Maneuvering target tracking system and method based on hybrid coding and UKF decoding

Abstract

The invention relates to the technical field of target tracking, and discloses a maneuvering target tracking system and method based on hybrid coding and UKF decoding. The maneuvering target tracking method comprises the steps of obtaining radar measurement data of a target position at each moment based on a target trajectory; the method comprises the following steps: encoding radar measurement data by adopting an attention-GRU network to obtain a target state encoding value; based on the target state coding value, obtaining a corrected target state noise covariance matrix, a corrected target state measurement noise covariance matrix and a corrected state initial covariance matrix by adopting an expectation maximization algorithm; and performing UKF decoding calculation on radar measurement data by adopting the corrected target state noise covariance matrix, the corrected target state measurement noise covariance matrix and the corrected state initial covariance matrix to obtain a target position. Model parameters such as noise variance information do not need to be preset, the tracking precision of the maneuvering target can be effectively improved, and noise interference is weakened.

Description

Technical Field

[0001] The present invention belongs to the technical field of target tracking, relates to radar target tracking, and particularly relates to a maneuvering target tracking system and method based on hybrid coding and UKF decoding. Background Art

[0002] Maneuvering target tracking estimates the state of a target at subsequent moments by using the target position state measured by a sensor as a reference, so as to achieve stable continuation of the target track. With the improvement of target maneuverability, the uncertainty of the target motion state and measurement source increases significantly, which not only affects the tracking accuracy but may also lead to target loss of tracking. Therefore, researching high-precision target tracking technology plays a key role in real-time monitoring of positions, which can support rapid decision-making and actions and is of great significance in the field of situation awareness.

[0003] Traditional algorithms need to set prior information. For example, parameters such as system process noise and measurement noise are usually set according to the target motion model or human experience. Representative algorithms include: Kalman Filter (KF), Particle Filter (PF), Extended Kalman Filter (EKF), Cubature Kalman Filter (CKF), Unscented Kalman Filter (UKF), etc. Most of these methods are based on Markov property and observation condition independence, which usually does not hold in actual situations. Especially when a radar detects a maneuvering target, the change in the dynamic geometric relationship between the target and the radar may generate non-Gaussian noise, directly affecting the effectiveness of these tracking methods.

[0004] In recent years, with the rapid development of deep learning, non-parametric algorithms based on Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gate Recurrent Unit (GRU), etc. have been proposed. These algorithms establish mapping relationships in a data-driven manner and can effectively cope with the uncertainty of target motion. However, when tracking a trajectory with strong maneuverability, sufficient data is required to help the model learn the transition law between violently changing states in the deep learning algorithm, resulting in an increase in target state estimation error.

[0005] Therefore, in view of the problems of complex maneuvering characteristics of high-speed targets, unknown prior statistical characteristics of noise, and low radar tracking accuracy, there is an urgent need for a new maneuvering target tracking algorithm to improve the tracking performance of strongly maneuvering targets. Summary of the Invention

[0006] The object of the present invention is to provide a maneuvering target tracking system and method based on hybrid coding and UKF decoding for the technical defect that traditional tracking algorithms are difficult to adapt to various sudden maneuvers in the case of inability to obtain prior statistical characteristics of noise and complex maneuver characteristics, as well as the problems of large amounts of data training required for predicting trajectories by deep learning or possible error accumulation. According to the encoding-decoding architecture, the attention-GRU is used to encode the radar measurement data, analyze the correlation information between different states, enhance the maneuvering ability of the learning target, and weaken the Markov property of the state and the conditional independence assumption of the observation. Different weights are assigned to different states, the influence degrees of different factors on the GRU encoding during the movement of the maneuvering target are concerned, and the output is used to dynamically estimate the noise covariance matrix of the system and the covariance matrix of the measurement noise through the Expectation Maximization (EM) algorithm. The modified parameters are used for decoding through the UKF state prediction and update process to obtain high-precision state estimation.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] In the first aspect, the present invention provides a maneuvering target tracking method based on hybrid coding and UKF decoding, including:

[0009] S10. Obtain the radar measurement data of the target position at each moment based on the target trajectory;

[0010] S11. Encode the radar measurement data by using the attention-GRU network to obtain the target state encoding value;

[0011] S12. Based on the target state encoding value, obtain the modified target state noise covariance matrix, the modified target state measurement noise covariance matrix, and the modified state initial covariance matrix by using the expectation maximization algorithm;

[0012] S13. Perform UKF decoding calculation on the radar measurement data by using the modified target state noise covariance matrix, the modified target state measurement noise covariance matrix, and the modified state initial covariance matrix to obtain the target position.

