A multi-target tracking method in non-Gaussian noise environment

Abstract

This invention provides a multi-target tracking method in a non-Gaussian noise environment. A spatial coordinate system is established to obtain the sensor's position coordinates and monitoring area. The sensor periodically acquires a measurement set, including the number of targets and the associated measurement set for each target. Based on a target state-measurement model, the associated measurement set is iteratively updated using a Dirichlet process-hidden Markov chain hybrid model to determine the target state's posterior probability density. Based on the iterative updates of each frame, the track of each target is output. This invention significantly reduces tracking error and improves track estimation accuracy and track integrity. By adaptively and in real time jointly estimating the probability density distribution of the target state and observation noise through variational inference, tracking error is significantly reduced and track estimation accuracy and track integrity are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of target tracking, and in particular to a multi-target tracking method in a non-Gaussian noise environment. Background Art

[0002] With the development of technology, manned or unmanned aircraft are more and more active in the air and underwater, so the monitoring of aircraft is particularly important.

[0003] When tracking and locating targets such as unmanned underwater vehicles, the signal received by the receiving end in the observation channel is subject to multipath interference due to the influence of factors such as the unevenness of the transmission medium and reflections from objects and interfaces, resulting in multimodal statistical characteristics of the sensor node's received measurements. At the same time, the random time-varying and space-varying characteristics of the observation channel will seriously affect the communication link. As well as the presence of clutter or false targets in the transmission medium, missed detections and false alarms often occur.

[0004] Standard random finite set algorithms assume a priori knowledge of sensor noise statistics and that they follow a Gaussian distribution. However, in practice, it is often difficult to accurately model sensor noise. Failure to account for this non-Gaussian measurement error in real systems can lead to model mismatch in multi-target tracking methods, resulting in degraded tracking performance and even completely incorrect target state estimates.

[0005] Therefore, accurately tracking and locating multiple targets in a non-Gaussian noise environment has become a technical problem that needs to be solved urgently. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, the present invention provides a multi-target tracking method in a non-Gaussian noise environment. A spatial coordinate system is established to obtain the position coordinates of the sensor and the monitoring area. The sensor periodically obtains a measurement set, which is a set of the position of the real target and the false target position caused by clutter. The set of measurement information in the i-th frame is recorded as Z i The number of targets and the associated measurement sets for each target are obtained. Based on the target state-measurement model, the associated measurement sets are used to iteratively update the target state posterior probability density using a Dirichlet process-hidden Markov chain hybrid model. Based on the iterative updates for each frame, the track of each target is output. This invention utilizes a generalized label multi-Bernoulli filter and a Dirichlet process-hidden Markov chain hybrid model to significantly reduce tracking error and improve track estimation accuracy and track integrity.

[0007] Aiming at the problems of track jump, high track missed detection rate and low tracking accuracy in the traditional standard random finite set algorithm under non-Gaussian noise environment, the present invention provides a multi-target tracking method in non-Gaussian noise environment to improve the accuracy and robustness of multi-target tracking.

[0008] The technical solution adopted by the present invention to solve the technical problem includes the following steps:

[0009] Step 1: Establish a spatial coordinate system to obtain the sensor's location coordinates and monitoring area;

[0010] Step 2: The sensor periodically acquires a measurement set, which is a set of the true target position and the false target position caused by clutter. The set of measurement information at time i is recorded as Z i ;

[0011] Step 3: Based on the measurement set obtained by the sensor, obtain the number of targets and the associated measurement set of each target through generalized label multi-Bernoulli filtering;

[0012] Step 4: Based on the target state-measurement model, a Dirichlet process-hidden Markov chain hybrid model is established. Based on the associated measurement set, the posterior probability density of the target state is iteratively updated using the variational inference solution method.

[0013] Step 5: Based on the iterative update of the posterior probability density of each target state, the estimated track of each target is output.

