Passive multi-station time difference target positioning and tracking method

Abstract

The application relates to the technical field of target positioning and tracking, and provides a passive multi-station time difference target positioning and tracking method. The method comprises the following steps: initializing and assigning an initial algorithm by using an initial state of initialization data and a covariance matrix of a positioning target, inputting a model into an initialization interactive algorithm, carrying out model interaction to obtain a mixed state estimation and a covariance matrix of the post-interaction model; performing state prediction and updating by using an improved cubature Kalman filter to obtain a model likelihood function; updating a model probability according to the model likelihood function to obtain an updated Markov transition probability matrix by using an improved IMM algorithm; and calculating a state estimation and a covariance matrix of a final fusion of the interactive multi-model to obtain a target positioning and tracking result. The QS-CKF algorithm is introduced, the IMM algorithm is improved, the Markov transition probability matrix is dynamically adjusted, the problems of filter divergence or tracking lag are avoided, and the performance of target positioning and tracking is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target positioning and tracking, and provides a passive multi-station time difference target positioning and tracking method. BACKGROUND

[0002] In modern civilian fields, target positioning and tracking technology has important application value, especially in passive multi-station time difference positioning (TDOA, Time Difference of Arrival) systems. The TDOA system measures the time difference of target signals arriving at multiple receiving stations to calculate the position information of the target. However, the high mobility and complex motion pattern of the target make it difficult for traditional positioning and tracking algorithms to meet the requirements of accuracy and real-time performance. The key to accurately tracking a mobile target lies in the selection of a suitable nonlinear filter and the optimization design of the tracking model.

[0003] Traditional target tracking algorithms (such as Kalman filter and its improved algorithms) have the following problems when dealing with nonlinear, non-Gaussian noise and high mobility targets:

[0004] Nonlinear problem: The observation equation of the TDOA positioning system is nonlinear, and traditional linear filtering algorithms (such as KF) are difficult to handle. At the same time, the traditional nonlinear filtering such as EKF (Extended Kalman Filter) will bring a large truncation error in the process of Taylor expansion approximation linearization.

[0005] High mobility problem: The target motion state may change rapidly, and traditional single model filtering algorithms cannot adapt to such changes. The classical target motion model mainly includes: CV model (Constant Velocity Model, constant velocity model), CA model (Constant Acceleration Model, constant acceleration model) and CT model (Constant Turn Rate Model, turning model). These target dynamic models are used to predict the future position and state of the target in target tracking, and provide the tracker with an estimate of the target's motion. The specific model used depends on the target's motion pattern and task requirements.

[0006] Noise problem: There are various noises (such as measurement noise, process noise) in the actual environment, and the robustness of traditional algorithms to noise is poor.

[0007] Therefore, there is an urgent need for a target positioning and tracking algorithm that can effectively handle nonlinearity, high mobility and noise interference. SUMMARY

[0008] The present application aims to at least solve one of the technical problems existing in the related art. To this end, the present application provides a passive multi-station time difference target positioning tracking method, which increases the transition probability of the matching model in the model stable stage, reduces the transition probability of the non-matching model, and accelerates the model switching speed in the model transition stage, thereby improving the tracking accuracy and response speed for high-maneuvering targets.

[0009] The present application provides a passive multi-station time difference target positioning tracking method, comprising the following steps:

[0010] S1: initializing and assigning the initial algorithm using the initial state of the initialization data and the covariance matrix of the positioning target, to obtain the initialization algorithm;

[0011] S2: inputting the model into the initialization algorithm, and performing model interaction to obtain the mixed state estimation and covariance matrix of the post-interaction model;

[0012] S3: performing model filtering on the mixed state estimation and covariance matrix of the post-interaction model, using Markov transition update for state prediction and update, to obtain the likelihood function of the model;

[0013] S4: performing model state fusion according to the likelihood function of the model, to obtain the target prediction model;

[0014] S5: inputting the model to be predicted into the target prediction model, to obtain the positioning tracking result of the target.

[0015] According to the passive multi-station time difference target positioning tracking method provided by the present application, the model comprises a uniform speed model, a uniform acceleration model, a turning model, a Singer model and a Jerk model. According to the passive multi-station time difference target positioning tracking method provided by the present application.