[0013] As a possible implementation manner, S10 is specifically: adding ranging error and angle measurement error to the target trajectory to obtain the radar measurement data of the target position at each moment, and the radar measurement data includes: distance, speed, acceleration, azimuth angle, and elevation angle.

[0014] As a possible implementation manner, S11 includes:

[0015] S110. Input the radar measurement data of the target position at the previous moment into the attention-GRU network to obtain the target hidden state at the current moment;

[0016] S111. Estimate the target hidden state at the next moment based on the target hidden state at the current moment and the radar measurement data of the target position at the next moment;

[0017] S112. Calculate the attention weights of the target hidden state at each moment using a non-linear attention mechanism;

[0018] S113. Weightedly sum the target hidden state at each moment with its corresponding attention weights to obtain the target state coding value.

[0019] As a possible implementation, S12 includes:

[0020] S120. Initialize the UKF, input the target state coding value into the initialized UKF to obtain the state estimate of the UKF;

[0021] S121. Execute the expectation step in the expectation-maximization algorithm to obtain the UKF parameters based on the state estimate of the UKF. The UKF parameters include the target state noise covariance matrix, the target state measurement noise covariance matrix, and the state initial covariance matrix;

[0022] S122. Based on the target state noise covariance matrix and the target state measurement noise covariance matrix, execute the maximization step in the expectation-maximization algorithm to update the target state noise covariance matrix and the target state measurement noise covariance matrix;

[0023] S123. Iteratively execute steps S121 - S122 until the UKF parameters converge to obtain the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix.

[0024] As a possible implementation, the convergence of the UKF parameters specifically means: the changes in the target state noise covariance matrix and the target state measurement noise covariance matrix are less than 10 -3 , and the maximum number of iterations is 100.

[0025] As a possible implementation, S13 includes:

[0026] S130. Generate a set of Sigma points based on the radar measurement data at the previous moment, the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix;

[0027] S131. Predict the next state value of each Sigma point based on the initial Sigma points in a set of Sigma points to obtain a set of predicted Sigma points;

[0028] S132. Based on a set of predicted Sigma points, use weighted average calculation to obtain the predicted state value and covariance matrix of the UKF;

[0029] S133. Calculate the Kalman gain according to the difference between the predicted state value and the prediction of the radar measurement data at the current moment;

[0030] S134. Update the predicted state value and covariance matrix based on the Kalman gain to obtain the updated predicted state value and the updated covariance matrix. The updated predicted state value is the target position at the current moment;

[0031] S135. Replace the corrected state initial covariance matrix in S130 with the updated covariance matrix, and repeat steps S130 - S134 to obtain the target position at the next moment.

[0032] In a second aspect, the present invention provides a maneuvering target tracking system based on hybrid coding and UKF decoding, which is used to execute the maneuvering target tracking method based on hybrid coding and UKF decoding in the first aspect. The maneuvering target tracking system includes: an attention - GRU coding module, an EM parameter estimation module, and a UKF decoding module;

[0033] The attention - GRU coding module is used to encode the radar measurement data to obtain the target state coding value;

[0034] The EM parameter estimation module, based on the target state coding value, runs the expectation - maximization algorithm to obtain the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix;

[0035] The UKF decoding module uses the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix to perform UKF decoding calculation on the radar measurement data to obtain the target position.

[0036] As a possible implementation, the attention-GRU encoding module includes a GRU, an attention unit, and a fully connected layer. The GRU estimates the target hidden state at the current moment based on the radar measurement data at the previous moment, and estimates the target hidden state at the next moment based on the target hidden state at the current moment and the radar measurement data at the next moment. The attention unit is used to calculate the attention weights of the target hidden state at each moment using the Softmax non-linear attention mechanism. The fully connected layer is used to perform a weighted sum of the target hidden state at each moment and its corresponding attention weights to obtain the target state encoding value.

[0037] As a possible implementation, the GRU includes an update gate and a reset gate. The update gate is used to control the proportion of retaining the target hidden state at the previous moment, and the reset gate is used to control the influence of the target hidden state at the current moment on the target hidden state at the next moment.