[0014] Furthermore, in step 4, the steps for solving the Dirichlet process-hidden Markov chain hybrid model and variational inference are:

[0015] The construction process of the Dirichlet process-hidden Markov chain hybrid model is as follows:

[0016] 1: At time t, the measurement information periodically obtained by the sensor The measurement information is the real target position information Position information of false targets caused by clutter Right now i is the time index, t represents the time t; the number of real targets N and the real target state set in the monitoring area are obtained through generalized label multi-Bernoulli filtering. and with Related measurements Where X is x j is the state of the jth target, Y is y j is the jth target x j The corresponding measurement, j is the target label index, x j,i is the state of the jth target at time i, y j,i is the measurement of the jth target at time i;

[0017] For the jth target The corresponding measurement The target state-measurement model is as follows:

[0018] x j,t =Fx j,t-1 +w j,t

[0019] y j,t =Hx j,t +ξ j,t

[0020] w j,t ~N(0,Q)

[0021]

[0022] F is the state transfer matrix, H is the measurement matrix; N(·) represents Gaussian distribution, w j,t is the state noise of the j-th target at time t, which obeys the zero-mean Gaussian distribution, and Q is the variance of the state noise; j,t is the non-Gaussian distribution measurement noise of the jth target at time t, which is fitted by the mixed Gaussian model GMM; K is the number of Gaussian components in the mixed Gaussian model, e is the Gaussian component index, and w e is the amplitude of the e-th Gaussian component, satisfying (μ ξe ,Σ ξe ) are the distribution parameters of the e-th Gaussian component, which are the mean and variance respectively; x j,t-1 is the state of the jth target at time t-1, x j,t is the state of the jth target at time t, y j,t is the measurement corresponding to the jth target at time t;

[0023] Based on the target state-measurement model, a Dirichlet process-hidden Markov chain hybrid model is established, and the broken stick model is used to construct a mathematical form of discrete distribution that obeys the Dirichlet process. The specific mathematical construction process is as follows:

[0024] V k ~Beta(1,α)

[0025]

[0026] α is the concentration parameter, which indicates the degree of discreteness of the generated distribution and is a scalar greater than 0. Beta(·) is the Beta distribution, k is the primary index of the number of cutoffs, and V k is the intermediate parameter of the kth truncation of the Beta distribution with parameters (1,α), and r is the secondary index of the number of truncation. V r is the intermediate parameter of the rth truncation of the Beta distribution with parameters (1,α). k is the weight coefficient of the kth truncation, satisfying s j,t is the jth target state x at time t j,t The indicator factor; λ is the base distribution parameter, G(λ) is the base distribution; θ k is the basis obtained by the k-th sampling, represents the infinite number of bases obtained by sampling; Σ is the covariance matrix of the Gaussian distribution obeyed by the measurement.

[0027] Furthermore, in step 4, the variational inference solution of the Dirichlet process-hidden Markov chain hybrid model is specifically as follows:

[0028] The posterior probability density p(x j,t ,s j,t ,V,θ|y j,1:t ):

[0029]

[0030] V is the intermediate parameter V of infinite truncation k A collection of y j,1:t is the measurement set of the jth target in the 1:t time period, y j,1:t-1 is the measurement set of the jth target in the time period 1:t-1,

[0031] Through mean field theory, the variational distribution family is used to replace the true posteriori at time t and t-1:

[0032]

[0033] where q p (x j,t )=∫q(x j,t-1 )p(x j,t |x j,t-1 )dx j,t-1 , s j,t-1 is the jth target state x at time t-1 j,t-1 indicator factor.

