Target Tracking Method and Related Devices Based on Variable Structure TS Fuzzy Semantic Model

Abstract

According to the target tracking method and related device based on the variable structure T-S fuzzy semantic model disclosed in the embodiment of the present invention, the state prediction value of the i-th fuzzy model is estimated; based on the preset model conditional filtering formula, the state estimation of the model is calculated; in combination with the observation matrix and the state estimation, the predicted observation of each model is calculated; based on the preset membership update expression, the first fuzzy membership of the predicted observation is calculated; the antecedent parameters are updated, the second fuzzy membership of the antecedent variables is calculated, and the weight of each model is calculated based on the second fuzzy membership; the target state and covariance are calculated in combination with the state estimation, the first fuzzy membership, and the weight. Through the implementation of the present invention, feedback learning is introduced into the strong tracking filtering method to realize the identification of the fuzzy model's consequent parameters, the EM algorithm and the fuzzy subspace clustering are combined to construct the fuzzy clustering objective function, the antecedent parameters are identified, the estimation performance and the accuracy of the motion state determination are improved, and the computational burden is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of target tracking, and in particular to a target tracking method based on a variable structure TS fuzzy semantic model and a related device. Background Art

[0002] Due to the randomness of the movement of maneuvering targets, a fixed single motion model is difficult to accurately reflect the maneuvering state information of the target. Multiple motion models are needed to describe the maneuvering state of the maneuvering target. Therefore, a target detection and tracking algorithm based on the TS fuzzy semantic model is proposed. However, the disadvantage of this method is that the computational burden is very large. As the language value increases, the number of fuzzy rules increases, and more motion models increase the computational burden. More importantly, more models do not necessarily improve the estimation performance, but do reduce the estimation performance due to competition with unnecessary models. In addition, different rules assume that the target's motion model follows a fixed turning rate and process noise standard deviation within a certain period of time, which cannot well reflect the true target motion state. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide a target tracking method and related devices based on a variable structure TS fuzzy semantic model, which can at least solve the problems of the target detection and tracking algorithm provided in the related art, such as high computational burden, low estimation performance, and poor accuracy of the determined motion state.

[0004] To achieve the above object, a first aspect of an embodiment of the present invention provides a target tracking method based on a variable structure TS fuzzy semantic model, the method comprising:

[0005] Combine the state transfer matrix and the target state at the k-1th moment to estimate the state prediction value of the i-th fuzzy linear model;

[0006] Calculating the state estimation of the i-th fuzzy linear model based on a preset model condition filtering formula;

[0007] Calculating the predicted observations of each fuzzy linear model in combination with the observation matrix and the state estimate;

[0008] Based on a preset membership updating expression, calculating a first fuzzy membership of the predicted observation to the fuzzy linear model;

[0009] Updating antecedent parameters, calculating a second fuzzy membership of antecedent variables, and calculating a weight of each of the fuzzy linear models based on the second fuzzy membership;

[0010] The target state and covariance at the kth moment are calculated by combining the state estimation, the first fuzzy membership, and the weight.

[0011] A second aspect of an embodiment of the present application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements each step of the method provided in the first aspect of the embodiment of the present application.

[0012] A third aspect of an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, each step of the method provided in the first aspect of the embodiment of the present application is implemented.

[0013] According to the target tracking method and related device based on the variable structure TS fuzzy semantic model provided by the embodiment of the present invention, the state prediction value of the i-th fuzzy linear model is estimated by combining the state transfer matrix and the target state at the k-1th moment; the state estimation of the i-th fuzzy linear model is calculated based on the preset model conditional filtering formula; the predicted observation of each fuzzy linear model is calculated by combining the observation matrix and the state estimation; the first fuzzy membership of the predicted observation belonging to the fuzzy linear model is calculated based on the preset membership update expression; the antecedent parameter is updated, the second fuzzy membership of the antecedent variable is calculated, and the weight of each fuzzy linear model is calculated based on the second fuzzy membership; the target state and covariance at the kth moment are calculated by combining the state estimation, the first fuzzy membership and the weight. Through the implementation of the present invention, feedback learning is introduced into the strong tracking filtering method, a robust strong tracking filtering algorithm is constructed, the identification of the TS fuzzy model post-parameters is realized, and the EM algorithm and fuzzy subspace clustering are combined to construct the fuzzy clustering objective function, the identification of the TS fuzzy model antecedent parameters is realized, the estimation performance is effectively improved, the calculation burden is reduced, and the accuracy of the motion state determination is improved.