[0016] According to the passive multi-station time difference target positioning tracking method provided by the present application, step S1 comprises:

[0017] S11: calculating the state initial mean value according to the initial state, and performing QR decomposition on the covariance matrix to obtain the covariance square root;

[0018] S12: performing time update according to the state initial mean value and the covariance square root, to obtain the time updated state mean value and the time updated covariance square root;

[0019] S13: performing measurement update on the time updated state mean value and the time updated covariance square root, to obtain the state estimation of the initialization algorithm and the covariance matrix square root of the initialization algorithm.

[0020] According to the passive multi-station time difference target positioning tracking method provided by the present application, step S13 comprises:

[0021] S131: calculating a cubature point according to the time-updated state mean and the time-updated covariance square root;

[0022] S132: calculating a measurement prediction value according to the cubature point;

[0023] S133: calculating a square root of a measurement one-step prediction error covariance matrix and a cross-covariance according to the measurement prediction value;

[0024] S134: performing SVD decomposition on the cross-covariance to obtain an autocorrelation covariance;

[0025] S135: calculating a gain matrix according to the autocorrelation covariance, the square root of the measurement one-step prediction error covariance matrix and the cross-covariance;

[0026] S136: calculating a state estimation of the initialization algorithm according to the autocorrelation covariance and the gain matrix;

[0027] S137: calculating a covariance matrix square root of the initialization algorithm according to the autocorrelation covariance.

[0028] According to the passive multi-station time difference target positioning and tracking method provided by the application, step S2 comprises:

[0029] S21: calculating a mixed probability of the model;

[0030] S22: calculating a mixed state estimation of the model according to the mixed probability and the state estimation of the initialization algorithm;

[0031] S23: calculating a covariance matrix of the model according to the mixed probability and the covariance matrix square root of the initialization algorithm.

[0032] According to the passive multi-station time difference target positioning and tracking method provided by the application, step S3 comprises:

[0033] S31: performing model updating according to the mixed state estimation and the covariance matrix of the post-interaction model to obtain an updated model probability;

[0034] S32: performing Markov transition updating on the updated model probability to obtain a normalized probability model probability;

[0035] S33: performing secondary correction on the normalized probability model probability to obtain a normalized TPM diagonal element;

[0036] S34: inputting the normalized TPM diagonal element as a transition probability of the next time and continuously updating the model corresponding to the normalized probability model until the model converges to obtain a likelihood function of the model.

[0037] According to the passive multi-station time difference target positioning tracking method provided by the application, step S31 comprises:

[0038] S311: calculating a likelihood function corresponding to the model according to the mixed state estimation of the model and the covariance matrix polarity model of the model;

[0039] S312: updating the model according to the likelihood function corresponding to the model to obtain a model probability.

[0040] According to the passive multi-station time difference target positioning tracking method provided by the application, step S32 comprises:

[0041] S321: obtaining a model probability change amount at different time according to the updated model probability;

[0042] S322: calculating a model probability change amount between different models at the same time according to the obtained model probability change amount at different time;

[0043] S323: performing probability transition correction on the model probability change amount between different models at the same time according to the correction function to obtain a normalized model probability.

[0044] According to the passive multi-station time difference target positioning tracking method provided by the application, step S33 comprises:

[0045] S331: normalizing the likelihood function corresponding to the model to obtain a normalized likelihood function;

[0046] S332: calculating a likelihood function change amount according to the normalized likelihood function;

[0047] S333: calculating a prediction error according to the likelihood function change amount and normalizing the prediction error to obtain a normalized prediction error;

[0048] S334: judging whether to enter a model transition stage according to the normalized prediction error, if entering the model transition stage, executing step S335, if not entering the model transition stage, returning to step S331;

[0049] S335: calculating a TPM diagonal element according to the prediction error and normalizing the calculation result to obtain a normalized TPM diagonal element.

[0050] According to the passive multi-station time difference target positioning tracking method provided by the application, step S4 comprises:

[0051] S41: updating the probability of the model according to the likelihood function of the model to obtain an updated model probability;

[0052] S42: dynamically correct the Markov transition matrix using the correction coefficient according to the updated model probability, and adjust the model strategy;

[0053] S43: obtain a target prediction model by fusing the mixed state estimation of the model according to the updated model probability.