[0034] By means of conjugate prior, (x j,t ,s j,t ,V,θ) sets the prior distribution at time t:

[0035]

[0036] is the jth target x at time t j,t The mean and variance of the conjugate Gaussian prior, is the j-th target x at time t-1j,t The mean and variance of the conjugate Gaussian prior, m j,t for The one-step forecast mean, Σ j,t for The one-step prediction variance of Mult(·) is a multinomial distribution. is the parameter of the multinomial distribution at time t, is the parameter of the k-th truncated multinomial distribution at time t; (u k ,v k ) is the parameter of the beta distribution; is the parameter of Gaussian distribution; since it needs to be updated iteratively over time, (u k ,v k )and Define as follows: is the parameter of the kth truncated beta distribution at time t-1, is the parameter of the kth truncated beta distribution at time t, is the parameter of the kth truncated Gaussian distribution at time t-1, is the parameter of the kth truncated Gaussian distribution at time t.

[0037] According to (x j,t ,s j,t ,V,θ) and its prior distribution are subjected to variational inference to obtain the recursive formula of each prior distribution parameter at time t:

[0038]

[0039] Furthermore, in step 5, the estimated track of each target is output based on the iterative update of the posterior probability density of each target state. The specific steps are:

[0040] For multi-target associated track sets: and Using the variational inference solution, at time 1:t, x is iteratively updated by the variational inference solution. j,i The posterior probability of each target is finally obtained Based on the estimated track set obtained above, the estimated track of each target as it changes over time can be presented through visualization operations.

[0041] An electronic device comprises one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method described above.

[0042] A computer-readable storage medium stores program code, which can be called by a processor to execute the method described above.

[0043] The beneficial effect of the present invention lies in that standard random finite set algorithms fail to account for non-Gaussian measurement errors, which can cause model mismatch in multi-target tracking methods, leading to degraded tracking performance and even completely erroneous target state estimates. By utilizing a generalized label multi-Bernoulli filter and a Dirichlet process-hidden Markov chain hybrid model, the probability density distribution of the target state and observation noise is adaptively estimated in real time through variational inference, significantly reducing tracking error and improving track estimation accuracy and track integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of an embodiment of the present invention;

[0045] Figure 2 is a flow chart of a specific embodiment of the present invention;

[0046] Figure 3 Schematic diagram of the actual tracks of multiple targets within the monitoring area in the simulation example of the present invention;

[0047] Figure 4 Schematic diagram of the probability distribution function of non-Gaussian measurement noise in a simulation example of the present invention;

[0048] Figure 5 Schematic diagram showing the comparison of multi-target measurement in the x-direction and trajectory estimation using the DP-HMM-JointGLMB and JointGLMB methods in a simulation example of the present invention;

[0049] Figure 6 Schematic diagram showing the comparison of multi-target measurement in the y direction and trajectory estimation using the DP-HMM-JointGLMB and JointGLMB methods in the simulation example of the present invention;

[0050] Figure 7 Schematic diagram comparing OSPA distance error, OSPA positioning error, and OSPA potential error using the DP-HMM-JointGLMB and JointGLMB methods in the simulation example of the present invention;

[0051] Figure 8 Schematic diagram comparing the number of multi-target estimates using the DP-HMM-JointGLMB and JointGLMB methods in the simulation example of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described below with reference to the accompanying drawings and examples.

[0053] like Figures 1-8 As shown, the present invention provides a multi-target tracking method in a non-Gaussian noise environment, comprising the steps of:

[0054] S1: Scene setting: Establish a two-dimensional or three-dimensional space coordinate system. Take the two-dimensional space coordinate system xoy as an example to obtain the position coordinates of the sensor (x s ,y s ) and monitoring area x∈[r x1 ,r x2 ],y∈[r y1 ,r y2 ];

[0055] S2: The sensor periodically obtains a measurement set. The set of measurement information in the i-th frame is denoted as Z i ; For the measurement information periodically acquired by the sensor within the 1:t time period The measurement information is the real target position information Position information of false targets caused by clutter Right now

[0056] S3: Based on the measurement set obtained by the sensor, the number of targets N and the associated measurement set of each target are obtained through the generalized labeled multi-Bernoulli filter (BNVo, BTVo, and HGHoang, "An efficient implementation of the generalized labeled multi-Bernoulli filter," IEEE Transactions on Signal Processing, vol. 65, no. 8, pp. 1975–1987, Apr. 2017.): target state set and its associated measurements Where X is x j is the state of a certain target, Y is y j For the target x j Corresponding measurements.

