Pattern Assignment (GOSPA) error can also measure the errors of missed and false alarm targets, and has therefore been incorporated into simulation evaluation.

[0169] Assume there are four targets in the simulation scenario, with their birth and death times t and t respectively. birth =[3,10,1,1],t death=[70,81,41,60], the initial state of the new target is set as follows: In addition, the initial inverse RCS parameters of the four targets are set as follows: In the initial stage of the simulation, the four targets are relatively dispersed and gradually approach each other. At time k=40, the four targets meet, then target 4 disappears, while the remaining targets continue to exist, and targets 1 and 2 change their direction of motion. The actual trajectory of the targets and the multi-target state estimation results of the RCSI-PMBM algorithm under a Monte Carlo simulation experiment are as follows. Figure 2 and Figure 3 As shown.

[0170] See Figure 4 The graph shows a comparison of the potential estimation and OSPA error results after 100 Monte Carlo simulations. To better illustrate the superiority of the proposed scheme, RCSI-CPHD, RCSI-CBMeMBer, and PMBM filters were compared with the proposed filter RCSI-PMBM. Figure 4 It can be seen that among the filters that incorporate RCS information, the RCSI-PMBM filter has the most accurate potential estimation and the lowest OSPA error. The PMBM filter has the largest OSPA tracking error, which causes false alarms due to the influence of clutter environment, leading to overestimation of potential.

[0171] In addition, GOSPA error, position error, missed detection error, and false alarm error, such as Figure 5 As shown, RCSI-PMBM still exhibits the smallest GOSPA error, with even lower errors in both false alarm and missed detection scenarios. Figure 4 Similarly, the PMBM filter exhibits the worst performance, prone to generating numerous false alarms due to clutter, resulting in the largest error under false alarm conditions. Except for the birth / death moments, all four filters show a sudden increase in GOSPA error at time k=40. This is partly due to the increased likelihood of missed target errors when targets approach each other, and partly because target 4 disappears at time k=41, further increasing the false alarm error. However, the introduction of RCS adds another dimension of information besides position, which can help improve multi-target tracking performance.

[0172] PMBM filters implicitly maintain trajectory continuity, thus target tracks can be extracted by incorporating unique label elements into the state space. Therefore, embodiments of the present invention can extract tracks from both PMBM and RCSI-PMBM filters. To analyze the track-level errors of the two filters in estimating surviving targets, the root mean square error of track-level multi-target tracking and its decomposition error are as follows: Figure 6 As shown. By Figure 6It can be seen that the root mean square trajectory error of the RCSI-PMBM filter is consistently lower than that of the PMBM filter. Since the RCSI-PMBM filter introduces additional RCS information, it can better distinguish between targets and clutter, thus reducing the error. Furthermore, the errors of both filters increase when the target disappears, specifically at times k = 41, 60, and 71.

[0173] in, Figure 7 The simulation displays track position error, track missed target error, track false alarm target error, and target exchange error. It can be seen that the RCSI-PMBM filter has a significant advantage across all metrics. Furthermore, for both PMBM and RCSI-PMBM filters, track position error increases and decreases with target creation and disappearance. At target creation, the track missed target error is larger. However, the RCSI-PMBM filter has a lower error compared to the PMBM filter. Similarly, when the target disappears, the track false alarm target error increases. Additionally, at the beginning of the simulation, due to clutter background, multiple tracks are likely to be established, resulting in a larger target exchange error. Introducing RCS information, utilizing the difference in RCS information distribution between the target and clutter, can effectively mitigate the target exchange error.