[0054] The one or more technical solutions in the embodiments of the present application have at least one of the following technical effects:

[0055] The passive multi-station time difference target positioning and tracking method provided by the present application significantly improves the numerical stability and calculation efficiency of the nonlinear filter by introducing the QS-CKF algorithm based on QR decomposition and SVD decomposition, and solves the problems of covariance matrix ill-conditioning and high computational complexity of the traditional CKF (Cubature Kalman Filter) in high-dimensional systems. At the same time, the improved IMM (Interacting Multiple Model Algorithm) algorithm dynamically adjusts the Markov transition probability matrix (TPM), increases the transition probability of the matching model and reduces the transition probability of the non-matching model in the model stable stage, and speeds up the model switching speed in the model transition stage, thereby improving the tracking accuracy and response speed of the high-maneuvering target. In addition, the combination of the QS-CKF and the improved IMM algorithm shows stronger robustness and stability in complex environments, avoids the common filter divergence or tracking lag problem in traditional algorithms, and significantly improves the performance of target positioning and tracking.

[0056] Additional aspects and advantages of the application will be described in part below, some will become apparent from the following description, or will be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0058] Figure 1 is a flowchart of a passive multi-station time difference target positioning and tracking method provided by the present application.

[0059] Figure 2 is a RMSE comparison chart of the present application and other methods.

[0060] Figure 3 is a filter average error comparison chart of the present application and other methods. DETAILED DESCRIPTION

[0061] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present application. The following embodiments are used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0062] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0063] The present application will be described below in combination with Figures 1 to 3 The present application is described.

[0064] Embodiments

[0065] As Figure 1 shown, Figure 1 a flowchart of a passive multi-station time difference target positioning and tracking method provided by the present application. Specifically, it includes:

[0066] S1: using the initial state of the initialization data and the covariance matrix of the positioning target to initialize and assign the initial algorithm, to obtain the initialization algorithm;

[0067] S2: inputting the model into the initialization algorithm, and performing model interaction to obtain the mixed state estimation and covariance matrix of the post-interaction model;

[0068] S3: performing model filtering on the mixed state estimation and covariance matrix of the post-interaction model, using Markov transition update to perform state prediction and update, to obtain the likelihood function of the model;

[0069] S4: performing model state fusion according to the likelihood function of the model, to obtain the target prediction model;

[0070] S5: inputting the model to be predicted into the target prediction model, to obtain the positioning and tracking result of the target.

[0071] Specifically, the models include the uniform speed model, the uniform acceleration model, the turning model, the Singer model, and the Jerk model.

[0072] Specifically, step S1 includes:

[0073] S11: Calculate the initial mean of the state based on the initial state, and perform QR decomposition on the covariance matrix to obtain the square root of the covariance. The specific process is as follows:

[0074] The initial mean of the state is The error covariance is QR decomposition By decomposing the roots, we obtain the following square roots:

[0075]

[0076] in, Covariance The square root of the covariance obtained by QR decomposition. for The expected predicted value, To obtain the expected function, This indicates the matrix transpose.

[0077] S12: Based on the initial mean and square root of the covariance of the state, perform time updates to obtain the updated mean and square root of the covariance. The specific process is as follows:

[0078] Calculate the volume point:

[0079]

[0080]

[0081] in, For the first The corresponding volume point of the first The mean of the state updated at the next time interval. For the first One volume point, For the first The expected predicted value at the next time update. For the dimensions of state variables, It is an identity matrix.

[0082] Propagation volume point:

[0083]

[0084] in, The mean of the states after the propagation volume point. Update the propagation function over time;

[0085] The state prediction value is calculated:

[0086]

[0087] For The corresponding state prediction value;

[0088] The square root of the state one-step prediction error covariance matrix is calculated :

[0089]

[0090] Wherein, The QR decomposition method is is the square root of the process noise covariance matrix, the first weighted center matrix is:

[0091]

[0092] S13: The time-updated state mean and the time-updated covariance square root are measurement-updated to obtain the state estimation of the initialization algorithm and the covariance matrix square root of the initialization algorithm, and step S13 comprises:

[0093] S131: The volume point is calculated according to the time-updated state mean and the time-updated covariance square root;

[0094]

[0095] The first The state mean of the first measurement update corresponding to the volume point;

[0096] S132: The measurement prediction value is calculated according to the volume point;

[0097]

[0098]

[0099] Wherein, The measurement state mean after propagating the volume point, The measurement propagation function is The measurement prediction value of the measurement propagation.