[0057] S4: Based on the target state-measurement model, a Dirichlet process-hidden Markov chain hybrid model is established, and based on the associated measurement set, the posterior probability density of the target state is iteratively updated using the variational inference solution method;

[0058] S4.1: For the jth target The corresponding measurement The target state-measurement model is as follows:

[0059] xj,t =Fx j,t-1 +w j,t

[0060] y j,t =Hx j,t +ξ j,t

[0061] w j,t ~N(0,Q)

[0062]

[0063] F is the state transfer matrix, H is the measurement matrix; N(·) represents Gaussian distribution, w j,t is the state noise of the j-th target at time t, which obeys the zero-mean Gaussian distribution, and Q is the variance of the state noise; j,t is the measurement noise of the jth target non-Gaussian distribution at time t, which is fitted by the mixed Gaussian model GMM; K is the number of Gaussian components in the mixed Gaussian model, e is the Gaussian component index, and we is the amplitude of the eth Gaussian component, satisfying (μ ξe ,Σ ξe ) are the distribution parameters of the e-th Gaussian component, which are the mean and variance respectively; x j,t-1 is the state of the jth target at time t-1, x j,t is the state of the jth target at time t, y j,t is the measurement corresponding to the jth target at time t. Based on the target state-measurement model, a Dirichlet process-hidden Markov chain hybrid model is established, and the broken stick model is used to construct the mathematical form of the discrete distribution that obeys the Dirichlet process. The specific mathematical construction process is as follows:

[0064] V k ~Beta(1,α)

[0065]

[0066] α is the concentration parameter, which indicates the degree of discreteness of the generated distribution and is a scalar greater than 0. Beta(·) is the Beta distribution, k is the primary index of the number of cutoffs, and V k is the intermediate parameter of the kth truncation of the Beta distribution with parameters (1,α), and r is the secondary index of the number of truncation. V r is the intermediate parameter of the rth truncation of the Beta distribution with parameters (1,α). k is the weight coefficient of the kth truncation, satisfying s j,t is the jth target state x at time t j,t The indicator factor; λ is the base distribution parameter, G(λ) is the base distribution; θ kis the basis obtained by the k-th sampling, represents the infinite number of bases obtained by sampling; Σ is the covariance matrix of the Gaussian distribution obeyed by the measurement.

[0067] Furthermore, the variational inference solution of the Dirichlet process-hidden Markov chain hybrid model is specifically as follows:

[0068] S4.2: The solution for variational inference using the Dirichlet process-hidden Markov chain hybrid model is as follows:

[0069] The posterior probability density p(x j,t ,s j,t ,V,θ|y j,1:t ):

[0070]

[0071] V is the intermediate parameter V of infinite truncation k A collection of y j,1:t is the measurement set of the jth target in the 1:t time period, y j,1:t-1 is the measurement set of the jth target in the time period 1:t-1,

[0072] Through mean field theory, the variational distribution family is used to replace the true posteriori at time t and t-1:

[0073]

[0074] where q p (x j,t )=∫q(x j,t-1 )p(x j,t |x j,t-1 )dx j,t-1 , sj,t-1 is the indicator factor of the j-th target state xj,t-1 at time t-1.