[0014] Other features and corresponding effects of the present invention are described in the latter part of the specification, and it should be understood that at least part of the effects become obvious from the description in the specification of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 A schematic diagram of a basic flow chart of a target tracking method provided by a first embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a program module of a target tracking device provided by a second embodiment of the present invention;

[0018] Figure 3 A schematic structural diagram of an electronic device provided in a third embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0020] For the TS fuzzy semantic model for discrete dynamic systems, the discrete nonlinear dynamic system is assumed to be as follows:

[0021] x k =f(x k-1 )+e k-1 (1)

[0022] z k =h(x k )+v k (2)

[0023] Among them, x k ∈R n represents the n-dimensional state vector at time k, z k ∈R m represents an m-dimensional observation vector, and f(.) and h(.) represent appropriate nonlinear functions. k-1 It means the mean is 0 and the covariance is The process noise, v k The mean is 0 and the covariance is R vk of observation noise.

[0024] In general, the TS fuzzy model believes that any nonlinear system can be represented by the following M fuzzy linear models:

[0025] Rule i: IF is and is …and is then

[0026]

[0027] Among them, θ k represents the antecedent variable of the rule, represents the fuzzy membership function corresponding to the antecedent variable, and They represent the state transfer matrix and the observation matrix respectively. From formula (3), we can see that M fuzzy models are all linear time-invariant models. Therefore, the global fuzzy model can be expressed as follows:

[0028]

[0029]

[0030] in Represents x k The fuzzy membership degree belonging to the i-th linear model can be calculated as follows:

[0031]

[0032] in, Representation variables Belongs to model set F k i The membership degree of

[0033] Usually the fuzzy membership function The following bell-shaped membership function can be used:

[0034]

[0035] in, and They represent the mean and standard deviation of the j-th membership function of the ith rule respectively.

[0036] The disadvantage of the target tracking method based on the TS fuzzy model algorithm is that the computational burden is very large. As the language value increases, the number of fuzzy rules increases, and more motion models increase the computational burden. More importantly, more models do not necessarily improve the estimation performance, but do reduce the estimation performance due to competition with unnecessary models, and different rules assume that the target's motion model follows a fixed turning rate and process noise standard deviation within a certain period of time, which cannot well reflect the true target motion state. A good solution is rule selection, which can optimize the use of the rule base to represent system characteristics. Therefore, how to select the fuzzy set in the TS fuzzy model is the key to solving the problem.

[0037] First embodiment:

[0038] In order to solve the technical problems of the target detection and tracking algorithms provided in the related art, such as the large computational burden, low estimation performance, and poor accuracy of the determined motion state, this embodiment proposes a target tracking method based on a variable structure TS fuzzy semantic model, such as Figure 1FIG. 2 is a schematic diagram of a basic flow chart of a target tracking method provided in this embodiment. The target tracking method provided in this embodiment includes the following steps:

[0039] Step 101: Estimate the state prediction value of the i-th fuzzy linear model by combining the state transfer matrix and the target state at the k-1th moment.

[0040] Specifically, the state transfer matrix and the target state at the k-1th moment are input into a preset state prediction value calculation formula to estimate the state prediction value of the i-th fuzzy linear model.

[0041] According to the definition of formula (3), assuming that the state of each model at time k-1 is The calculation formula of the state prediction value is expressed as:

[0042]

[0043] in, represents the state transfer matrix, T represents the time interval, represents the angular velocity of the target in the i-th fuzzy linear model, represents the target state of the i-th fuzzy linear model at the k-1th moment, Represents the state prediction value of the i-th fuzzy linear model.

[0044] It should be understood that, in this embodiment, before the state prediction step, the relevant parameters may be initialized, specifically including: defining the weighted index m=2, the clustering stop threshold ε=0.00001, x0 represents the actual starting position of the target, represents the observation noise variance, and I represents the identity matrix.

[0045] Step 102: Calculate the state estimate of the i-th fuzzy linear model based on a preset model condition filtering formula.

