A target tracking method, system, device, medium and data processing terminal

Abstract

The present invention belongs to the technical field of state estimation and target tracking in signal processing, and discloses a target tracking method, system, equipment, medium and data processing terminal. The target tracking method includes: establishing a discrete system equation and initializing parameters; predicting the state at the next moment according to the previous state, and predicting and estimating the measurement according to the predicted state; inputting the measurement data measured by the sensor to obtain the measurement information; filtering the measurement through a smooth variable structure filter to obtain the measurement gain term; obtaining the coefficient corresponding to the state and the measurement gain term according to the covariance; estimating the state according to the coefficient corresponding to the state and the measurement gain term and the measurement gain term. The present invention proposes a new method for estimating the state without measurement value corresponding to the smooth variable structure filter, and provides different solutions for linear and nonlinear systems. The method proposed by the present invention has high accuracy and good robustness, good scalability, and can be used in various projects.

Description

Technical Field

[0001] The present invention belongs to the technical field of state estimation and target tracking in signal processing, and in particular relates to a target tracking method, system, equipment, medium and data processing terminal. Background Art

[0002] At present, state estimation has a wide range of applications in signal processing, navigation, battery management, fault diagnosis and target tracking. With more and more technological breakthroughs, the target state to be estimated is more and more prone to rapid changes. For example, in target tracking, weapons such as aircraft can quickly change speed and acceleration, enabling them to complete complex movements such as turning, snaking, accelerating, rolling, etc., making the target's maneuvering trajectory more and more complex. Traditional target tracking methods have difficulty adapting to these uncertain maneuvering changes, because most of the filters in these target tracking methods are based on statistical information, and they have difficulty handling the uncertainty caused by random changes in motion patterns.

[0003] At present, filters based on Kalman filters are widely used in target tracking, mainly linear Kalman filters, nonlinear extended Kalman filters, unscented Kalman filters and volumetric Kalman filters, etc. These filters are all derived based on Gaussian noise and require the establishment of an accurate state model. However, when the actual target is maneuvering, it is difficult to establish a state model consistent with the motion model for the filtering system because the motion model may change at any time. If the established state model is inconsistent with the actual motion model, the system is easy to diverge, resulting in unsatisfactory state estimation results. In addition, noise in actual engineering often does not obey Gaussian distribution, which makes it difficult for this type of method to meet actual needs.

[0004] The smooth variable structure filter (SVSF) is a new type of filter based on sliding mode control theory. It can maintain good robustness in the presence of model uncertainty and noise interference. When the target is maneuvering, even if there is an error between the state model of the smooth variable structure filter and the actual motion model, when the uncertainty and noise caused by the error have an upper bound, the estimated state will gradually approach the true state and converge within a bounded range. The filter has no requirements on the type of measurement noise, only requires the noise to have an upper bound value, and can be applied to non-Gaussian systems. With the continuous improvement of this theory, smooth variable structure filters are widely used in navigation, unmanned driving, fault diagnosis and other fields.

[0005] However, existing smooth variable structure filters also have problems: First, they are generally applied to systems where there is a one-to-one correspondence between states and measurement values, while in reality the state dimension is generally higher than the measurement dimension. For example, in radar target tracking, the target state that needs to be estimated may include position, velocity, and acceleration, while the radar can only measure position information and cannot directly measure other states such as velocity and acceleration. Smooth variable structure filters cannot directly estimate these states without measurement values. For this type of problem, existing smooth variable structure filters have two solutions: (1) construct pseudo-measurements, such as dividing the measurement values ​​twice by the sampling time to obtain pseudo-measurement values ​​of velocity. (2) construct a low-order Luenberger estimator to estimate states without measurement values. Both methods are mainly applied to linear systems. Second, the existing methods require that the measurement equations are linear, while the actual engineering field is more of a nonlinear system, and the relationship between the state and the measurement is also nonlinear, which severely limits the method in practical applications. Therefore, there is an urgent need for a new target tracking method and system to make up for the shortcomings of the existing technology.