[0038] Compared with the prior art, the beneficial effects produced by the present invention are as follows:

[0039] 1. For the maneuvering target tracking method based on hybrid encoding and UKF decoding provided by the present invention, the attention-GRU network is used to encode the radar measurement data. After obtaining the target state encoding value, the expectation maximization algorithm is used to obtain the corrected target state noise covariance matrix, the corrected target state measurement noise covariance matrix, and the corrected state initial covariance matrix, and then through the unscented Kalman filter, the Kalman filter gain is calculated, so as to calculate the filtered value of the target position information. There is no need to preset model parameters such as noise variance information, effectively improving the tracking accuracy and weakening the noise interference.

[0040] 2. For the maneuvering target tracking method and system based on hybrid encoding and UKF decoding provided by the present invention, encoding is carried out relying on the GRU. The GRU only includes two gate structures, namely the update gate and the reset gate, which solves the problem of long-term time dependence, improves the flexibility of information processing, and captures target sudden maneuvers by assigning weights to the inputs at different time steps through the attention mechanism, improving the adaptability to target maneuvers, and can effectively weaken the Markov property of the state and the assumption of conditional independence of observations.

[0041] 3. For the maneuvering target tracking system based on hybrid encoding and UKF decoding provided by the present invention, relying on the EM parameter estimation module, dynamic estimation of noise distribution parameters is realized. Compared with the KF and UKF with fixed noise covariance, this module has an adaptive ability. By correcting the noise covariance matrix, the covariance matrix of the measurement noise, and the initial covariance matrix, the target position is calculated. This method effectively weakens the noise interference, so as to achieve more accurate target tracking in a glint noise scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings described herein are used to provide a further understanding of the present invention and form a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0043] Figure 1 is a flowchart of a maneuvering target tracking method based on hybrid coding and UKF decoding in an embodiment of the present invention;

[0044] Figure 2 is a maneuvering target trajectory and observation diagram in an embodiment of the present invention;

[0045] Figure 3 is the position root mean square error curve of four different models under Gaussian noise scenario in an embodiment of the present invention;

[0046] Figure 4 is the position root mean square error curve of four different models under flicker noise scenario in an embodiment of the present invention;

[0047] Figure 5 is the average absolute error in the X direction of the maneuvering target tracking method provided by the present invention under Gaussian noise scenario in an embodiment of the present invention;

[0048] Figure 6 is the average absolute error in the Y direction of the maneuvering target tracking method provided by the present invention under Gaussian noise scenario in an embodiment of the present invention;

[0049] Figure 7 is the average absolute error in the Z direction of the maneuvering target tracking method provided by the present invention under Gaussian noise scenario in an embodiment of the present invention;

[0050] Figures 8 to 9 is a schematic structural diagram of a maneuvering target tracking system provided by an embodiment of the present invention;

[0051] Figure 10 is a schematic structural diagram of an attention-GRU coding module in the maneuvering target tracking system provided by an embodiment of the present invention;

[0052] Figure 11 is a schematic structural diagram of GRU in the maneuvering target tracking system provided by an embodiment of the present invention.

[0053] Reference numerals

[0054] 1 - attention-GRU coding module, 10 - GRU, 100 - update gate, 101 - reset gate, 11 - attention unit, 12 - fully connected layer, 2 - EM parameter estimation module, 3 - UKF decoding module. Detailed implementation manners

[0055] In order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish identical or similar items with basically the same functions and roles. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and do not limit their sequence. Those skilled in the art can understand that terms such as "first" and "second" do not limit the quantity and execution order, and the terms "first" and "second" do not necessarily limit being different.

[0056] It should be noted that in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" aims to present relevant concepts in a specific manner.

[0057] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. The following at least one (item) or its similar expression refers to any combination of these items, including any combination of single item (item) or plural items (items). For example, at least one (item) of a, b or c can represent: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b and c can be single or multiple.