[0046] In the interactive TS multi-model filtering algorithm, the final fusion estimate is only used as an output and is not fed back to the next moment for state estimation. However, the final fusion estimate is often more accurate than the estimates of each mode dependency, so it can be considered as a reference item for state estimation at the next moment. Based on this, this embodiment provides a TS interactive multi-model adaptive filtering algorithm that performs feedback learning on the fusion estimate.

[0047] The feedback learning term is defined based on the interactive TS multi-model filtering method, namely:

[0048]

[0049] in, and They are obtained by the final fusion at time k-1 and the mixed initial state estimation of model i through the system state transfer matrix, namely:

[0050]

[0051]

[0052] Final fusion estimate at time k-1 The state estimation at time k and the state estimation of multi-model filtering at time k are respectively:

[0053]

[0054]

[0055] in, and are the filter gains of the two estimates respectively. Let c1 be the proportion coefficient of the final fusion estimate at time k-1 to the proportion coefficient of the multi-model filter at time k. Then 1-c1 is the proportion coefficient of the multi-model filter at time k. The state estimate at time k is the sum of the state estimate of the final fusion estimate at time k-1 and the state estimate of the multi-model filter at time k, that is:

[0056]

[0057] Therefore, we can get:

[0058]

[0059] For the calculation of the consequent parameter, since this parameter is a multidimensional variable, so it cannot be calculated using weighted recursive least squares (WRLS). In the algorithm proposed in this embodiment, the parameters of each linear model defined in (3) are calculated by a recursive least squares estimator based on feedback learning terms. Considering that the feedback learning terms are used for the state estimation at the next moment, the above formula can be rewritten as:

[0060]

[0061] in, is the gain parameter of the feedback learning item. The gain parameter of the feedback learning item refers to the degree to which the feedback learning item is used for state estimation at the next moment. Generally, the more accurate the current observation information is, or the less the weight of the prior information is, The smaller the gain, the smaller the gain. To this end, the following error is defined:

[0062]

[0063]

[0064]

[0065] According to the update estimation formula, the relationship between the three errors can be obtained, namely:

[0066]

[0067] The corresponding covariance matrix is:

[0068]

[0069] where I is the dimensionally consistent identity matrix, and:

[0070]

[0071] Optimal Gain Matrix It can be obtained by minimizing the covariance matrix trace, that is:

[0072]

[0073] Optimal Gain Matrix This can be obtained by minimizing the covariance matrix trace:

[0074]

[0075] Therefore, the optimal gain matrix is:

[0076]

[0077] It should be noted that the optimal gain matrix requires the calculation of the covariance matrix, and the calculation of this matrix involves the interaction covariance matrix, so the calculation is more complicated. To avoid calculating the matrix, when the gain parameter of the feedback learning item is sufficiently small, the calculation can be ignored, so a suboptimal gain matrix can be used:

[0078]

[0079] At this time, the error covariance matrix is:

[0080]

[0081] Therefore, combined with the fading factor introduced in the improved strong tracking method, in general, the more accurate the current observation information is, or the less weight the prior information has, the smaller the fading factor λ should be, and vice versa. The model conditional filtering formula in the TS interactive multi-model filtering method can be expressed as:

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] Since the feedback learning term and its gain parameter are introduced in the update estimation, it is necessary to design a reasonable gain parameter. If the gain parameter is too large, the filtering algorithm may diverge, and if it is too small, the feedback learning term may not work. Therefore, it is necessary to select a reasonable feedback parameter. In addition, the introduction of this feedback learning term is similar to the distributed consistent filtering algorithm process. In the distributed consistent filtering algorithm, the difference in the prediction estimation of different sensors is introduced into the update estimation to improve the estimation performance, which is equivalent to the difference between the final fusion estimation and the different mode-dependent estimation in the multi-model adaptive filtering algorithm. The difference is introduced into the update estimation to improve the estimation performance.

[0092] Step 103: Combine the observation matrix and the state estimation to calculate the predicted observation of each fuzzy linear model.

[0093] Specifically, the observation matrix and the state estimate are input into a preset prediction observation calculation formula to calculate the prediction observation of each fuzzy linear model;

[0094] The prediction observation calculation formula is expressed as:

[0095]

[0096] in, represents the predicted observation based on the i-th fuzzy linear model at time k, represents the observation matrix, Represents the state estimate.