[0100] S133: The square root of the measurement one-step prediction error covariance matrix and the cross-covariance are calculated according to the measurement prediction value:

[0101]

[0102] wherein, is a square root of the one-step prediction error covariance matrix, is a cross-covariance, the second weighted centering matrix is:

[0103]

[0104] S134: performing singular value decomposition (SVD) on the cross-covariance to obtain an autocorrelation covariance:

[0105]

[0106] wherein, is a left singular matrix, is a diagonal singular matrix, is a right singular matrix, is a singular value decomposition.

[0107] S135: calculating a gain matrix from the first weighted centering matrix and the second weighted centering matrix;

[0108]

[0109] S136: calculating a state estimate of the initialization algorithm from the autocorrelation covariance and the gain matrix ;

[0110] .

[0111] S137: calculating a covariance matrix square root of the initialization algorithm from the autocorrelation covariance, a square root of a one-step prediction error covariance matrix of the measurement, and the cross-covariance.

[0112]

[0113]

[0114] wherein, is an updated state prediction value, is an updated gain matrix.

[0115] Specifically, the step S2 comprises:

[0116] S21: calculating a mixing probability of the model;

[0117] S22: calculating a mixed state estimate of the post-interaction model from the mixing probability and the state estimate of the initialization algorithm;

[0118] S23: calculating the covariance matrix of the post-interaction model according to the mixing probability and the square root of the covariance matrix of the initialization algorithm.

[0119] Suppose there are N models, the mixing probability of model j is

[0120]

[0121]

[0122] wherein represents the Markov transition probability of the model from to at the moment of time , that is, the first row and the first column of the Markov transition matrix. represents the probability of model as the matching model at the moment of time is the mixing normalization constant, and N is the total number of input models in the interactive multiple model algorithm.

[0123] The mixing state estimation and the covariance matrix of model are calculated by using the state estimation and the covariance matrix of model at the last moment of time:

[0124]

[0125]

[0126] is the state estimation of model at the moment of time , is the state estimation of model at the moment of time , and is the covariance matrix of model at the moment of time

[0127] Specifically, step S3 comprises:

[0128] S31: performing model updating according to the mixing state estimation and the covariance matrix of the post-interaction model to obtain updated model probabilities, comprising:

[0129] S311: calculating the likelihood function corresponding to the model according to the mixing state estimation of the model and the covariance matrix of the model.

[0130] calculating the likelihood function of the first model:​​​​​

[0131]

[0132] .

[0133] wherein, is the likelihood function of the th model, is the prediction difference, is the updated measurement propagated value at the end of the update.

[0134] S312: updating the model according to the likelihood function corresponding to the model to obtain an updated model probability

[0135]

[0136]

[0137]

[0138] wherein, is the model probability of the updated th model, is a normalization constant, is a normalized weighting coefficient, denotes the transition probability from model to model .

[0139] S32: using Markov transition to update the updated model probability to obtain a normalized probability model probability, comprising:

[0140] S321: obtaining a model probability change amount at different time according to the updated model probability;

[0141] calculating the model probability change amount at different time :

[0142]

[0143]

[0144] wherein denotes the posterior probability of the th model, denotes the prior probability of the th model, is a trend variable.

[0145] when the model j matches, is positive, otherwise is negative. Therefore The value has the following definition

[0146]

[0147] S322: Calculate the change in model probability between different models at the same time based on the change in model probability at different times;

[0148] Calculate the change in model probability between different models at the same time:

[0149]

[0150] in, For the model Comparison Model The change in model probability between them For exponential parameters, Indicates the first The posterior probability of each model.

[0151] When the model Comparison Model A better match, i.e., time k. ,but When the model Comparison Model More matching, i.e. ,but This reduces the probability of the matched model transitioning to the unmatched model while increasing the probability of the unmatched model transitioning to the matched model.

