A variable structure multi-model target tracking method based on directed graph switching

Abstract

A variable structure multiple model target tracking method based on directed graph switching. Step 1: a variable structure multiple model filtering framework is established based on a ground moving target; Step 2: an adaptive strategy is carried out based on a model set in the variable structure multiple model filtering framework of step 1; Step 3: the variable structure multiple model filtering framework of step 1 is combined with the model set adaptive strategy of step 2, so that the variable structure multiple model target tracking method is obtained; Step 4: the variable structure multiple model target tracking method of step 3 is used to realize strong maneuvering target tracking. The application is used to solve the problem of strong maneuvering target tracking.

Description

Technical Field

[0001] This invention belongs to the field of target tracking, specifically relating to a variable structure multi-model target tracking method based on directed graph switching. Background Technology

[0002] Ground-based moving targets exhibit high maneuver intensity and diverse maneuver patterns, posing significant challenges to tracking algorithms. For fixed-structure multi-model methods, employing more models to cover all possible maneuver patterns leads to increased computational load and model competition, resulting in decreased accuracy. To address this issue, variable-structure multi-model methods are commonly used for tracking maneuvering targets. The main idea is to use several model groups to form a target motion model set, with only one model group running at any given time. Tracking of different maneuver patterns is achieved through two key steps: model set adaptation and adaptive estimation based on the model set. Summary of the Invention

[0003] This invention provides a variable structure multi-model target tracking method based on directed graph switching to solve the problem of tracking highly maneuverable targets. A complete model set is established according to the motion form of ground moving targets, and an adaptive switching strategy for the model set is designed based on the edge model probability. The variable structure multi-model filtering framework is integrated to obtain a variable structure multi-model method based on directed graph switching with less model competition and fewer unset threshold parameters.

[0004] This invention provides an electronic device.

[0005] This invention provides a computer-readable storage medium.

[0006] This invention is achieved through the following technical solution:

[0007] A variable structure multi-model target tracking method based on directed graph switching, the variable structure multi-model target tracking method includes the following steps:

[0008] Step 1: Establish a variable structure multi-model filtering framework based on ground moving targets;

[0009] Step 2: Implement an adaptive strategy based on the model set within the variable structure multi-model filtering framework of Step 1;

[0010] Step 3: Combine the variable structure multi-model filtering framework from Step 1 with the model set adaptive strategy from Step 2 to obtain the variable structure multi-model target tracking method;

[0011] Step 4: Implement high-maneuverability target tracking using the variable structure multi-model target tracking method from Step 3.

[0012] A variable structure multi-model target tracking method based on directed graph switching, wherein step 1 of establishing the variable structure multi-model filtering framework specifically includes the following steps:

[0013] Step 1.1: Mode Mixing;

[0014] Step 1.2: State interaction;

[0015] Step 1.3: State recursion;

[0016] Step 1.4: Status Update;

[0017] Step 1.5: Fusion.

[0018] A variable structure multi-model target tracking method based on directed graph switching, wherein step 1 of establishing the variable structure multi-model filtering framework specifically includes the following steps:

[0019] Step 1.1: Complete the evolution of the model probability from time k-1 to time k; assuming the model set is Θ, then have

[0020]

[0021] In the formula, m k-1 and m k The motion patterns at time k-1 and time k are respectively; Given the measurement sequence up to time k-1, P(m) k =j|m k-1 =i,y k-1 ) represents the model transition probability from model i to model j;

[0022] Step 1.2: Generate the initial state probability density; based on conditional probability and total probability theory, have

[0023]

[0024] Step 1.3: Propagate the state probability density from time k-1 to time k to obtain the prior distribution at time k; according to the state transition equation, have

[0025] p(x k |m k =j,y k-1 )=∫p(x k |x k-1 ,m k =j,y k-1 )p(x k-1 |m k =j,y k-1 )dx k-1 (3);

[0026] Step 1.4: Obtain the likelihood equation, correct prior information, and obtain the posterior state distribution at time k. Define the likelihood function p(y). k |x k According to Bayes' criterion, the posterior distribution can be obtained as follows:

[0027] p(x k ,m k =j|y k )∝P(m k =j|y k-1 )p(x k |m k =j,y k-1 )p(y k |x k (4);

[0028] Step 1.5: Complete the calculation of the model probability at time k; according to Bayes' criterion

[0029]

[0030] Combining the formulas and applying the law of total probability, we obtain the form of the posterior probability density function of the state.