[0174] In summary, analysis shows that the introduction of RCS information can effectively help distinguish between targets, clutter, and nearby intersecting targets, thus improving the accuracy of multi-target tracking. Specifically, a Bayesian RCS information recursive estimation mechanism is constructed. This mechanism uses a gamma distribution to represent the RCS correlation distribution, defines the probability density function of the inverse RCS parameter 'a' as following a gamma distribution, and derives the prediction update expression for the target RCS posterior density using the conjugate of the gamma distribution and the Bayesian criterion. This Bayesian RCS information recursive estimation mechanism is then introduced into a Poisson-Dobernoli hybrid filter, and the target state vector is augmented. The prediction part of the gamma-Gaussian mixture implementation process of the Poisson-Dobernoli hybrid filter assisted by RCS information is derived, expressing the multi-target posterior density in a gamma-Gaussian mixture form. The corresponding prediction densities in the gamma-Gaussian mixture form are derived for both the Poisson and Dobernoli mixture parts. The derivation of the RCS information-assisted... The update part of the Poisson-Dobernoli mixture filter's gamma-Gaussian mixture implementation process expresses the multi-target posterior density as a gamma-Gaussian mixture form. The update density of the corresponding gamma-Gaussian mixture form is derived for both the Poisson and Dobernoli mixture parts. The Poisson part represents targets not yet detected, while the Dobernoli mixture part represents missed targets, existing targets, and newly detected targets. Combining the update densities of the corresponding gamma-Gaussian mixture forms from the Poisson and Dobernoli mixture parts, a cost matrix and allocation matrix for the global hypothesis are constructed. After obtaining the cost and allocation matrices, the Murty algorithm is used to generate k optimal global hypotheses and their corresponding weights. The global hypothesis with the largest weight is selected to obtain the multi-target motion state estimate. Existing global hypotheses are then pruned and merged. This invention, with the assistance of RCS information, can effectively reduce multi-target OSPA and GOSPA errors, improve multi-target estimation accuracy, and thus improve multi-target tracking performance.

[0175] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0176] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0177] Example 2

[0178] See Figure 8 Embodiment 2 of the present invention provides a multi-target tracking device assisted by RCS information in a clutter environment, employing a multi-target tracking method assisted by RCS information in a clutter environment as described in the above embodiment, comprising:

[0179] The recursive estimation mechanism construction module 1 is used to construct a Bayesian RCS information recursive estimation mechanism. The Bayesian RCS information recursive estimation mechanism uses a gamma distribution to represent the RCS correlation distribution, defines the probability density function of the inverse RCS parameter a as following a gamma distribution, and uses the conjugate of the gamma distribution and the Bayesian criterion to derive the prediction update expression of the target RCS posterior density.

[0180] The prediction analysis module 2 is used to introduce the Bayesian RCS information recursive estimation mechanism into the Poisson-Dobernoli mixture filter, augment the target state vector, derive the prediction part of the RCS information-assisted Poisson-Dobernoli mixture filter in the gamma-Gaussian mixture implementation process, express the multi-target posterior density in the gamma-Gaussian mixture form, and derive the corresponding gamma-Gaussian mixture form prediction density for the Poisson part and the Dobernoli mixture part respectively.

[0181] Update analysis module 3 is used to derive the update part in the gamma-Gaussian mixture implementation process of the Poisson-Dobernouri mixture filter assisted by RCS information. It expresses the posterior density of multiple targets as a gamma-Gaussian mixture form, and derives the update density of the corresponding gamma-Gaussian mixture form for the Poisson part and the Dobernouri mixture part respectively. The Poisson part represents targets that have not yet been detected, and the Dobernouri mixture part represents missed targets, existing targets, and newly generated targets.

[0182] The global hypothesis processing module 4 is used to combine the update density of the gamma-Gaussian mixture form corresponding to the Poisson part and the multi-Bernoulli mixture part to construct the cost matrix and allocation matrix of the global hypothesis.

[0183] The motion state estimation module 5 is used to generate k optimal global hypotheses and their corresponding weights using the Murty algorithm after obtaining the cost matrix and allocation matrix. The global hypothesis with the largest weight is selected to obtain the motion state estimation of the multi-target target.

[0184] The global hypothesis post-processing module 6 is used to trim and merge existing global hypotheses.

[0185] It should be noted that the information interaction and execution process between the modules of the above-mentioned device are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0186] Example 3

[0187] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a multi-target tracking method in an RCS-assisted clutter environment. The program code includes instructions for executing the multi-target tracking method in an RCS-assisted clutter environment of Embodiment 1 or any possible implementation thereof.

[0188] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives, SSDs).