[0152] S323: Based on the correction function, perform probability transfer correction on the change in model probability between different models at the same moment to obtain the normalized probability model probability.

[0153] By combining the two correction functions mentioned above, each element of the probability transition matrix is ​​corrected:

[0154]

[0155] in, express The first step of the transition probability matrix at time step line, number Column elements, Indicates after modification The transition probabilities at time step 1. Furthermore, the elements in the transition probability matrix are non-negative, and due to the properties of a time-homogeneous Markov chain, the sum of all elements in each row equals 1.

[0156]

[0157] in Normalized to:

[0158]

[0159] wherein, is the normalized .

[0160] S33: twice modifying the normalized probability model probability to obtain a normalized TPM diagonal element, comprising:

[0161] To this end, the main diagonal element is set to limit When , do twice modification to

[0162]

[0163] wherein, represents the element of the row, the column of the transition probability matrix at the moment.

[0164] S331: normalizing the likelihood function corresponding to the model to obtain a normalized likelihood function;

[0165] First, calculate the likelihood function of the model and normalize it:

[0166]

[0167]

[0168] is the likelihood function of the th model, is the normalized likelihood function.

[0169] S332: calculating the likelihood function change amount according to the normalized likelihood function;

[0170] In calculating the likelihood function change amount of the model :

[0171]

[0172] wherein, is the normalized likelihood function before updating.

[0173] S333: calculating the prediction error according to the likelihood function change amount, and normalizing to obtain a normalized prediction error;

[0174] After that, for each model, calculate its prediction error :

[0175]

[0176] wherein, is the system measurement value at the current time, is the state prediction value of the i-th model, is the corresponding state prediction value of the i-th model, represents a modulo operation.

[0177] The prediction error is normalized:

[0178]

[0179] wherein, is the normalized prediction error.

[0180] S334: Determine whether to enter the model transition stage according to the normalized prediction error, if entering the model transition stage, execute step S335, if not entering the model transition stage, return to step S331;

[0181] Determine whether to enter the model transition stage by setting a first threshold and a second threshold :

[0182]

[0183]

[0184] wherein, is a maximum value function, is a minimum value function, is a mean value function, is a standard deviation function.

[0185] If one of the following conditions is met, enter the model transition stage:

[0186]

[0187] wherein, is an absolute value.

[0188] S335: Calculate the TPM diagonal element according to the prediction error, and normalize the calculation result to obtain the normalized TPM diagonal element.

[0189] Dynamically adjust the main diagonal element of the TPM:

[0190]

[0191] wherein, is the element of the main diagonal at the current time, is the element of the main diagonal at the last time.​ is a regulating factor, controlling the overall adjustment range, and is usually set as . is a first weight coefficient, is a second weight coefficient, respectively controlling the influence of the prediction error and the change of the likelihood function on the adjustment.

[0192] Adjusting the non-diagonal elements:

[0193]

[0194] wherein, is the element of the non-main diagonal at the current time, represents the element of the j-th row and the i-th column of the transition probability matrix at the current time.

[0195] Normalizing the adjusted TPM:

[0196]

[0197] wherein, is the element of the non-main diagonal after normalization.

[0198] S34: input the diagonal element of the normalized TPM as the transition probability of the next time, continue to update the model corresponding to the normalized probability model until the model converges, and obtain the likelihood function of the model.

[0199] Specifically, step S4 includes:

[0200] S41: update the probability of the model according to the likelihood function of the model, and obtain an updated model probability;

[0201] S42: dynamically correct the Markov transition matrix using a correction coefficient according to the updated model probability, and adjust the model strategy;

[0202] S43: fuse the mixed state estimation of the model according to the updated model probability, and obtain a target prediction model.

[0203] Specifically, step S5 includes: inputting the model to be predicted into the target prediction model to obtain the target positioning tracking result. Since the method proposed in the embodiment of the application is used to predict the model itself to obtain the target tracking result corresponding to the model.

[0204] The method proposed in the application and the method proposed in other applications are compared as shown in Table 1:

[0205] Table 1 Comparison results table

[0206] ​​

[0207] As shown in Table 1, it can be seen that the RMSE (Root Mean Square Error) and MAE (Mean Absolute Error) errors of the present application are less than those of other models, and the switching speed and matching probability of the model are also relatively high.