[0031]

[0032] A variable structure multi-model target tracking method based on directed graph switching, wherein the adaptive strategy for the model set in step 2 specifically includes the following steps:

[0033] Step 2.1: Run the IMM algorithm;

[0034] Let the current model set be Θ. k The running model set is Θ k The IMM algorithm;

[0035] Step 2.2: Model set activation;

[0036] Set activation criteria for the model set. If the activation condition is met, activate a new model subset Θ that better matches the target motion pattern. n Together with the original model set, they form a new model set Θ. a , where Θ a =Θ n ∪Θ k If not satisfied, then retain the original model set Θ. k ;

[0037] Step 2.3: Termination of model set;

[0038] Set a model set termination criterion. If the termination condition is met, terminate the subset of models Θ that does not match the target motion pattern. n Or Θk If the termination condition is not met, the original model set will be retained.

[0039] Step 2.4: Let k = k + 1, and return to step 2.1.

[0040] A variable structure multi-model target tracking method based on directed graph switching, wherein the adaptive strategy for the model set in step 2 is as follows:

[0041] Let M denote the model; Θ denote the model set; let S denote the set of all models, called the complete model set; the complete model set S consists of M1, M2, ..., M5, and the real-time model set at the current time is Θ[M3] = {M2, M3, M4}, where M3 is the central model; define the model M at time k. j Model probability In the form of

[0042]

[0043] Calculate the probability of each model in the model set Θ[M3], when the marginal model probability satisfies the following condition within the memory interval:

[0044]

[0045] In the formula, T = {kd, k-d+1, ..., k} is the memory interval, and d is the memory depth;

[0046] If the target moves frequently, then d takes a larger value; if the target moves infrequently, then a smaller value is taken.

[0047] In addition, the model probability must also satisfy

[0048]

[0049] When the model set is switched to M j The model set Θ[M] centered on the model j ];

[0050] Where μ th The preset probability threshold must satisfy μ. th >1 / N, where N is the number of models in Θ[M3]; μ th The larger the value of μ, the lower the model set switching frequency. th The smaller the value, the higher the switching frequency.

[0051] A variable structure multi-model target tracking method based on directed graph switching, wherein step 3 of the variable structure multi-model target tracking method specifically includes the following steps:

[0052] Step 3.1: Mode blending;

[0053] Step 3.2: State interaction;

[0054] Step 3.3: State recursion;

[0055] Step 3.4: Model probability update;

[0056] Step 3.5: Interactive output;

[0057] Step 3.6: Adaptive switching of model sets.

[0058] The aforementioned variable structure multi-model target tracking method based on directed graph switching, specifically step 3.1, involves:

[0059] Define the transition probabilities between models in the model set.

[0060] π ij =P(m) k =j|m k-1 =i,y k-1 ),i,j∈Θ (10)

[0061] definition Let be the probability of model i at time k-1, i.e.

[0062]

[0063] The probabilities of each mode can then be obtained according to the formula.

[0064]

[0065] Step 3.2 specifically involves defining the hybrid weights.

[0066]

[0067] Substituting into the equation yields the estimated mixed state value.

[0068]

[0069] In the formula, E[·] represents the expectation operation, x k Represents the target state vector. This represents the state estimate based on model j at time k-1;

[0070] definition Let the state error covariance based on model j at time k-1 be the state error covariance. Then, calculate the mixed state error covariance.

[0071]

[0072] Specifically, step 3.3 involves each sub-filter performing state recursion based on the state equation.