[0189] Example 4

[0190] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0191] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute the RCS information-assisted multi-target tracking method in clutter environments according to Embodiment 1 or any possible implementation thereof by calling the program instructions.

[0192] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0193] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0194] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0195] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A multi-target tracking method assisted by RCS information in a cluttered environment, characterized in that, include: A Bayesian RCS information recursive estimation mechanism is constructed, which uses a gamma distribution to represent the RCS correlation distribution and defines the inverse RCS parameter. The probability density function follows a gamma distribution. The prediction update expression of the target RCS posterior density is derived by using the conjugate of the gamma distribution and the Bayesian criterion. The Bayesian RCS information recursive estimation mechanism is introduced into the Poisson-Dobernoli mixture filter, and the target state vector is augmented. The prediction part of the Gamma-Gaussian mixture implementation process of the Poisson-Dobernoli mixture filter assisted by RCS information is derived. The multi-target posterior density is expressed as a Gamma-Gaussian mixture form. The prediction density of the corresponding Gamma-Gaussian mixture form is derived for the Poisson part and the Dobernoli mixture part respectively. The update part of the Poisson-Dobernoli mixture filter with RCS information assistance is derived in the process of gamma-Gaussian mixture. The posterior density of the multi-target is expressed as a gamma-Gaussian mixture. The update density of the corresponding gamma-Gaussian mixture is derived for the Poisson part and the Dobernoli mixture part respectively. The Poisson part represents the target that has not yet been detected, and the Dobernoli mixture part represents the missed target, the existing target and the newly generated target. By combining the update density of the Gamma-Gaussian mixture form corresponding to the Poisson part and the multi-Bernoulli mixture part, the cost matrix and assignment matrix of the global hypothesis are constructed. After obtaining the cost matrix and allocation matrix, the Murty algorithm is used to generate k optimal global hypotheses and their corresponding weights. The global hypothesis with the largest weight is selected to obtain the multi-target motion state estimate. The existing global assumptions are trimmed and merged.

2. The multi-target tracking method assisted by RCS information in a cluttered environment according to claim 1, characterized in that, In constructing the Bayesian RCS information recursive estimation mechanism, the inverse RCS parameter is defined. It is a random variable that follows a gamma distribution, assuming The initial distribution parameters are known, i.e. ;but The probability density function is: The evolutionary iteration formula is: In the formula, It is an exponentially independent and identically distributed random variable with parameter . , It is a random variable that follows a Poisson distribution, with parameter . ; For shape parameters; For scale parameters; This is a standard gamma function; assumed The RCS observations of the target at time t follow a gamma distribution, expressed as: In the formula, For shape parameters, For the inverse RCS parameter of the gamma distribution defined above, select... The value makes the probability density function cover all Swerling models; The RCS measurement value for the target.

3. The multi-target tracking method assisted by RCS information in a cluttered environment according to claim 2, characterized in that, set up Following a gamma distribution, the predicted distribution obtained using Bayes' theorem is: In the formula, The inverse RCS parameter, Inverse RCS parameter The state transition density function; Using a moment matching strategy, the prediction distribution is approximated as a gamma distribution, i.e.: In the formula, For iterative generation A parameter in the distribution, for Shape parameters in the time-varying gamma distribution; for Scale parameters in the time-varying gamma distribution; for Shape parameters in the time-varying gamma distribution; for The scaling parameter in the time-major gamma distribution; using Bayes' theorem, the updated distribution follows a gamma distribution, i.e.: In the formula, To determine the detection threshold, the inverse RCS parameter is calculated. And the minimum mean square error estimate of the local average RCS.