[0058] The embodiments of the present invention aim at the technical defects that traditional tracking algorithms cannot accurately model various maneuvers and are difficult to adapt to various sudden maneuvers, as well as the problems of large amount of data training required for predicting trajectories by deep learning or possible error accumulation. A maneuvering target tracking system and method based on hybrid coding and UKF decoding are proposed. According to the encoding-decoding architecture, the radar measurement data is encoded by using attention-GRU, the association information between different states is analyzed, the maneuvering ability of the learning target is enhanced, and the Markov property of the state and the conditional independence assumption of the observation are weakened. Different weights are assigned to different states, the influence degree of different factors on the GRU encoding during the movement of the maneuvering target is concerned, and the output is used to dynamically estimate the noise covariance matrix Q of the system and the covariance matrix R of the measurement noise through the EM algorithm. The decoded state prediction and update process is performed by using the corrected parameters to obtain high-precision state estimation.

[0059] In a first aspect, an embodiment of the present invention provides a maneuvering target tracking method based on hybrid coding and UKF decoding. Refer to Figure 1 , including:

[0060] S10. Obtain radar measurement data of the target position at each moment based on the target trajectory;

[0061] As a possible implementation, S10 is specifically: adding ranging error and angle measurement error to the target trajectory to obtain radar measurement data, where the radar measurement data includes: distance, speed, acceleration, azimuth angle, and elevation angle. Specifically in implementation, a jerk motion model is used to model a high-speed maneuvering target with a speed of Mach 8 and an acceleration of up to 20G. Generate the target trajectory as follows: set the trajectory parameters, construct the jerk model motion trajectory, and the trajectory information consists of position coordinates in a three-dimensional plane, and use it as the ground truth of the data set. Add ranging error and angle measurement error to this target trajectory to obtain radar measurement data containing ranging error and angle measurement error.

[0062] S11. Encode the radar measurement data using an attention-GRU network to obtain the target state encoding value;

[0063] As a possible implementation, S11 includes:

[0064] S110. Input the radar measurement data of the target position at the previous moment into the attention-GRU network to obtain the target hidden state at the current moment;

[0065] As an example, for the initial moment, input the radar measurement data of the target position at the initial moment into the attention-GRU network. For the k-th moment, the input of the attention-GRU network is the radar measurement data of the target position at the (k - 1)-th moment.

[0066] S111. Estimate the target hidden state at the next moment based on the target hidden state at the current moment and the radar measurement data of the target position at the next moment;

[0067] Specifically in implementation, the attention-GRU network includes an update gate and a reset gate. The update gate is used to control the proportion of retaining the target hidden state at the previous moment, and the reset gate is used to control the influence of the target hidden state at the current moment on the target hidden state at the next moment.

[0068] As an example, the update gate in the attention-GRU network performs the following operation:

[0069] z k =σ(W z x k +U z h k_1 +bz ) (1)

[0070] The reset gate performs the following operations:

[0071] r k = σ(W z x k + U r h k_1 + b r ) (2)

[0072] The target hidden state at time k is:

[0073]

[0074] where

[0075] In equations (1) to (4), is the intermediate state that combines the current input and the reset gate, k represents the time, which is a positive integer; x k represents the radar measurement data of the target position at time k, h k is the target hidden state at time k, h k-1 is the target hidden state at time k - 1; σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function; ⊙ represents the dot product; W and U are the weight matrices of the input and hidden layers respectively; b is the corresponding bias vector.

[0076] S112. Calculate the attention weights of the target hidden state at each time using a non - linear attention mechanism;

[0077] As an example, the non - linear attention uses Softmax attention.

[0078] S113. Weighted - sum each target hidden state with its corresponding attention weight to obtain the target state encoding value.

[0079] S12. Based on the target state encoding value, use the expectation - maximization algorithm to obtain the corrected target state noise covariance matrix, the corrected target state measurement noise covariance matrix, and the corrected state initial covariance matrix;

[0080] As a possible implementation, S12 includes:

[0081] S120. Initialize the UKF, input the target state encoding value into the initialized UKF, and obtain the state estimate of the UKF;

[0082] S121. Execute the expectation step in the expectation-maximization algorithm to obtain UKF parameters based on the state estimation of UKF. The UKF parameters include the target state noise covariance matrix, the target state measurement noise covariance matrix, and the state initial covariance matrix;

[0083] S122. Based on the target state noise covariance matrix and the target state measurement noise covariance matrix, execute the maximization step in the expectation-maximization algorithm to update the target state noise covariance matrix and the target state measurement noise covariance matrix;

[0084] S123. Iteratively execute steps S121 - S122 until the UKF parameters converge, and obtain the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix.