[0097] Step 104: Calculate the first fuzzy membership of the predicted observation belonging to the fuzzy linear model based on a preset membership update expression.

[0098] Parameter identification is very important for modeling TS fuzzy model. In this embodiment, Gaussian mixture model (GMM) is used to fit its parameters using expectation maximization (EM) algorithm. In the process of building TS fuzzy semantic model, EM algorithm is used to realize the identification of antecedent parameters of TS fuzzy semantic model. The membership function of antecedent variables obeys Gaussian distribution. EM algorithm is a general method for maximum likelihood estimation of model parameters. The parameter model likelihood function is constructed according to the idea of ​​mixed Gaussian EM as follows:

[0099]

[0100] According to the gradient descent method, the updated mean and standard deviation of the antecedent membership function are obtained:

[0101]

[0102]

[0103] Identification of antecedent parameters and non-negative fuzzy parameters about, therefore, The value of is the key issue to be solved in the next step. This embodiment uses fuzzy rule antecedent learning based on fuzzy subspace clustering to solve this problem. Assume is an observation set, is a set of predicted observations, z k,l Indicates l th Observe, while Indicates that k is based on fuzzy model i th The weighted entropy term is introduced into the fuzzy subspace clustering algorithm to balance the membership and transform the high-dimensional data into the relevant subspace for clustering. k It contains rich information that reflects the target motion trend in real time, so combined with the spatial constraint information θ k , the objective function of the entropy adjustment method is defined as follows:

[0104]

[0105] Among them, λ k is the Lagrange multiplier vector, β is a constant, represents the weight of feature m in model i, is k time l th Observation belongs to i th The fuzzy membership of the model satisfies represents the metric function between observation l and the observation predicted by model i, which is expressed as follows:

[0106]

[0107] Given target state The observation z k,l Likelihood function.

[0108] According to the objective function Find the partial derivative and get the degree of membership The update expression is:

[0109]

[0110] in, Indicates that the predicted observation belongs to the first fuzzy membership of the fuzzy linear model, β represents a constant, represents the weight of feature m in the i-th fuzzy linear model, represents the weight of feature m in the qth fuzzy linear model, Represents spatial constraint information, represents the metric function between observation l and the predicted observation in the i-th fuzzy linear model, represents the metric function between observation l and the predicted observation in the qth fuzzy linear model.

[0111] In an optional implementation of this embodiment, in order to improve the accuracy of rule selection for identifying the TS model, this embodiment, based on the traditional fuzzy set rule selection, introduces a local multi-granularity rough set method in which the upper and lower approximations are only related to the equivalence classes induced by samples in the target concept. The lower approximation must be included in the target concept, so the local multi-granularity rough set model has stronger robustness for model simplification in target tracking. Combined with the simplification method based on residual new information, the most accurate model at the current moment is jointly screened out. The ideal state of the entire feature simplification process is that when it is detected that the target is maneuvering, all models are working at the current moment, and rough set simplification is turned on to simplify the fuzzy rules that are most suitable for this moment. When the target is not maneuvering, the TS model uses the fuzzy set that is simplified during the maneuver, and no feature simplification is performed.

[0112] If m equivalence relations R1, R2, L are given, For any Its lower approximation is defined as:

[0113]

[0114] in, is determined by the relationship R i The equivalence class of induced x; yes The conditional probability with respect to X, α is a constraint parameter.

[0115]

[0116] The non-empty finite set consisting of three fuzzy sets is taken as the domain U. In the TS fuzzy model, the non-empty finite set A representing the attribute is the heading angle error The three language values ​​of Negative Large (NL), Small (S) and Positive Large (PL), the range of attribute values ​​of attribute a∈A is V a is the value of the membership function corresponding to the heading angle error and The heading angle error of the target is used as a feature to divide each fuzzy model, and the normalized weight W of the consequent parameter in each rule is obtained as the decision attribute in the system calculation. The knowledge system is shown in the following table:

[0117] Table 1 Attributes of knowledge system

[0118]

[0119] Set the W of the activated model set to 1 and the W of the terminated model set to 0, and the resulting knowledge system is as follows:

[0120] Table 2 Attributes of knowledge system

[0121]

[0122] In neighborhood rough set data processing, data usually have differences in order of magnitude and dimension. In order to obtain accurate data processing results, the original data needs to be normalized before data processing. The commonly used normalization method is the maximum and minimum method, and its formula is:

[0123]

[0124] Among them, x max and x min are the maximum and minimum values ​​of the sample array respectively. After normalization, the data all fall in the interval [0,1]. After normalization of the data in the neighborhood decision system, the results shown in Table 3 are obtained:

[0125] Table 3 Attributes of knowledge system

[0126]

[0127] Assume that α=0.6, β=0.4, X={x2,x3}, X is the target fuzzy set, that is, the fuzzy set with decision attributes greater than 0. R1={NL}, R2={S}, R3={PL}. The standard deviation of the attributes NL, S and PL after normalization is calculated, and the standard deviations of NL, S and PL are Std(NL)=0.4041, Std(S)=0.2852, and Std(PL=0.5774. Among them, Std(NL) is the standard deviation of the attribute NL data, and λ is a set parameter used to adjust the neighborhood size according to the data classification accuracy. It is usually taken between 2 and 4. In the example, λ=2, and the neighborhood radius of the two attributes is δ(NL)=0.2020, δ(S)=0.1426, and δ(PL=0.2887. By calculation, we can get:

[0128]

[0129] According to the definition of local rough sets, we can get the upper and lower approximations of X with respect to R1:

[0130]

[0131] Similarly, the upper and lower approximations of X with respect to R2 and R3 are:

[0132]

[0133] Then the intersection of the lower approximation of R1, R2 and R3 is x3, which means that the equivalence class of the samples in the target fuzzy set X is x3. Therefore, fuzzy set 3 is retained and fuzzy sets 1 and 2 are simplified.

[0134] Step 105: Update the antecedent parameters, calculate the second fuzzy membership of the antecedent variables, and calculate the weight of each fuzzy linear model based on the second fuzzy membership.

[0135] Specifically, this embodiment inputs the second fuzzy membership into a preset weight calculation formula to calculate the weight of each fuzzy linear model;

[0136] The weight calculation formula is expressed as:

[0137]

[0138] in, Represents the antecedent variable Belongs to the fuzzy membership function F in the fuzzy linear model k i The second fuzzy membership degree of Represents the weight of each fuzzy linear model.

[0139] Step 106: Calculate the target state and covariance at the kth moment by combining the state estimation, the first fuzzy membership, and the weight.

[0140] Specifically, this embodiment inputs the aforementioned state estimation, the first fuzzy membership, and the weight into a preset target state calculation formula to calculate the target state at the kth moment, and inputs the state estimation, the first fuzzy membership, the weight, and the target state at the kth moment into a preset covariance calculation formula to calculate the covariance;

[0141] The target state calculation formula is expressed as:

[0142]

[0143] The covariance calculation formula is expressed as:

[0144]

[0145] in, represents the state estimate, represents the first fuzzy membership, represents the weight, represents the target state at the kth moment, represents the error covariance matrix, P k represents the covariance.

[0146] According to the target tracking method based on the variable structure TS fuzzy semantic model provided by the embodiment of the present invention, the state prediction value of the i-th fuzzy linear model is estimated by combining the state transfer matrix and the target state at the k-1th moment; the state estimation of the i-th fuzzy linear model is calculated based on the preset model conditional filtering formula; the predicted observation of each fuzzy linear model is calculated by combining the observation matrix and the state estimation; the first fuzzy membership of the predicted observation belonging to the fuzzy linear model is calculated based on the preset membership update expression; the antecedent parameter is updated, the second fuzzy membership of the antecedent variable is calculated, and the weight of each fuzzy linear model is calculated based on the second fuzzy membership; the target state and covariance at the kth moment are calculated by combining the state estimation, the first fuzzy membership and the weight. Through the implementation of the present invention, feedback learning is introduced into the strong tracking filtering method, a robust strong tracking filtering algorithm is constructed, the identification of the TS fuzzy model post-parameters is realized, and the EM algorithm and fuzzy subspace clustering are combined to construct the fuzzy clustering objective function, the identification of the TS fuzzy model antecedent parameters is realized, the estimation performance is effectively improved, the calculation burden is reduced, and the accuracy of the motion state determination is improved.