[0208] As shown in Figure 2 and Figure 3 , Figure 2 is the RMSE comparison chart of the present application and other methods. Among them, Figure 2 (a) in the X direction RMSE, Figure 2 (b) in the Y direction RMSE, it can be seen that the RMSE value of the improved method of the present patent is the smallest.

[0209] Figure 3 is the filter average error comparison chart of the present application and other methods. Among them, Figure 3 (a) in the X direction filter average error, Figure 3 (b) in the Y direction filter average error, it can be seen that the filter average error value of the improved method of the present patent is the smallest.

[0210] From the above two charts, it can be seen that the RMSE and filter average error of the present application are less than those of other methods, and have better accuracy.

[0211] The present application introduces the QS-CKF algorithm based on QR decomposition and SVD decomposition, significantly improves the numerical stability and calculation efficiency of nonlinear filtering, solves the problem of covariance matrix ill-conditioning and high computational complexity of traditional CKF in high-dimensional system. At the same time, the improved IMM algorithm adjusts the Markov transition probability matrix (TPM) dynamically, increases the transition probability of the matching model and reduces the transition probability of the non-matching model in the model stable stage, and speeds up the model switching speed in the model transition stage, thereby improving the tracking accuracy and response speed of the high maneuvering target. In addition, the algorithm combining QS-CKF and improved IMM shows stronger robustness and stability in complex environment, avoids the common filter divergence or tracking lag problem in traditional algorithm, and significantly improves the performance of target positioning and tracking.

[0212] Finally, it should be pointed out that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

[0213] It should be noted that embodiments of the present disclosure can be realized by hardware, software, or a combination of software and hardware. The hardware portion can be realized with a special logic, and the software portion can be stored in a memory and executed by a proper instruction execution system, such as a microprocessor or a specially designed hardware. Those skilled in the art can understand that the above-described apparatus and method can be realized using computer executable instructions and / or included in processor control codes, for example, such codes are provided on a programmable memory or a data carrier such as an optical or electronic signal carrier.

[0214] Furthermore, although the operations of the method of the present disclosure are described in a particular order in the drawings, this is not required or implied in any way, as these operations can be performed in any order, or some of the operations can be performed concurrently, that the desired result can be achieved. Conversely, certain steps can be omitted, combined, or further divided into additional steps. It should also be noted that features and functions of two or more devices according to the present disclosure can be embodied in one device. Conversely, features and functions of one device described above can be further divided into multiple devices.