[0085] As an example, the expression of the target state noise covariance matrix is as follows:

[0086]

[0087] The expression of the target state measurement noise covariance matrix is as follows:

[0088]

[0089] In equations (5) - (6), x k is the state vector at time k, y k is the measurement vector at time k, the nonlinear functions f(·) and h(·) are the process function and the measurement function respectively, and N represents the total number of time instants.

[0090] As a possible implementation, set the maximum number of iterations to 100 times. The convergence of the UKF parameters specifically means that the changes in both the target state noise covariance matrix and the target state measurement noise covariance matrix are less than 10 -3 .

[0091] S13. Use the corrected target state noise covariance matrix, the corrected target state measurement noise covariance matrix, and the corrected state initial covariance matrix to perform UKF decoding calculation on the radar measurement data to obtain the target position.

[0092] As a possible implementation, S13 includes:

[0093] S130. Generate a set of Sigma points based on the radar measurement data at the previous time instant, the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix;

[0094] As an example, the expression of the Sigma points is as follows:

[0095]

[0096] Among them, represents the mean of the state variables, n represents the dimension of the state variables, λ controls the scaling ratio of the Sigma point set, and n + λ cannot be equal to zero. Usually, n + λ = 3.

[0097] S131. Predict the next state value of each Sigma point based on the initial Sigma point in a set of Sigma points to obtain a set of predicted Sigma points;

[0098] As an example, the expression of the predicted Sigma point is as follows:

[0099]

[0100] Among them, is the predicted Sigma point, and the nonlinear function f(·) is the process function.

[0101] S132. Based on a set of predicted Sigma points, use weighted average calculation to obtain the predicted state value and covariance matrix of the UKF;

[0102] As an example, the expressions of the predicted state value and covariance matrix are as follows:

[0103]

[0104] Among them, are 2n + 1 sigma points; i represents the i-th sigma point, and w i is the weight of the i-th sigma point; is the predicted state value.

[0105] S133. Calculate the Kalman gain according to the difference between the predicted state value and the prediction of the radar measurement data at the current moment;

[0106] As an example, the expression of the prediction of the radar measurement data at the current moment is as follows:

[0107]

[0108] Among them, Y k+1|k is the prediction of the radar measurement data at the current moment (abbreviated as measurement prediction), and the nonlinear function h(·) is the measurement function.

[0109]

[0110] Among them, P yy is the measurement prediction covariance matrix, and P xy is the cross-covariance matrix between the predicted state value and the measurement prediction.

[0111] As an example, the Kalman gain expression is as follows:

[0112]

[0113] Among them, K k+1 represents the Kalman gain matrix.

[0114] S134. Update the predicted state value and covariance matrix based on the Kalman gain to obtain the updated predicted state value and the updated covariance matrix. The updated predicted state value is the target position at the current moment;

[0115] As an example, the expression for the updated predicted state value is as follows:

[0116]

[0117] Among them, Y k+1 is the radar measurement data at the current moment.

[0118] The expression for the updated covariance matrix is as follows:

[0119]

[0120] S135. Replace the corrected state initial covariance matrix in S130 with the updated covariance matrix, and repeat steps S130 - S134 to obtain the target position at the next moment.

[0121] Next, a simulation experiment is carried out by combining the maneuvering target tracking method provided by the present invention with the maneuvering target tracking method of adaptive UKF decoding, and the maneuvering scenarios of Gaussian noise and flicker noise are respectively tracked. Then, in the same maneuvering scenario, the target is tracked by using Kalman filter, unscented Kalman filter, LSTM combined with Kalman filter, and LSTM combined with unscented Kalman filter. Taking the root mean squared error (RMSE) and average root mean squared error (ARMSE) of the position as evaluation indicators, the advantages of the present invention are elaborated in combination with the experimental data.