[0006] Through the above analysis, the problems and defects of the prior art are as follows:

[0007] (1) The maneuvering trajectory of the target is becoming more and more complex. Since most filters in existing target tracking methods are based on statistical information, it is difficult to handle the uncertainty caused by the random changes in the motion pattern.

[0008] (2) Existing Kalman filters are all based on Gaussian noise and require the establishment of an accurate state model. However, when the actual target is maneuvering, the motion model may change at any time, making it difficult to establish a state model for the filtering system that is consistent with the motion model. This causes the system to easily diverge and the estimation effect to be poor. In addition, in actual situations, noise often does not obey the Gaussian distribution, which makes it difficult for this type of method to meet actual needs.

[0009] (3) Existing smooth variable structure filters are mainly used in systems where there is a one-to-one correspondence between states and measurement values. However, in reality, the state dimension is generally higher than the measurement dimension. In radar target tracking, smooth variable structure filters cannot directly estimate states without measurement values.

[0010] (4) The existing smooth variable structure filter requires that the measurement equation is linear. In actual engineering fields, there are more nonlinear systems. The relationship between measurement and state is nonlinear, that is, the measurement equation is nonlinear, which greatly limits the application of smooth variable structure filters.

[0011] The difficulty in solving the above problems and defects is as follows: in the existing literature on methods based on smooth variable structure filters, almost no scholars have proposed effective methods to estimate unmeasured states when both the state equation and the measurement equation are nonlinear, and these are often nonlinear in actual engineering. The method of the present invention solves this limitation.

[0012] The significance of solving the above problems and defects: The method of the present invention enables the smooth variable structure filter to be used in both linear and nonlinear systems, so that it is no longer limited by the restriction that the measurement equation must be linear. Like the Kalman filter, it has good scalability. For nonlinear systems, the method of the present invention and its improved method can be used in various systems, which is easier for engineering practice. The method of the present invention is for model uncertain systems, has good robustness and high accuracy, and can well estimate the state that does not correspond to the measurement value. Summary of the invention

[0013] In view of the problems existing in the prior art, the present invention provides a target tracking method, system, device, medium and data processing terminal, and more particularly, relates to a target tracking method, system, device, medium and data processing terminal based on a smooth variable structure filter.

[0014] The present invention is implemented as follows: a target tracking method, the target tracking method comprising the following steps:

[0015] Step 1: According to the tracking target system, a discrete linear system or a nonlinear system is established, a suitable target tracking motion model is selected for the system, and the system is initialized;

[0016] Step 2: predict the state of the target at the next moment through the state equation to obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement based on the measurement equation and the predicted value of the state;

[0017] Step 3: Input the sensor's measurement data at the next moment, and obtain new information based on the measurement data and the predicted value of the measurement;

[0018] Step 4, using the new information, calculate the measurement gain term according to the smooth variable structure filter;

[0019] Step 5: Obtain the coefficients corresponding to the state and measurement gain terms through the covariance between the state and measurement and the covariance between the measurement prediction value; obtain the correction term of the predicted state through this coefficient, that is, the coefficient corresponding to each state and gain term, and estimate the final value of the state.

[0020] Step 6, obtaining the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and outputting the state;

[0021] Step 7: After outputting the state, return to step 2, iterate, and estimate the state at subsequent moments;

[0022] Further, in step 1, the discrete system equation is as follows:

[0023] x k+1 =f(x k )+w k ;

[0024] z k+1 =h(x k+1 )+υ k+1 ;

[0025] In the formula, x k represents the state at time k, f represents the transition relationship from state k to the next time k+1, and w k represents process noise, h represents the corresponding relationship between state and measurement, z k+1 represents the measurement at time k+1, υ k+1 represents the measurement noise; in step 2, the predicted value of the state and the predicted value of the measurement Calculate using the following formulas:

[0026]

[0027]

[0028] In step 3, the measured new information e k+1|k Calculated by the following formula:

[0029]

[0030] In the formula, z k+1 Represents the measurement value measured by the system through the sensor at time k+1.