[0073]

[0074] In the formula, Let i be the state transition matrix of model i; Let be the process noise matrix of model i.

[0075] A variable structure multi-model target tracking method based on directed graph switching, wherein step 3.4 specifically comprises:

[0076] After parallel filtering is completed, the probabilities of each model are updated based on the measurement residuals and residual covariance information output by the sub-filters.

[0077]

[0078] In the formula, The likelihood function value of the model

[0079]

[0080] Step 3.5 specifically involves,

[0081] Using the new model probability State estimates of the output of each sub-filter and state error covariance The global state estimate is calculated using a weighted summation method. Covariance P k|k ;

[0082] Global state estimate

[0083]

[0084] Global state error covariance P k|k

[0085]

[0086] Step 3.6 specifically involves using the updated model probabilities obtained from the formula to determine whether the model set switching conditions are met; if...

[0087]

[0088] If both conditions are met, then switch to the real-time model set Θ[M] j Otherwise, keep the current real-time model set.

[0089] An electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0090] Memory, used to store computer programs;

[0091] When a processor executes a program stored in memory, it implements the steps of the method described above.

[0092] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0093] The beneficial effects of this invention are:

[0094] This invention constructs a directed graph of model sets with clear transformation relationships, and directly switches between model sets based on the probability of edge models, without the need to activate and terminate model sets separately. This maintains the number of model sets available for computation. Therefore, compared with interactive multi-model algorithms with the same number of complete model sets, it can effectively reduce the amount of computation while achieving the same tracking accuracy. Attached Figure Description

[0095] Figure 1 This is a schematic diagram of the switching of the directed graph model set in this invention.

[0096] Figure 2 This is a flowchart of the method of the present invention.

[0097] Figure 3 This is a schematic diagram of the movement curve of the ground moving target according to the present invention.

[0098] Figure 4 This is a schematic diagram of the ground moving target position curve of the present invention.

[0099] Figure 5 This is a schematic diagram of the ground moving target speed curve of the present invention.

[0100] Figure 6 This is a schematic diagram of the position estimation error curve of the present invention.

[0101] Figure 7 This is a schematic diagram of the velocity estimation error curve of the present invention. Detailed Implementation

[0102] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0103] A variable structure multi-model target tracking method based on directed graph switching, the variable structure multi-model target tracking method includes the following steps:

[0104] Step 1: Establish a variable structure multi-model filtering framework based on ground moving targets;

[0105] Step 2: Implement an adaptive strategy based on the model set within the variable structure multi-model filtering framework of Step 1;

[0106] Step 3: Combine the variable structure multi-model filtering framework from Step 1 with the model set adaptive strategy from Step 2 to obtain the variable structure multi-model target tracking method;

[0107] Step 4: Implement high-maneuverability target tracking using the variable structure multi-model target tracking method from Step 3.

[0108] A variable structure multi-model target tracking method based on directed graph switching, wherein step 1 of establishing the variable structure multi-model filtering framework specifically includes the following steps:

[0109] Step 1.1: Mode Mixing;

[0110] Step 1.2: State interaction;

[0111] Step 1.3: State recursion;

[0112] Step 1.4: Status Update;

[0113] Step 1.5: Fusion.

[0114] A variable structure multi-model target tracking method based on directed graph switching, wherein step 1 of establishing the variable structure multi-model filtering framework specifically includes the following steps:

[0115] Step 1.1: Complete the evolution of the model probability from time k-1 to time k; assuming the model set is Θ, then have

[0116]

[0117] In the formula, m k-1 and m k The motion patterns at time k-1 and time k are respectively; Given the measurement sequence up to time k-1, P(m)k =j|m k-1 =i,y k-1 ) represents the model transition probability from model i to model j;

[0118] Step 1.2: Generate the initial state probability density; based on conditional probability and total probability theory, have

[0119]

[0120] Step 1.3: Propagate the state probability density from time k-1 to time k to obtain the prior distribution at time k; according to the state transition equation, have