[0075] By means of conjugate prior, (x j,t ,s j,t ,V,θ) sets the prior distribution at time t:

[0076]

[0077] is the jth target x at time t j,t The mean and variance of the conjugate Gaussian prior, is the j-th target x at time t-1 j,t The mean and variance of the conjugate Gaussian prior, m j,t for The one-step forecast mean, Σ j,t for The one-step prediction variance of Mult(·) is a multinomial distribution. is the parameter of the multinomial distribution at time t, is the parameter of the k-th truncated multinomial distribution at time t; (u k ,v k ) is the parameter of the beta distribution; is the parameter of Gaussian distribution; since it needs to be updated iteratively over time, (u k ,v k )and Define as follows: is the parameter of the kth truncated beta distribution at time t-1, is the parameter of the kth truncated beta distribution at time t, is the parameter of the kth truncated Gaussian distribution at time t-1, is the parameter of the kth truncated Gaussian distribution at time t.

[0078] According to (x j,t ,s j,t ,V,θ) and its prior distribution are subjected to variational inference to obtain the recursive formula of each prior distribution parameter at time t:

[0079]

[0080] S5: Based on the iterative update of the posterior probability density of each target state, the estimated track of each target is output:

[0081] For multi-target associated track sets: and Using the above variational inference solution, at time 1:t, x is iteratively updated by the variational inference solution. j,i The posterior probability of each target is finally obtained is the jth target x at time i j,t The mean of the conjugate Gaussian prior. Based on the estimated track set obtained above, the estimated track of each target as it changes over time can be visualized.

[0082] In order to better illustrate the technical solution of the present invention, the present invention is further described below in conjunction with simulation experiments:

[0083] The present invention proposes a multi-target tracking method for non-Gaussian measurements based on a Dirichlet process-hidden Markov chain hybrid model on the basis of generalized labeled multi-Bernoulli (GLMB) filtering.

[0084] 1. Simulation conditions

[0085] In a two-dimensional space coordinate system, consider a single sensor node with a position coordinate of (0,0) and tracking three targets within a monitoring area of ​​[-1000,1000]m×[-1000,1000]m. The sensor measurement covariance matrix is The sampling period is T = 1s, and the total tracking time is 100s. The starting time of different targets is {1, 1, 30}, and the extinction time is {70, 100, 70}. The real track is as follows Figure 3 shown.

[0086] The parameters of the generalized label multi-Bernoulli filter are set as follows, including a single target state with position l and velocity v:

[0087] C=x=[l x ,v x ,l y ,v y ];

[0088] Single target survival probability p s =0.99, detection probability p d =0.98. The target state transfer matrix F and covariance matrix Q are

[0089]

[0090] In the above formula, q is the standard deviation of the target process noise. In this example, it is taken as 0.1;

[0091] The measurement function is:

[0092]

[0093] The target generation follows the GLMB distribution, and the parameter set is in mean and the covariance matrix P B for:

[0094]

[0095] P B =diag([5,5,5,5]);

[0096] The maximum number of tracks is 1000, the maximum number of update step assumptions is 100, and the track cutoff threshold is 10 -15 , single track Gaussian component cutoff threshold 10 -3 .

[0097] Through the above GLMB filter, we can get the target estimation number N, the target state set and its associated measurements Where X is x j is the state of a certain target, Y is y j For the target x j Corresponding measurements.

[0098] For the jth target The corresponding measurement The following Dirichlet process-hidden Markov chain hybrid model is established:

[0099] V k ~Beta(1,α)

[0100]

[0101] In this example, α=2, λ=[5,20], and the base distribution G(·) is a Gaussian distribution.

[0102] Since k∈[1,+∞), It is necessary to truncate k so that k∈[1,K]; construct V through the broken stick model: k ~Beta(1,α), When k=K, if but Satisfied at this time

[0103] Solved by variational inference method:

[0104] Initialization parameters:

[0105] x j,0 is the initial value of the jth target state,

[0106]

[0107]

[0108] For multi-target associated track sets: Using the above variational inference solution, at time 1:t, x is iteratively updated by the variational inference solution. j,i The posterior probability of each target is finally obtained is the x of the jth target at time i j,tThe mean of the conjugate Gaussian prior. Based on the estimated track set obtained above, the estimated track of each target as it changes over time can be visualized.