[0031] In step 4, the gain term G corresponding to the measurement k+1 Calculated by the following formula:

[0032]

[0033] In the formula, |e| means the absolute value of e; γ means the convergence rate, which is 0≤γ<1; represents the multiplication of corresponding elements of the matrix; sat represents the saturation function; ψ represents the width of the smooth bounded layer;

[0034] In step 5, the covariance P of the state and measurement is obtained through the system equation xz and the covariance P of the measured predicted value zz Then, the coefficient C corresponding to the state and measurement gain term is obtained.k+1 for:

[0035] C k+1 =P xz *(P zz ) -1 .

[0036] In step 6, the final state estimate is Calculated by the following formula:

[0037]

[0038] Further, in step 5, obtaining the coefficients corresponding to the state and the measurement gain term includes:

[0039] There are different methods for different systems. When the system equation in step 1 is a linear equation, the covariance P of the state prediction value in step 2 k+1|k Calculated according to the following formula:

[0040] P k+1|k =FP k|k F T +Q k ;

[0041] In the formula, the state transfer matrix F is the specific value of f, F T is the transpose of the matrix F, P k|k represents the covariance of the state at time k, Q k represents the process noise w k The covariance of .

[0042] Then the covariance P between the state and the measurement is xz Calculated by the following formula:

[0043] P xz =P k+1|k H T ;

[0044] Where H is the conversion matrix between state and measurement, which is the specific value of h in the linear system. T is the transpose of H. Then the covariance P of the measured predicted value is zz Calculated by the following formula:

[0045] P zz =HP k+1|k H T ;

[0046] Then the coefficient C corresponding to the state and measurement gain term is k+1 Calculated by the following formula:

[0047] C k+1 =Pxz *(P zz ) -1 .

[0048] Further, in step 5, obtaining the coefficients corresponding to the state and the measurement gain term includes:

[0049] When the system equation in step 1 is a nonlinear equation, the system equation needs to be processed when calculating the covariance, such as performing linear approximation on the nonlinear system, calculating the first-order Taylor expansion of the system equation, and the formulas for calculating the Jacobian matrices of f and h are as follows:

[0050]

[0051]

[0052] The covariance P of the state prediction value in step 2 k+1|k Calculated according to the following formula:

[0053]

[0054] Then the covariance P between the state and the measurement is xz and the covariance P of the measured predicted value zz Calculate using the following formulas:

[0055]

[0056]

[0057] Then the coefficient C corresponding to the state and measurement gain term is k+1 Calculated by the following formula:

[0058] C k+1 =P xz *(P zz ) -1 .

[0059] Further, in step 5, obtaining the coefficients corresponding to the state and the measurement gain term includes:

[0060] When the system equation in step 1 is a nonlinear equation, there are many methods for dealing with the nonlinear system equation. For example, the traceless transformation method can be used to calculate its mean and covariance.

[0061] According to the status and P k|k Get 2n+1 sampling points, sampling point X i,k|k And its corresponding weight W i The following formulas are used to calculate:

[0062]

[0063]

[0064] Where κ is a scale parameter, is (n+κ)P k|k The i-th row or i-th column of the RMS matrix, n is the dimension of the state, and i is the sampling point.

[0065] Sampling point X i,k|k The predicted value of the corresponding state Calculated by the following formula:

[0066]

[0067] Then the state prediction estimate in step 2 is and the state prediction covariance P k+1|k It is obtained by the following formula:

[0068]

[0069]

[0070] Then the predicted value of the measurement in step 2 is Predicted value through sampling points It is obtained by the following formula:

[0071]

[0072]

[0073] In the formula, It is the sampling point The corresponding measured predicted value, W i is its weight.

[0074] Then the covariance P between the state and the measurement is xz and the covariance P of the measured predicted value zz Calculate using the following formulas:

[0075]

[0076]

[0077] Then the coefficient C corresponding to the state and measurement gain term is k+1 Calculated by the following formula:

[0078] C k+1 =P xz *(P zz ) -1 .