4. The multi-target tracking method assisted by RCS information in a cluttered environment according to claim 1, characterized in that, The expression for augmenting the target state vector is: In the formula, location information , , Indicates position coordinates, , Represents velocity; the measurement vector is represented as In the formula, This indicates the measurement value representing the target's location information; The measured value of the target's RCS; assumed A Poisson process is used to characterize undetected targets at all times. The intensity function of the Poisson process is a gamma-Gaussian mixture, expressed as: The predicted Poisson process intensity is also a gamma-Gaussian mixture, expressed as: In the formula, , ; express The number of Gaussian components predicted for an undetected target at any given time. express The number of Gaussian components of undetected targets at any given time. Indicates the target number of new students. express Undetected target at any time The weight values ​​for each Gaussian component prediction. This represents the probability of the target's existence. express Undetected target at any time The weight values ​​of each Gaussian component. express Undetected target at any time Shape parameters predicted by Gaussian components express Undetected target at any time Shape parameters of each Gaussian component express Undetected target at any time The scaling parameter for Gaussian component prediction. express Undetected target at any time The scaling parameter of each Gaussian component. express Undetected target at any time The mean of the Gaussian component predictions. express Undetected target at any time The mean of the Gaussian components, The state transition matrix represents the target. express Undetected target at any time The covariance matrix predicted by each Gaussian component. express Undetected target at any time The covariance matrix of Gaussian components, express The process noise matrix at each time step; The Dobernouri mixture represents the potentially detected target, if Time of the first Under the global assumption, the first The Bernoulli density of an objective is of a gamma-gaussian form, namely: The corresponding prediction density is: In the formula, ; express Time of the first Under the global assumption, the first The weight values ​​for each target prediction. express Time of the first Under the global assumption, the first The weight values ​​of each objective. express Time of the first Under the global assumption, the first The probability of the existence of each target prediction express Time of the first Under the global assumption, the first The probability of the existence of each target express Time of the first Under the global assumption, the first The shape parameters of the target are predicted. express Time of the first Under the global assumption, the first The shape parameters of the target express Time of the first Under the global assumption, the first The scale parameter for target prediction. express Time of the first Under the global assumption, the first The scale parameters of an object. express Time of the first Under the global assumption, the first The mean of the predicted targets, express Time of the first Under the global assumption, the first The mean of the targets, express Time of the first Under the global assumption, the first The covariance matrix of the target predictions express Time of the first Under the global assumption, the first The covariance matrix of each objective.

5. The multi-target tracking method assisted by RCS information in a cluttered environment according to claim 1, characterized in that, The updated density in the corresponding gamma-Gaussian mixture form is derived from the Poisson part, and the updated intensity function is expressed as: In the formula, , ; express The number of Gaussian components of the target that has not been updated at any given time. express The number of Gaussian components predicted for an undetected target at any given time. express Undetected target at any time The updated weight values ​​of each Gaussian component. express Undetected target at any time The weight values ​​for each Gaussian component prediction. This represents the threshold value of the inverse RCS parameter. express Undetected target at any time The shape parameters are updated using Gaussian components. express Undetected target at any time Shape parameters predicted by Gaussian components express Undetected target at any time The scaling parameters updated for each Gaussian component. express Undetected target at any time The scaling parameter for Gaussian component prediction. express Undetected target at any time The mean of each Gaussian component is updated. express Undetected target at any time The mean of the Gaussian component predictions. express Undetected target at any time The covariance matrix updated by Gaussian components. express Undetected target at any time The covariance matrix predicted by each Gaussian component.

6. The multi-target tracking method assisted by RCS information in a cluttered environment according to claim 5, characterized in that, The update density of the corresponding gamma-Gaussian mixture form is derived from the multi-Bernoulli mixture part, considering the update of the Bernoulli term for three types of objectives: (I) Section Under the global assumption, the first Update on missed detections for each target: In the formula, ; express In the case of missed detection at any time, the first Under the global assumption, the first The updated weight values ​​for each target. express In the case of missed detection at any time, the first Under the global assumption, the first The probability of an updated target. express In the case of missed detection at any time, the first Under the global assumption, the first The shape parameters of each target are updated. express In the case of missed detection at any time, the first Under the global assumption, the first The scale parameter for updating each target. express In the case of missed detection at any time, the first Under the global assumption, the first The mean of each target update. express In the case of missed detection at any time, the first Under the global assumption, the first The covariance matrix updated for each objective; (II) Regarding measurement The Under the global assumption, the first Update on the existence status of each target: In the formula, , ; express When the time exists, the first Under the global assumption, the first The updated weight values ​​for each target. express When the time exists, the first Under the global assumption, the first The probability of an updated target. express When the time exists, the first Under the global assumption, the first The shape parameters of each target are updated. express When the time exists, the first Under the global assumption, the first The scale parameter for updating each target. express When the time exists, the first Under the global assumption, the first The mean of each target update. The measurement matrix representing the target. The Kalman gain of the target is represented. Represents the identity matrix. express When the time exists, the first Under the global assumption, the first The covariance matrix updated for each objective; (III) Regarding measurement Update on the first detected target: In the formula, , , ; express The weight value of the target in the case of the first detection at any given moment. express The first detection case at the moment The weight values ​​of each Gaussian component. express The weight value in the case of the first detection at time step. Indicates clutter intensity. express The probability of the target's existence during the first detection at any given moment. express The first detection case at the moment Shape parameters of each Gaussian component express The first detection case at the moment The scaling parameter of each Gaussian component. express The first detection case at the moment The mean of the Gaussian components, express The first detection case at the moment The covariance matrix of Gaussian components, express Measurement noise matrix of the target at any given time.