[0122] The target trajectory and observations are as Figure 2 shown, and the RMSE curves of the above four different models in the two scenarios are as Figures 3 to 4 shown. Among them, Figure 3 is the root mean squared error of the position in the Gaussian noise scenario, Figure 4 is the root mean squared error of the position in the flicker noise scenario, Figures 5 to 7They are the average absolute errors in the X, Y, and Z directions under the Gaussian noise scenario. After 100 Monte Carlo simulations, the position ARMSE results of different tracking models are shown in Table 1 as follows:

[0123] Table 1 Comparison of position ARMSE of five methods under different noise scenarios

[0124] Original KF UKF LSTM-KF LSTM-UKF This method model Gaussian noise 0.787 0.513 0.448 0.444 0.360 0.230 Flicker noise 0.422 0.294 0.254 0.247 0.200 0.145

[0125] It can be seen that the method provided by the present invention can effectively track high-speed maneuvering targets with a speed of up to Mach 8 and an acceleration of up to 20G. Under the Gaussian noise scenario, compared with tracking models such as KF, UKF, LSTM-KF, and LSTM-UKF, the ARMSE of this method is reduced by 36.0%, 27.7%, 27.2%, and 21.2% respectively. Under the flicker noise scenario, the ARMSE of this method is significantly lower than that of the other four models, with reductions of 36.49%, 27.02%, 25.41%, and 14.22% respectively. The results show that the method of the present invention significantly improves the tracking accuracy of maneuvering targets, enhances the adaptability of the algorithm to maneuvers, and verifies its effectiveness.

[0126] In a second aspect, the present invention provides a maneuvering target tracking system based on hybrid coding and UKF decoding, which is used to execute the maneuvering target tracking method based on hybrid coding and UKF decoding in the first aspect. Refer to Figures 8 to 9 , the maneuvering target tracking system includes: an attention-GRU encoding module 1, an EM parameter estimation module 2, and a UKF decoding module 3;

[0127] The attention-GRU encoding module 1 is used to encode radar measurement data to obtain the target state encoding value;

[0128] Based on the target state encoding value, the EM parameter estimation module 2 runs the expectation-maximization algorithm to obtain a corrected target state noise covariance matrix, a corrected target state measurement noise variance matrix, and a corrected state initial covariance matrix;

[0129] The UKF decoding module 3 uses the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix to perform UKF decoding calculations on the radar measurement data to obtain the target position.

[0130] Refer to Figure 10, as a possible implementation, the attention-GRU encoding module 1 includes a GRU 10, an attention unit 11, and a fully-connected layer 12. The GRU 10 estimates the target hidden state at the current moment based on the radar measurement data at the previous moment, and estimates the target hidden state at the next moment based on the target hidden state at the current moment and the radar measurement data at the next moment. The attention unit 11 is used to calculate the attention weight of the target hidden state at each moment by adopting the Softmax non-linear attention mechanism. The fully-connected layer 12 is used to perform weighted summation of the target hidden state at each moment and its corresponding attention weight to obtain the target state encoding value.

[0131] See Figure 11 , as a possible implementation, the GRU 10 includes an update gate 100 and a reset gate 101. The update gate 100 is used to control the proportion of retaining the target hidden state at the previous moment, and the reset gate 101 is used to control the influence of the target hidden state at the current moment on the target hidden state at the next moment.

[0132] Although the present invention has been described in conjunction with various embodiments, however, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure content, and the like. In the specification, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude the case of multiple. A single processor or other unit can implement several functions listed in the specification. Certain measures are described in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0133] Although the present invention has been described in conjunction with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present invention. Accordingly, the present specification and the drawings are merely exemplary descriptions of the present invention and are considered to have covered any and all modifications, variations, combinations, or equivalents within the scope of the present invention. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A maneuvering target tracking method based on hybrid coding and UKF decoding, characterized in that Including: S10. Obtain the radar measurement data of the target position at each moment based on the target trajectory; S11. Encode the radar measurement data using an attention-GRU network to obtain the target state encoding value; S12. Based on the target state encoding value, use the expectation-maximization algorithm to obtain a corrected target state noise covariance matrix, a corrected target state measurement noise covariance matrix, and a corrected state initial covariance matrix; S13. Perform UKF decoding calculation on the radar measurement data using the corrected target state noise covariance matrix, the corrected target state measurement noise covariance matrix, and the corrected state initial covariance matrix to obtain the target position.

2. The maneuvering target tracking method based on hybrid coding and UKF decoding according to claim 1, characterized in that The specific content of S10 is: Add ranging error and angle measurement error to the target trajectory to obtain the radar measurement data of the target position at each moment. The radar measurement data includes: distance, speed, acceleration, azimuth angle, and elevation angle.