[0079] Another object of the present invention is to provide a target tracking system applying the target tracking method, the target tracking system comprising:

[0080] The system initialization module is used to establish a discrete linear system or a nonlinear system according to the target tracking system, select a suitable target tracking motion model for the system, and initialize the system;

[0081] The prediction value calculation module is used to predict the state of the target at the next moment through the state equation, obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement according to the measurement equation and the predicted value of the state;

[0082] The measurement new information obtaining module is used to obtain new information by inputting the measurement data of the sensor at the next moment and according to the measurement data and the predicted value of the measurement;

[0083] A measurement gain item obtaining module is used to obtain the measurement gain item according to the smooth variable structure filter through the new information;

[0084] A coefficient calculation module, used to obtain the coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value;

[0085] A state estimation result acquisition module is used to obtain the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and output the state;

[0086] The iterative calculation module is used to return to the prediction value calculation module after outputting the state, iterate, and estimate the state at the subsequent time.

[0087] Another object of the present invention is to provide a computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, the processor performs the following steps:

[0088] (1) According to the target tracking system, a discrete linear system or a nonlinear system is established, a suitable target tracking motion model is selected for the system, and the system is initialized;

[0089] (2) Predict the state of the target at the next moment through the state equation to obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement based on the measurement equation and the predicted value of the state;

[0090] (3) Input the sensor's measurement data at the next moment, and obtain new information based on the measurement data and the predicted value of the measurement; and obtain the measurement gain term based on the new information according to the smooth variable structure filter;

[0091] (4) Obtain the coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value;

[0092] (5) Obtain the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and output the state;

[0093] (6) After outputting the state, return to step (2), iterate, and estimate the state at subsequent times.

[0094] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps:

[0095] (1) According to the target tracking system, a discrete linear system or a nonlinear system is established, a suitable target tracking motion model is selected for the system, and the system is initialized;

[0096] (2) Predict the state of the target at the next moment through the state equation to obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement based on the measurement equation and the predicted value of the state;

[0097] (3) Input the sensor's measurement data at the next moment, and obtain new information based on the measurement data and the predicted value of the measurement; and obtain the measurement gain term based on the new information according to the smooth variable structure filter;

[0098] (4) Obtain the coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value;

[0099] (5) Obtain the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and output the state;

[0100] (6) After outputting the state, return to step (2), iterate, and estimate the state at subsequent times.

[0101] Another object of the present invention is to provide an information data processing terminal, which is used to implement the target tracking system.

[0102] Another object of the present invention is to provide an application of the target tracking system in the fields of aerospace, radar tracking and navigation positioning.

[0103] Combined with all the above technical solutions, the advantages and positive effects of the present invention are as follows: the target tracking method provided by the present invention is based on the basic principle of the smooth variable structure filter, and uses its covariance to obtain the state without measurement value. Through different methods, the system can be applied to linear and nonlinear Gaussian and non-Gaussian systems. The present invention estimates the state without measurement value with high accuracy and good system robustness, breaking through the limitations of the original method. It can be applied to various types of state estimation systems like Kalman filters, greatly expanding the scope of application of the method, having theoretical and engineering value, and can be applied to aerospace, radar tracking, navigation and positioning and other fields.

[0104] The purpose of the present invention is to propose a new method to estimate the state without measurement value, and at the same time break through the limitation of traditional method that the measurement equation must be in a linear system, expand the application scope of this type of method, and enable it to be applied to linear and nonlinear, Gaussian and non-Gaussian systems.