[0121] p(x k |m k =j,y k-1 )=∫p(x k |x k-1 ,m k =j,y k-1 )p(x k-1 |m k =j,y k-1 )dx k-1 (3);

[0122] Step 1.4: Obtain the likelihood equation, correct prior information, and obtain the posterior state distribution at time k. Define the likelihood function p(y). k |x k According to Bayes' criterion, the posterior distribution can be obtained as follows:

[0123] p(x k ,m k =j|y k )∝P(m k =j|y k-1 )p(x k |m k =j,y k-1 )p(y k |x k (4);

[0124] Step 1.5: Complete the calculation of the model probability at time k; according to Bayes' criterion

[0125]

[0126] Combining the formulas and applying the law of total probability, we obtain the form of the posterior probability density function of the state.

[0127]

[0128] A variable-structure multi-model target tracking method based on directed graph switching is proposed. This method employs a variable-structure multi-model filtering framework based on the IMM algorithm, adding an adaptive step to the model set. It uses a real-time variable model set containing combinations of different models. At each filtering time step, the model set is optimized according to a specific switching criterion, and then the state estimate is obtained based on the optimized best model set. The variable-structure multi-model filtering comprises two parts: the IMM algorithm and a model set adaptive strategy. The adaptive strategy for the model set in step 2 specifically includes the following steps:

[0129] Step 2.1: Run the IMM algorithm;

[0130] Let the current model set be Θ. k The running model set is Θ k The IMM algorithm;

[0131] Step 2.2: Model set activation;

[0132] The purpose of model set activation is to increase the number of models that may have better tracking accuracy;

[0133] Set activation criteria for the model set. If the activation condition is met, activate a new model subset Θ that better matches the target motion pattern. n Together with the original model set, they form a new model set Θ. a , where Θ a =Θ n ∪Θ k If not satisfied, then retain the original model set Θ. k ;

[0134] Step 2.3: Termination of model set;

[0135] The purpose of terminating the model set is to remove models with poor tracking accuracy.

[0136] Set a model set termination criterion. If the termination condition is met, terminate the subset of models Θ that does not match the target motion pattern. n Or Θ k If the termination condition is not met, the original model set will be retained.

[0137] Step 2.4: Let k = k + 1, and return to step 2.1.

[0138] A variable-structure multi-model target tracking method based on directed graph switching, wherein the adaptive strategy for the model set in step 2 is as follows: the model set adaptive strategy is the core of the variable-structure multi-model algorithm. Current mainstream adaptive strategies based on model set probability and model set likelihood ratio mainly suffer from the following problems:

[0139] Issue 1: Too many threshold parameters to be set. Model set adaptation requires presetting two model set activation threshold parameters and two model set termination threshold parameters.

[0140] Problem 2: The adaptive strategy is complex. It requires separate model set activation and termination operations, which cause changes in the number of models in the model set, and the algorithm's time consumption also changes accordingly, which is not conducive to practical engineering applications.

[0141] To avoid the problems of too many threshold parameters and complex adaptive strategies, this paper adopts a combined approach of merging model set activation and termination operations. Specifically, a directed graph of model sets with explicit transformation relationships is constructed, and model set switching is directly completed based on the probability of marginal models, eliminating the need for separate activation and termination. Furthermore, a model set with a fixed number of models is used, transforming the model set adaptation process into a continuous process of selecting a fixed-size model set from a complete model set, thus increasing the number of models without increasing computational cost. To avoid erroneous model switching, a model switching criterion with memory depth is designed.

[0142] A variable structure multi-model target tracking method based on directed graph switching, wherein the adaptive strategy for the model set in step 2 is as follows:

[0143] Let M denote the model; Θ denote the model set; let S denote the set of all models, called the complete model set; the complete model set S consists of M1, M2, ..., M5, and the real-time model set at the current time is Θ[M3] = {M2, M3, M4}, where M3 is the central model; define the model M at time k. j Model probability In the form of

[0144]

[0145] Calculate the probability of each model in the model set Θ[M3], when the marginal model probability satisfies the following condition within the memory interval:

[0146]

[0147] In the formula, T = {kd, k-d+1, ..., k} is the memory interval, and d is the memory depth;

[0148] If the target moves frequently, then d takes a larger value; if the target moves infrequently, then a smaller value is taken.