3. The maneuvering target tracking method based on hybrid coding and UKF decoding according to claim 1, characterized in that The S11 includes: S110. Input the radar measurement data of the target position at the previous moment into the attention-GRU network to obtain the target hidden state at the current moment; S111. Estimate the target hidden state at the next moment based on the target hidden state at the current moment and the radar measurement data of the target position at the next moment; S112. Calculate the attention weight of the target hidden state at each moment using a non-linear attention mechanism; S113. Perform weighted summation of the target hidden state at each moment and its corresponding attention weight to obtain the target state encoding value.

4. The maneuvering target tracking method based on hybrid coding and UKF decoding according to claim 1, wherein The S12 includes: S120. Initialize UKF, input the target state encoding value into the initialized UKF to obtain the state estimate of UKF; S121. Execute the expectation step in the expectation-maximization algorithm to obtain UKF parameters based on the state estimate of UKF. The UKF parameters include the target state noise covariance matrix, the target state measurement noise covariance matrix, and the state initial covariance matrix; S122. Based on the target state noise covariance matrix and the target state measurement noise covariance matrix, execute the maximization step in the expectation-maximization algorithm to update the target state noise covariance matrix and the target state measurement noise covariance matrix; S123. Iteratively execute steps S121 - S122 until the UKF parameters converge to obtain a corrected target state noise covariance matrix, a corrected target state measurement variance matrix, and a corrected state initial covariance matrix.

5. The maneuvering target tracking method based on hybrid coding and UKF decoding according to claim 1, characterized in that The UKF parameter convergence specifically means that the changes in the target state noise covariance matrix and the target state measurement noise covariance matrix are less than 10 -3 , and the maximum number of iterations is 100.

6. The maneuvering target tracking method based on hybrid coding and UKF decoding according to claim 1, characterized in that, The S13 includes: S130. Generate a set of Sigma points based on the radar measurement data at the previous moment, the corrected target state noise covariance matrix, the corrected target state measurement variance matrix, and the corrected state initial covariance matrix; S131. Predict the next state value of each Sigma point based on the initial Sigma point in the set of Sigma points to obtain a set of predicted Sigma points; S132. Based on the set of predicted Sigma points, use weighted average calculation to obtain the predicted state value and covariance matrix of UKF. S133. Calculate the Kalman gain according to the difference between the predicted state value and the prediction of the radar measurement data at the current moment; S134. Update the predicted state value and the covariance matrix based on the Kalman gain to obtain the updated predicted state value and the updated covariance matrix, and the updated predicted state value is the target position at the current moment; S135. Replace the corrected state initial covariance matrix in S130 with the updated covariance matrix, and repeat steps S130 - S134 to obtain the target position at the next moment.

7. A maneuvering target tracking system based on hybrid coding and UKF decoding, characterized in that A maneuvering target tracking system for executing the maneuvering target tracking method based on hybrid coding and UKF decoding according to any one of claims 1 to 6, the maneuvering target tracking system comprising: an attention - GRU coding module, an EM parameter estimation module, and a UKF decoding module; The attention - GRU coding module is used to encode the radar measurement data to obtain a target state coding value; The EM parameter estimation module runs the expectation - maximization algorithm based on the target state coding value to obtain a corrected target state noise covariance matrix, a corrected target state measurement noise variance matrix, and a corrected state initial covariance matrix; The UKF decoding module performs UKF decoding calculation on the radar measurement data by using the corrected target state noise covariance matrix, the corrected target state measurement noise variance matrix, and the corrected state initial covariance matrix to obtain the target position.

8. The maneuvering target tracking system based on hybrid coding and UKF decoding according to claim 7, wherein, The attention - GRU coding module includes a GRU, an attention unit, and a fully - connected layer. The GRU estimates the target hidden state at the current moment based on the radar measurement data at the previous moment, and estimates the target hidden state at the next moment based on the target hidden state at the current moment and the radar measurement data at the next moment; The attention unit is used to calculate the attention weight of the target hidden state at each moment by using the Softmax non - linear attention mechanism; The fully - connected layer is used to perform weighted summation of the target hidden state at each moment and its corresponding attention weight to obtain the target state coding value.

9. The maneuvering target tracking system based on hybrid coding and UKF decoding according to claim 8, characterized in that, The GRU includes an update gate and a reset gate. The update gate is used to control the proportion of retaining the target hidden state at the previous moment, and the reset gate is used to control the influence of the target hidden state at the current moment on the target hidden state at the next moment.

Images