7. The multi-target tracking method assisted by RCS information in a cluttered environment according to claim 1, characterized in that, Construct the first Cost matrix of global hypotheses for: In the formula, Weights for different types of objectives. This represents the weight matrix of previously detected targets. This represents the weight matrix for the first detected target; Allocation matrix It is a list containing either 0 or 1 items. Type matrix.

8. The multi-target tracking method assisted by RCS information in a cluttered environment according to claim 1, characterized in that, After obtaining the cost matrix and assignment matrix, the Murty algorithm is used to generate k optimal global hypotheses and their corresponding weights. The global hypothesis with the largest weight is then selected. The expression is: In the formula, express Time of the first The total weight value under each global assumption express Time of the first Under the global assumption, the first The weight values ​​of each objective; Based on the obtained optimal global assumptions, the corresponding multi-objective state estimates are extracted. .

9. A multi-target tracking method assisted by RCS information in a cluttered environment according to claim 1, characterized in that, During the pruning and merging of existing global hypotheses, global hypotheses with weights below a set threshold are pruned, and similar global hypotheses are merged.

10. A multi-target tracking device in a clutter environment assisted by RCS information, employing the multi-target tracking method in a clutter environment assisted by RCS information as described in any one of claims 1 to 9, characterized in that, include: The recursive estimation mechanism construction module is used to construct a Bayesian RCS information recursive estimation mechanism. This mechanism uses a gamma distribution to represent the RCS correlation distribution and defines the inverse RCS parameter. The probability density function follows a gamma distribution. The prediction update expression of the target RCS posterior density is derived by using the conjugate of the gamma distribution and the Bayesian criterion. The predictive analysis module is used to introduce the Bayesian RCS information recursive estimation mechanism into the Poisson-Dobernoli mixture filter, augment the target state vector, derive the prediction part of the Gamma-Gaussian mixture implementation process of the Poisson-Dobernoli mixture filter assisted by RCS information, express the multi-target posterior density in the form of Gamma-Gaussian mixture, and derive the corresponding prediction density of the Gamma-Gaussian mixture form for the Poisson part and the Dobernoli mixture part respectively. The update analysis module is used to derive the update part in the gamma-Gaussian mixture implementation process of the Poisson-Dobernouri mixture filter assisted by RCS information. It expresses the posterior density of multiple targets as a gamma-Gaussian mixture form, and derives the update density of the corresponding gamma-Gaussian mixture form for the Poisson part and the Dobernouri mixture part respectively. The Poisson part represents targets that have not yet been detected, and the Dobernouri mixture part represents missed targets, existing targets, and newly generated targets. The global hypothesis processing module is used to combine the update density of the gamma-Gaussian mixture form corresponding to the Poisson part and the multi-Bernoulli mixture part to construct the cost matrix and assignment matrix of the global hypothesis. The motion state estimation module is used to generate k optimal global hypotheses and their corresponding weights using the Murty algorithm after obtaining the cost matrix and allocation matrix. The global hypothesis with the largest weight is selected to obtain the motion state estimation of the multi-target target. The global hypothesis post-processing module is used to trim and merge existing global hypotheses.

Images