[0149] In addition, the model probability must also satisfy

[0150]

[0151] When the model set is switched to M j The model set Θ[M] centered on the modelj ];

[0152] Where μ th The preset probability threshold must satisfy μ. th >1 / N, where N is the number of models in Θ[M3]; μ th The larger the value of μ, the lower the model set switching frequency. th The smaller the value, the higher the switching frequency.

[0153] Taking the current model set Θ[M3] as an example, when the model probability When the model set is switched to Θ[M2]={M1,M2,M3}, and the model probability is... When switching, the model set is changed to Θ[M4] = {M3, M4, M5}; otherwise, Θ[M3] remains unchanged. It can be seen that the directed graph variable structure multi-model algorithm can maintain a constant number of models during switching, ensuring that the computation time does not increase with the tracking time and meeting the requirements for online computation. Furthermore, by switching the model set, the model that best matches the target motion is included in the tracking model, reducing the risk of losing the target.

[0154] A variable structure multi-model target tracking method based on directed graph switching, wherein step 3 of the variable structure multi-model target tracking method specifically includes the following steps:

[0155] Step 3.1: Mode blending;

[0156] Step 3.2: State interaction;

[0157] Step 3.3: State recursion;

[0158] Step 3.4: Model probability update;

[0159] Step 3.5: Interactive output;

[0160] Step 3.6: Adaptive switching of model sets.

[0161] The aforementioned variable structure multi-model target tracking method based on directed graph switching, specifically step 3.1, involves:

[0162] Define the transition probabilities between models in the model set.

[0163] π ij =P(m) k =j|m k-1 =i,y k-1 ),i,j∈Θ (10)

[0164] definition Let be the probability of model i at time k-1, i.e.

[0165]

[0166] The probabilities of each mode can then be obtained according to the formula.

[0167]

[0168] Step 3.2 specifically involves defining the hybrid weights.

[0169]

[0170] Substituting into the equation yields the estimated mixed state value.

[0171]

[0172] In the formula, E[·] represents the expectation operation, x k Represents the target state vector. This represents the state estimate based on model j at time k-1;

[0173] definition Let the state error covariance based on model j at time k-1 be the state error covariance. Then, calculate the mixed state error covariance.

[0174]

[0175] Specifically, step 3.3 involves each sub-filter performing state recursion based on the state equation.

[0176]

[0177] In the formula, Let i be the state transition matrix of model i; Let be the process noise matrix of model i.

[0178] A variable structure multi-model target tracking method based on directed graph switching, wherein step 3.4 specifically comprises:

[0179] After parallel filtering is completed, the probabilities of each model are updated based on the measurement residuals and residual covariance information output by the sub-filters.

[0180]

[0181] In the formula, The likelihood function value of the model

[0182]

[0183] Step 3.5 specifically involves,

[0184] Using the new model probability State estimates of the output of each sub-filter and state error covariance The global state estimate is calculated using a weighted summation method. Covariance P k|k ;

[0185] Global state estimate

[0186]

[0187] Global state error covariance P k|k

[0188]

[0189] Step 3.6 specifically involves using the updated model probabilities obtained from the formula to determine whether the model set switching conditions are met; if...

[0190]

[0191] If both conditions are met, then switch to the real-time model set Θ[M] j Otherwise, keep the current real-time model set.

[0192] An electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0193] Memory, used to store computer programs;

[0194] When a processor executes a program stored in memory, it implements any of the steps described above.

[0195] A computer-readable storage medium storing a computer program that, when executed by a processor, implements any of the steps described above.