[0109] For different methods, the error between the true track and the estimated track is measured based on the Optimal Sub-patten Assignment (OSPA), which measures the error between two sets A = {a1, a2, ..., a m} and B={b1,b2,…,b n},m,n∈{0,1,2,…}, the distance D p,c (A,B):

[0110]

[0111] d (c) (a,b)=min(c,||ab||);

[0112] Π n represents all permutations on the set {1,2,3,…,n};

[0113] OSPA distance It can be decomposed into positioning error distance and potential error distance:

[0114]

[0115] Take c = 100, p = 1 to obtain the OSPA distance between the true track and the estimated track.

[0116] 2. Simulation results analysis

[0117] The JointGLMB algorithm in the figure is a more efficient implementation algorithm proposed by (BNVo, BTVo, and HGHoang, “An efficient implementation of the generalized labeled multi-Bernoulli filter,” IEEE Transactions on Signal Processing, vol. 65, no. 8, pp. 1975–1987, Apr. 2017.) based on the GLMB algorithm.

[0118] Figure 5 and Figure 6 The estimated tracks of the proposed method DP-HMM-JointGLMB and JointGLMB algorithm in the x-direction and y-direction are given;

[0119] Figure 7 The estimated track OSPA distance, positioning error distance, and potential error distance of the proposed method DP-HMM-JointGLMB and the JointGLMB algorithm are given. As can be seen from the figure, the tracking accuracy of the DP-HMM-JointGLMB method is significantly better than that of the JointGLMB algorithm.

[0120] Figure 8 The target estimation numbers of the proposed method DP-HMM-JointGLMB and JointGLMB algorithm are given by Figure 8 It can be seen that the two methods have the same performance in estimating the number of targets.

[0121] It should be understood that the above-mentioned embodiments are merely illustrative and non-restrictive. Without departing from the basic principles of the present invention, various obvious or equivalent modifications or substitutions that can be made by those skilled in the art to the above-mentioned details will be included in the scope of the claims of the present invention.

Claims

1. A multi-target tracking method in a non-Gaussian noise environment, characterized in that The steps include: Step 1: Establish a spatial coordinate system to obtain the sensor's location coordinates and monitoring area; Step 2: The sensor periodically acquires a measurement set, which is a set of the true target position and the false target position caused by clutter. The set of measurement information at time i is recorded as Z i ; Step 3: Based on the measurement set obtained by the sensor, obtain the number of targets and the associated measurement set of each target through generalized label multi-Bernoulli filtering; Step 4: Based on the target state-measurement model, a Dirichlet process-hidden Markov chain hybrid model is established. Based on the associated measurement set, the posterior probability density of the target state is iteratively updated using the variational inference solution method. Step 5: Based on the iterative update of the posterior probability density of each target state, the estimated track of each target is output.