[0147] Second embodiment:

[0148] In order to solve the technical problems of the target detection and tracking algorithms provided in the related art, such as the heavy computational burden, low estimation performance, and poor accuracy of the determined motion state, this embodiment shows a target tracking device based on a variable structure TS fuzzy semantic model. For details, please refer to Figure 2 , the target tracking device of this embodiment includes:

[0149] An estimation module 201 is used to estimate the state prediction value of the i-th fuzzy linear model by combining the state transfer matrix and the target state at the k-1th moment;

[0150] A first calculation module 202, used to calculate the state estimation of the i-th fuzzy linear model based on a preset model condition filtering formula;

[0151] The second calculation module 203 is used to calculate the predicted observation of each fuzzy linear model by combining the observation matrix and the state estimation;

[0152] A third calculation module 204 is used to calculate the first fuzzy membership of the predicted observation belonging to the fuzzy linear model based on a preset membership update expression;

[0153] A fourth calculation module 205, used for updating the antecedent parameters, calculating the second fuzzy membership of the antecedent variables, and calculating the weight of each fuzzy linear model based on the second fuzzy membership;

[0154] The fifth calculation module 206 is used to calculate the target state and covariance at the kth moment by combining the state estimation, the first fuzzy membership, and the weight.

[0155] In an optional implementation of this embodiment, the estimation module is specifically used to: input the state transfer matrix and the target state at the k-1th moment into a preset state prediction value calculation formula to estimate the state prediction value of the i-th fuzzy linear model;

[0156] The calculation formula of the state prediction value is expressed as:

[0157]

[0158] in, represents the state transfer matrix, T represents the time interval, ω i represents the angular velocity of the target in the i-th fuzzy linear model, represents the target state of the i-th fuzzy linear model at the k-1th moment, Represents the state prediction value of the i-th fuzzy linear model.

[0159] In an optional implementation of this embodiment, the model condition filtering formula is expressed as:

[0160]

[0161]

[0162]

[0163]

[0164]

[0165]

[0166]

[0167]

[0168]

[0169] in, and The final fusion of k-1 moments and model i mixed initial state estimate After the system state transfer matrix get, represents the error covariance matrix of the mixed state, represents the covariance matrix of the observation noise in the ith model, P k-1 represents the filtering error covariance matrix at the previous moment, represents the fading factor, represents the state estimate, represents the measurement matrix, I represents the identity matrix, represents the optimal gain matrix, represents the error covariance matrix, Represents the gain parameter of the feedback learning term.

[0170] In an optional implementation of this embodiment, the second calculation module is specifically used to: input the observation matrix and the state estimate into a preset prediction observation calculation formula to calculate the prediction observation of each fuzzy linear model;

[0171] The prediction observation calculation formula is expressed as:

[0172]

[0173] in, represents the predicted observation based on the i-th fuzzy linear model at time k, represents the observation matrix, Represents the state estimate.

[0174] In an optional implementation of this embodiment, the membership update expression is expressed as:

[0175]

[0176] in, Indicates that the predicted observation belongs to the first fuzzy membership of the fuzzy linear model, β represents a constant, represents the weight of feature m in the i-th fuzzy linear model, represents the weight of feature m in the qth fuzzy linear model, Represents spatial constraint information, represents the metric function between observation l and the predicted observation in the i-th fuzzy linear model, represents the metric function between observation l and the predicted observation in the qth fuzzy linear model.

[0177] In an optional implementation of this embodiment, the fourth calculation module is specifically used to: input the second fuzzy membership into a preset weight calculation formula to calculate the weight of each fuzzy linear model;

[0178] The weight calculation formula is expressed as:

[0179]

[0180] in, Represents the antecedent variable Fuzzy membership function in fuzzy linear model The second fuzzy membership degree of Represents the weight of each fuzzy linear model.

[0181] Further, in an optional implementation of the present embodiment, the fifth calculation module is specifically used to: input the state estimation, the first fuzzy membership, and the weight into a preset target state calculation formula to calculate the target state at the kth moment, and input the state estimation, the first fuzzy membership, the weight, and the target state at the kth moment into a preset covariance calculation formula to calculate the covariance;

[0182] The target state calculation formula is expressed as:

[0183]

[0184] The covariance calculation formula is expressed as:

[0185]

[0186] in, represents the state estimate, represents the first fuzzy membership, represents the weight, represents the target state at the kth moment, represents the error covariance matrix, P k represents the covariance.