[0196] Two drones equipped with passive sensors are used to track a maneuvering ground target. The simulation scenario is as follows: Figure 3 As shown.

[0197] The motion position and velocity curves of the ground moving target are as follows: Figure 4 and Figure 5 As shown.

[0198] The target was located using an interactive multi-model and a directed graph variable structure multi-model, respectively. The interactive multi-model and the directed graph variable structure multi-model adopted a constant velocity model, a "current" statistical model, and a cooperative turning model, respectively. The specific parameters are shown in Table 1.

[0199] Table 1 Model set parameters

[0200]

[0201]

[0202] Model transition probability matrix between multiple models

[0203] Other simulation-related parameters are shown in Table 2.

[0204] Table 2 Simulation-related parameters

[0205]

[0206] The tracking filter selects six motion quantities of the maneuvering target in the ground system: position, velocity, and acceleration, as the state variables of the tracking system, with X = [xv]. x a x yv y a y ] T .

[0207] Based on the above simulation scenario and parameters, the proposed directed graph-based variable structure multi-model filtering algorithm was applied using capacitive Kalman filtering to track moving targets on the ground. One hundred independent Monte Carlo shooting experiments were conducted, and the tracking results are as follows: Figures 6 to 7 As shown.

[0208] The root mean square error of 100 experiments was calculated, and the results are shown below:

[0209] Table 3 Root Mean Square Error

[0210]

[0211] It can be seen that, since the variable structure multi-model can change the model set selection and parameters at any time according to the real-time situation, it can quickly and adaptively adjust when the target is maneuvering, and has higher tracking accuracy than the interactive multi-model.

Claims

1. A variable structure multiple model target tracking method based on directed graph switching, characterized in that, The variable-structure multi-model target tracking method comprises the following steps: Step 1: establishing a variable-structure multi-model filtering framework based on a ground moving target; Step 2: performing an adaptive strategy based on a model set in the variable-structure multi-model filtering framework of step 1; Step 3: combining the variable-structure multi-model filtering framework of step 1 with the model set adaptive strategy of step 2 to obtain a variable-structure multi-model target tracking method; Step 4: implementing target strong maneuvering by using the variable-structure multi-model target tracking method of step 3; The step 1 of establishing the variable-structure multi-model filtering framework comprises the following steps: Step 1.1: mode mixing; Step 1.2: state interaction; Step 1.3: state recursion; Step 1.4: state updating; Step 1.5: fusion; The step 1 of establishing the variable-structure multi-model filtering framework comprises the following steps: Step 1.1: Evolution of model probabilities from time k-1 to time k when the model set is completed; assuming the model set is Θ, then There are In the formula, m k-1 and m k The motion patterns at time k-1 and time k are respectively; Given the measurement sequence up to time k-1, P(m) k =j|m k-1 =i,y k-1 ) represents the model transition probability from model i to model j; Step 1.2: Generate initial state probability density; from conditional probabilities and law of total probability, There are Step 1.3: Propagate the state probability density from time k-1 to time k to obtain the prior distribution at time k; according to the state transition equation, There are p(x k |m k = j, y k-1 ) = ∫ p(x k |x k-1 , m k = j, y k-1 ) p(x k-1 |m k = j, y k-1 ) dx k-1 (3); Step 1.4: Obtain the likelihood equation, correct the prior information, and obtain the state posterior distribution at time k Define the likelihood function p(y k |x k ); according to the Bayes rule, the posterior distribution is in the form of p(x k |m k ) = p(x k |y k )P(m k-1 |y k ) (3) k k-1 k k ) (4)​​​ Step 1.5: completing calculation of model probabilities at the k moment; according to the Bayesian criterion Combining the formulas (1.4) and (1.5) and according to the total probability criterion, a state posterior probability density function form is obtained 2. The variable structure multiple model target tracking method based on directed graph switching according to claim 1, characterized in that, The step 2 of performing the adaptive strategy based on the model set comprises the following steps: Step 2.1: running the IMM algorithm; Let Θ be the current model set k , and Θ k be the IMM algorithm with running model set Step 2.2: model set activation; A model set activation criterion is set, if the activation condition is met, a new model subset Θ that matches the target motion pattern better is activated n , and the original model set is replaced by the new model set Θ a . Wherein Θ a = Θ n ∪ Θ k ; if not, the original model set Θ k is maintained Step 2.3: model set termination; setting a termination criterion for the model set, and if the termination criterion is met, terminating the model subset Θ that does not match the target motion pattern n or Θ k keeping the original model set; and if the termination criterion is not met, keeping the original model set. Step 2.4: setting k=k+1 and returning to step 2.