[0215] While the present disclosure has been described with reference to several particular embodiments, it should be understood that the present disclosure is not limited to the particular embodiments disclosed. The present disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A passive multistatic time difference of arrival target location tracking method, characterized by, The method comprises the following steps: S1: initializing and assigning an initial algorithm using an initial state of initialization data and a covariance matrix of a positioning target, wherein the initial algorithm is a cubature Kalman filter model, and step S1 comprises: S11: calculating a state initial mean value according to the initial state, and performing QR decomposition on the covariance matrix to obtain a covariance square root; S12: performing time updating according to the state initial mean value and the covariance square root to obtain a time-updated state mean value and a time-updated covariance square root; S13: performing measurement updating on the time-updated state mean value and the time-updated covariance square root to obtain a state estimation of the initial algorithm and a covariance matrix square root of the initial algorithm; S2: inputting a motion model into the initial algorithm to perform filter model interaction to obtain a mixed state estimation and a covariance matrix; S3: performing model filtering on the mixed state estimation and the covariance matrix, performing state prediction and updating using Markov transition updating to obtain a likelihood function of a filter model, and dynamically adjusting a Markov transition probability matrix TPM by using an improved IMM algorithm; Step S3 comprises: S31: performing model updating according to the mixed state estimation and the covariance matrix to obtain an updated model probability; S32: performing Markov transition updating on the updated model probability to obtain a normalized probability model probability; S33: performing secondary correction on the normalized probability model probability to obtain a normalized TPM diagonal element; step S33 comprises: S331: normalizing a likelihood function corresponding to the filter model to obtain a normalized likelihood function; S332: calculating a likelihood function change amount according to the normalized likelihood function; S333: calculating a prediction error according to the likelihood function change amount, and normalizing the prediction error to obtain a normalized prediction error; S334: judging whether to enter the model transition stage according to the normalized prediction error, if entering the model transition stage, executing step S335, if not entering the model transition stage, returning to step S331; judging whether to enter the model transition stage by setting a first threshold value and a second threshold value ​ wherein, is a maximum function, is a normalized likelihood function, is a prediction error is a minimum function, is a mean function, is a standard deviation function; If one of the following conditions is met, the model transition stage is entered: wherein, is the absolute value, is the change in the likelihood function of the th model; S335: calculating a TPM diagonal element according to the prediction error, normalizing the calculation result to obtain a normalized TPM diagonal element; wherein, is an element of the main diagonal at the current time instant, is an element of the main diagonal at the previous time instant, is a tuning factor, is a first weight coefficient, is a second weight coefficient, is the normalized likelihood function before the update, is the change in the normalized likelihood function; adjusting a non-diagonal element: wherein, is the element of the current time non-diagonal matrix, represents the element of the current time transition probability matrix, in the i-th row and the j-th column, in the i-th row and the j-th column. normalizing the adjusted TPM: wherein, is the normalized off-diagonal element, and N is the total number of input models in the interacting multiple model algorithm S34: inputting the normalized TPM diagonal element as a transition probability at the next moment to continue updating a model corresponding to the normalized probability model until the model converges, and obtaining a likelihood function of the filter model; S4: performing model state fusion according to the likelihood function of the filter model to obtain a target prediction model; S5: inputting a model to be predicted into the target prediction model to obtain a positioning and tracking result of the target.

2. The passive multi-station TDOA target positioning and tracking method according to claim 1, characterized in that, The motion model comprises a uniform speed model, a uniform acceleration model, a turning model, a Singer model and a Jerk model.

3. The passive multi-station TDOA target location tracking method according to claim 1, wherein, Step S13 comprises: S131: calculating a cubature point according to the time-updated state mean value and the time-updated covariance square root; S132: calculating a measurement prediction value according to the cubature point; S133: calculating a square root of a measurement one-step prediction error covariance matrix and a cross-covariance according to the measurement prediction value; S134: performing SVD decomposition on the cross-covariance to obtain an autocorrelation covariance; S135: calculating a gain matrix according to the autocorrelation covariance, a square root of a one-step prediction error covariance matrix and a cross-covariance; S136: calculating a state estimation of the initialization algorithm according to the autocorrelation covariance and the gain matrix; S137: calculating a square root of a covariance matrix of the initialization algorithm according to the autocorrelation covariance.

4. The passive multi-station TDOA target location tracking method according to claim 1, wherein, Step S2 comprises: S21: calculating a mixing probability of the model; S22: calculating a mixing state estimation of the post-interaction model according to the mixing probability and the state estimation of the initialization algorithm; S23: calculating a covariance matrix of the post-interaction model according to the mixing probability and the square root of the covariance matrix of the initialization algorithm.

5. The passive multistatic TDOA target location tracking method according to claim 1, characterized in that, Step S31 comprises: S311: calculating a likelihood function corresponding to the model according to the mixing state estimation of the model and a covariance matrix of the model; S312: updating the model to obtain an updated model probability according to the likelihood function corresponding to the model.

6. The passive multistatic TDOA target location tracking method of claim 1, wherein, Step S32 comprises: S321: obtaining a model probability variation amount at different time instants according to the updated model probability; S322: calculating a model probability variation amount between different models at the same time instant according to the obtained model probability variation amount at different time instants; S323: performing probability transition correction on the model probability variation amount between different models at the same time instant according to a correction function to obtain a normalized model probability.

7. The passive multistatic TDOA target location tracking method of claim 1, wherein, Step S4 comprises: S41: updating the probability of the model to obtain an updated model probability according to the likelihood function of the model; S42: dynamically correcting a Markov transition matrix using a correction coefficient according to the updated model probability to adjust a model strategy; S43: fusing the mixing state estimation of the model to obtain a target prediction model according to the updated model probability.

Images