[0187] Furthermore, in another optional implementation of this embodiment, the target tracking device further includes: a selection module, configured to select a fuzzy rule based on a preset fuzzy rule reduction algorithm;

[0188] The fuzzy rule reduction algorithm is expressed as:

[0189]

[0190]

[0191]

[0192]

[0193]

[0194]

[0195]

[0196]

[0197]

[0198]

[0199]

[0200] Among them, x max and x min Represent the maximum and minimum values ​​of the sample array, X is the target fuzzy set, R1, R2, L, represents m given equivalence relations, Represented by the relationship R i The equivalence class of induced x, express The conditional probability with respect to X.

[0201] It should be noted that the target tracking methods in the aforementioned embodiments can all be implemented based on the target tracking device provided in this embodiment. Ordinary technical personnel in the relevant field can clearly understand that for the convenience and conciseness of description, the specific working process of the target tracking device described in this embodiment can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0202] The target tracking device based on the variable structure TS fuzzy semantic model provided by the present embodiment is used to estimate the state prediction value of the i-th fuzzy linear model in combination with the state transfer matrix and the target state at the k-1th moment; the state estimation of the i-th fuzzy linear model is calculated based on the preset model conditional filtering formula; the predicted observation of each fuzzy linear model is calculated in combination with the observation matrix and the state estimation; the first fuzzy membership of the predicted observation belonging to the fuzzy linear model is calculated based on the preset membership update expression; the antecedent parameter is updated, the second fuzzy membership of the antecedent variable is calculated, and the weight of each fuzzy linear model is calculated based on the second fuzzy membership; the target state and covariance at the kth moment are calculated in combination with the state estimation, the first fuzzy membership, and the weight. Through the implementation of the present invention, feedback learning is introduced into the strong tracking filtering method, a robust strong tracking filtering algorithm is constructed, and the identification of the TS fuzzy model consequent parameters is realized, and the EM algorithm and fuzzy subspace clustering are combined to construct a fuzzy clustering objective function, and the identification of the TS fuzzy model antecedent parameters is realized, which effectively improves the estimation performance, reduces the calculation burden, and improves the accuracy of the motion state determination.

[0203] Third embodiment:

[0204] This embodiment provides an electronic device, see Figure 3 As shown, it includes a processor 301, a memory 302 and a communication bus 303, wherein: the communication bus 303 is used to realize the connection and communication between the processor 301 and the memory 302; the processor 301 is used to execute one or more computer programs stored in the memory 302 to implement at least one step in the target tracking method in the above-mentioned embodiment 1.

[0205] The present embodiment also provides a computer-readable storage medium, which includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer.

[0206] The computer-readable storage medium in this embodiment can be used to store one or more computer programs, and the one or more computer programs stored therein can be executed by a processor to implement at least one step of the method in the above-mentioned embodiment 1.

[0207] This embodiment also provides a computer program that can be distributed on a computer-readable medium and executed by a computing device to implement at least one step of the method in the above-mentioned embodiment one; and in some cases, at least one step shown or described can be executed in an order different from that described in the above-mentioned embodiment.

[0208] This embodiment further provides a computer program product, including a computer readable device, on which the computer program as shown above is stored. In this embodiment, the computer readable device may include the computer readable storage medium as shown above.

[0209] It can be seen that those skilled in the art should understand that all or some of the steps, systems, and functional modules / units in the above disclosed methods can be implemented as software (which can be implemented with computer program code executable by a computing device), firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component can have multiple functions, or a function or step can be performed by several physical components in cooperation. Some or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit.

[0210] In addition, it is well known to those skilled in the art that communication media generally contain computer readable instructions, data structures, computer program modules or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery media. Therefore, the present invention is not limited to any specific hardware and software combination.