1.

3. The variable structure multiple model target tracking method based on directed graph switching according to claim 2, characterized in that, The step 2 of performing the adaptive strategy based on the model set is specifically: M denotes a model; Θ denotes a model set; S denotes a set of all models, referred to as a complete model set; a complete model set S consisting of M1, M2, M3, M4, and M5, a real-time model set at a current time is Θ[M3] = {M2, M3, M4}, wherein M3 is a central model; a model M at k time is defined as j a model probability in a form of Calculating model probabilities in the model set Θ[M3], when the edge model probabilities in the memory interval all satisfy In the formula, T={k-d, k-d+1,..., k} is a memory interval, and d is a memory depth; If the target frequently maneuvers, a large value is taken for d; if the target does not frequently maneuver, a small value is taken; In addition, the model probabilities also need to satisfy At this time, the model set is switched to M j The model set Θ[M j ] is the model set for the center model. wherein μ th is a preset probability threshold, satisfying μ th > 1 / N, N being the number of models in Θ[M3]; the larger μ th is, the lower the model set switching frequency is, and the smaller μ th is, the higher the switching frequency is.

4. The variable structure multiple model target tracking method based on directed graph switching according to claim 1, characterized in that, The step 3 of the variable-structure multi-model target tracking method comprises the following steps: Step 3.1: mode mixing; Step 3.2: state interaction; Step 3.3: state recursion; Step 3.4: model probability updating; Step 3.5: interactive output; Step 3.6: model set adaptive switching.

5. The variable structure multiple model target tracking method based on directed graph switching according to claim 4, characterized in that, The step 3.1 is specifically, Defining transition probabilities between models in the model set π ij = P(m k = j | m k-1 = i, y k-1 ), i, j e Q (10) Definitions is the probability of model i at time k-1, i.e. Then, the mode probabilities can be obtained according to the formula (1.1) Said step 3.2 is in particular defining a mixing weight Substituting (1.2) into (1.1) gives the mixed state estimate where E[·] denotes the expectation operation, x k denotes the target state vector, denotes the state estimate based on model j at time k-1; Definitions is the state error covariance based on model j for time k-1, then the hybrid state error covariance is computed as The step 3.3 is specifically, wherein is the state transition matrix for model i; is the process noise matrix for model i.

6. The variable structure multiple model target tracking method based on directed graph switching according to claim 4, characterized in that, Each sub-filter completes state recursion based on a state equation The step 3.4 is specifically, In the formula, is the likelihood function value for the model After completing parallel filtering, according to information such as measurement residuals and residual covariance output by the sub-filter, updating of the model probabilities is completed, that is, Utilizing the new model probabilities state estimates from each sub-filter and state error covariances Computing the global state estimate and covariance P k|k ; global state estimate Global state error covariance P k|k The step 3.5 is specifically, If both are satisfied, switch to real-time model set Θ[M j ]; otherwise, keep the current real-time model set.

7. An electronic device, comprising: The step 3.6 is specifically, The updated model probabilities obtained from the formula (1.17) are used to determine whether the model set switching condition is satisfied; if The computer device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus; 8. A computer-readable storage medium, characterized in that, The memory is used for storing a computer program; The processor is used for executing the program stored on the memory to implement the method steps of any one of claims 1-6. The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the method steps of any one of claims 1-6.

Images