2. The multi-target tracking method in a non-Gaussian noise environment according to claim 1, characterized in that: In step 4, the construction process of the Dirichlet process-hidden Markov chain hybrid model is specifically as follows: 1: At time t, the measurement information periodically obtained by the sensor The measurement information is the real target position information Position information of false targets caused by clutter Right now i is the time index, t represents the time t; the number of real targets N and the real target state set in the monitoring area are obtained through generalized label multi-Bernoulli filtering. and with Related measurements Where X is x j is the state of the jth target, Y is y j is the jth target x j The corresponding measurement, j is the target label index, x j,i is the state of the jth target at time i, y j,i is the measurement of the jth target at time i; For the jth target The corresponding measurement The target state-measurement model is as follows: x j,t =Fx j,t-1 +w j,t y j,t =Hx j,t +ξ j,t w j,t ~N(0,Q) F is the state transfer matrix, H is the measurement matrix; N(·) represents Gaussian distribution, w j,t is the state noise of the j-th target at time t, which obeys the zero-mean Gaussian distribution, and Q is the variance of the state noise; j,t is the non-Gaussian distribution measurement noise of the jth target at time t, which is fitted by the mixed Gaussian model GMM; K is the number of Gaussian components in the mixed Gaussian model, e is the Gaussian component index, and w e is the amplitude of the e-th Gaussian component, satisfying (μ ξe ,Σ ξe ) are the distribution parameters of the e-th Gaussian component, which are the mean and variance respectively; x j,t-1 is the state of the jth target at time t-1, x j,t is the state of the jth target at time t, y j,t is the measurement corresponding to the jth target at time t; Based on the target state-measurement model, a Dirichlet process-hidden Markov chain hybrid model is established, and the broken stick model is used to construct a mathematical form of discrete distribution that obeys the Dirichlet process. The specific mathematical construction process is as follows: In k ~Beta(1,α) α is the concentration parameter, which indicates the degree of discreteness of the generated distribution and is a scalar greater than 0. Beta(·) is the Beta distribution, k is the primary index of the number of cutoffs, and V k is the intermediate parameter of the kth truncation of the Beta distribution with parameters (1, α), r is the secondary index of the number of truncation; V r is the intermediate parameter of the rth truncation of the Beta distribution with parameters (1,α); π k is the weight coefficient of the kth truncation, satisfying s j,t is the jth target state x at time t j,t The indicator factor; λ is the base distribution parameter, G(λ) is the base distribution; θ k is the basis obtained by the k-th sampling, represents the infinite number of bases obtained by sampling; Σ is the covariance matrix of the Gaussian distribution obeyed by the measurement.

3. The multi-target tracking method in a non-Gaussian noise environment according to claim 2, characterized in that: In step 4, the variational inference solution of the Dirichlet process-hidden Markov chain hybrid model is specifically as follows: The posterior probability density p(x j,t ,s j,t ,V,θ|y j,1:t ): V is the intermediate parameter V of infinite truncation k A collection of y j,1:t is the measurement set of the jth target in the 1:t time period, y j,1:t-1 is the measurement set of the jth target in the time period 1:t-1, Through mean field theory, the variational distribution family is used to replace the true posteriori at time t and t-1: where q p (x j,t )=∫q(x j,t-1 )p(x j,t |x j,t-1 )dx j,t-1 , s j,t-1 is the jth target state x at time t-1 j,t-1 indicator factors; By means of conjugate prior, (x j,t ,s j,t ,V,θ) sets the prior distribution at time t: is the jth target x at time t j,t The mean and variance of the conjugate Gaussian prior, is the j-th target x at time t-1 j,t The mean and variance of the conjugate Gaussian prior, m j,t for The one-step forecast mean, Σ j,t for The one-step prediction variance of Mult(·) is a multinomial distribution. is the parameter of the multinomial distribution at time t, is the parameter of the k-th truncated multinomial distribution at time t; (u k ,v k ) is the parameter of the beta distribution; is the parameter of Gaussian distribution; since it needs to be updated iteratively over time, (u k ,v k )and Define as follows: is the parameter of the kth truncated beta distribution at time t-1, is the parameter of the kth truncated beta distribution at time t, is the parameter of the kth truncated Gaussian distribution at time t-1, is the parameter of the kth truncated Gaussian distribution at time t; According to (x j,t ,s j,t ,V,θ) and its prior distribution are subjected to variational inference to obtain the recursive formula of each prior distribution parameter at time t:

4. The multi-target tracking method in a non-Gaussian noise environment according to claim 3, characterized in that: In step 5, the estimated track of each target is output based on the iterative update of the posterior probability density of each target state. The specific steps are: For multi-target associated track sets: and Using the variational inference solution, at time 1:t, x is iteratively updated using the variational inference solution. j,i The posterior probability of each target is finally obtained Based on the estimated track set obtained above, the estimated track of each target as it changes over time can be presented through visualization operations.

5. An electronic device, characterized in that: include: one or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to perform the method according to any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program code, and the program code can be called by a processor to execute the method according to any one of claims 1 to 4.

Images