[0211] The above contents are further detailed descriptions of the embodiments of the present invention in combination with specific implementation methods, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A target tracking method based on variable structure TS fuzzy semantic model, characterized in that: include: Combined with the state transfer matrix and k -1 time target state, estimate the i The state prediction value of a fuzzy linear model; Based on the preset model condition filtering formula, calculate the i State estimation of a fuzzy linear model; Calculating the predicted observations of each fuzzy linear model in combination with the observation matrix and the state estimate; Based on a preset membership updating expression, calculating a first fuzzy membership of the predicted observation to the fuzzy linear model; Updating antecedent parameters, calculating a second fuzzy membership of antecedent variables, and calculating a weight of each of the fuzzy linear models based on the second fuzzy membership; Combine the state estimation, the first fuzzy membership, and the weight to calculate the k The target state and covariance at each moment; The model condition filtering formula is expressed as: in, and Respectively by The final fusion of moments and Model Hybrid Initial State Estimation After the system state transfer matrix get, represents the error covariance matrix of the mixed state, Indicates The covariance matrix of the observation noise in the model, represents the filtering error covariance matrix at the previous moment, represents the fading factor, represents the state estimate, represents the measurement matrix, represents the identity matrix, represents the optimal gain matrix, represents the error covariance matrix, Represents the gain parameter of the feedback learning term.

2. The target tracking method according to claim 1, characterized in that: The combined state transfer matrix and the k -1 time target state, estimate the i The steps of predicting the state value of a fuzzy linear model include: The state transfer matrix and k The target state at time -1 is input into the preset state prediction value calculation formula to estimate the i The state prediction value of a fuzzy linear model; The state prediction value calculation formula is expressed as: ， ； in, represents the state transfer matrix, Indicates the time interval, represents the angular velocity of the target in the i-th fuzzy linear model, Indicates the i The fuzzy linear model is k -1 moment target state, Indicates the i The state prediction value of a fuzzy linear model.

3. The target tracking method according to claim 1, characterized in that: The step of calculating the predicted observation of each fuzzy linear model by combining the observation matrix and the state estimate comprises: Inputting the observation matrix and the state estimate into a preset prediction observation calculation formula to calculate the prediction observation of each fuzzy linear model; The prediction observation calculation formula is expressed as: ， in, express k Time based on the i The predicted observations of a fuzzy linear model, represents the observation matrix, represents the state estimate.

4. The target tracking method according to claim 1, characterized in that: The membership update expression is expressed as: in, Indicates that the predicted observation belongs to the first fuzzy membership of the fuzzy linear model, β Represents a constant, Indicates i The features in the fuzzy linear model m The weight of Indicates q The features in the fuzzy linear model m The weight of Represents spatial constraint information, Indicates observation l With i The metric function between the predicted observations in the fuzzy linear model is Indicates observation l With q The measure function between the predicted observations in a fuzzy linear model.

5. The target tracking method according to claim 1, characterized in that: The step of calculating the weight of each of the fuzzy linear models based on the second fuzzy membership comprises: Inputting the second fuzzy membership into a preset weight calculation formula to calculate the weight of each of the fuzzy linear models; The weight calculation formula is expressed as: in, Represents the antecedent variable The fuzzy membership function in the fuzzy linear model The second fuzzy membership degree of Represents the weight of each of the fuzzy linear models.

6. The target tracking method according to any one of claims 1 to 5, characterized in that: The method combines the state estimation, the first fuzzy membership, and the weight calculation k The target state at the moment and the steps of covariance include: The state estimation, the first fuzzy membership, and the weight are input into a preset target state calculation formula to calculate the k The target state at the moment, and the state estimation, the first fuzzy membership, the weight, the k The target state at the moment is input into the preset covariance calculation formula to calculate the covariance; The target state calculation formula is expressed as: The covariance calculation formula is expressed as: in, represents the state estimate, represents the first fuzzy membership, represents the weight, Indicates the k The target state at the moment, represents the error covariance matrix, Denotes the covariance.

7. The target tracking method according to any one of claims 1 to 5, characterized in that: Before the step of calculating the weight of each of the fuzzy linear models based on the second fuzzy membership, the method further includes: Selecting fuzzy rules based on a preset fuzzy rule reduction algorithm; The fuzzy rule reduction algorithm is expressed as: in, x max and x min Respectively represent the maximum and minimum values ​​of the sample array, X is the target fuzzy set, R 1, R 2, , R m R express m A given equivalence relation, Indicated by the relationship R i Induced x The equivalence class of express Relative to X The conditional probability of .

8. An electronic device, characterized in that: include: Memory, processor and communication bus; The communication bus is used to realize the connection and communication between the memory and the processor; The processor is used to execute the computer program stored in the memory; When the processor executes the computer program, the steps in the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 7 are implemented.

Images