[0105] The present invention proposes a new target tracking method using a smooth variable structure filter. Compared with the traditional method, the present invention has good scalability and can be applied to linear and nonlinear systems. In combination with different nonlinear processing methods, different types of state estimators can be obtained. The present invention is easy to implement and has good convergence. Compared with the traditional method of constructing pseudo-measurements and using low-order estimators to estimate states without measurement values, the present invention does not need to set smooth bounded layer width parameters for states without measurement values ​​when using the smooth variable structure filter theory, making the present invention more suitable for engineering practice. At the same time, the present invention has good robustness and estimation accuracy and can be applied to linear and nonlinear, Gaussian and non-Gaussian systems. In maneuvering target tracking, the method proposed by the present invention is more robust, and its estimation accuracy will not be significantly reduced when the target is maneuvering like the traditional method, and the tracking accuracy is higher.

[0106] Since the present invention improves the defect of the traditional smooth variable structure filter in the state dimension is greater than the measurement dimension estimation system, and at the same time makes up for the defect that the traditional method requires the measurement equation to be a linear equation. Therefore, the method proposed in the present invention enables the smooth variable structure filter to have good expansibility like the Kalman filter, and different types of methods can be extended based on the present invention, which will promote the extension of the present invention in theory, and different methods can also be selected according to different actual engineering needs for improvement, which will greatly promote the application of the present invention in the fields of signal processing, control engineering, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0108] Figure 1 It is a flow chart of a target tracking method provided by an embodiment of the present invention.

[0109] Figure 2 It is a schematic diagram of a target tracking method provided by an embodiment of the present invention.

[0110] Figure 3 is a structural block diagram of a target tracking system provided by an embodiment of the present invention;

[0111] In the figure: 1. System initialization module; 2. Prediction value calculation module; 3. Measurement new information acquisition module; 4. Measurement gain item acquisition module; 5. Coefficient calculation module; 6. State estimation result acquisition module; 7. Iterative calculation module.

[0112] Figure 4 It is a trajectory diagram of a simulation result of Example 1 provided in an embodiment of the present invention, including a real trajectory, measurement points, and trajectories processed by different filtering methods.

[0113] Figure 5 It is a schematic diagram of the absolute position error (m) of the estimation results of different methods on the x and y axes of Example 1 provided by an embodiment of the present invention.

[0114] Figure 6 It is a schematic diagram of the absolute speed error (m / s) of the estimation results of different methods on the x and y axes of Example 1 provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0115] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0116] In view of the problems existing in the prior art, the present invention provides a target tracking method, system, device, medium and data processing terminal, which are described in detail below with reference to the accompanying drawings.

[0117] like Figure 1 As shown, the target tracking method provided by the embodiment of the present invention includes the following steps:

[0118] S101, according to the target tracking system, a discrete linear system or a nonlinear system is established, a suitable target tracking motion model is selected for the system, and the system is initialized;

[0119] S102, predicting the state of the target at the next moment through the state equation to obtain the predicted value of the state at the next moment, and calculating the predicted value of the measurement according to the measurement equation and the predicted value of the state;

[0120] S103, inputting the measurement data of the sensor at the next moment, and obtaining new information based on the measurement data and the predicted value of the measurement; and obtaining the measurement gain term based on the new information according to the smooth variable structure filter;

[0121] S104, obtaining coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value;

[0122] S105, obtaining a final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and outputting the state;

[0123] S106, after outputting the state, returns to S102, iterates, and estimates the state at subsequent times.

[0124] The principle diagram of the target tracking method provided by the embodiment of the present invention is as follows Figure 2 shown.

[0125] like Figure 3 As shown, the target tracking system provided by the embodiment of the present invention includes:

[0126] The system initialization module 1 is used to establish a discrete linear system or a nonlinear system according to the target tracking system, select a suitable target tracking motion model for the system, and initialize the system;

[0127] Prediction value calculation module 2, used to predict the state of the target at the next moment through the state equation, obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement according to the measurement equation and the predicted value of the state;

[0128] The measurement new information obtaining module 3 is used to input the measurement data of the sensor at the next moment, and obtain the new information based on the measurement data and the predicted value of the measurement;

[0129] The measurement gain item obtaining module 4 is used to obtain the measurement gain item according to the smooth variable structure filter through the new information;

[0130] A coefficient calculation module 5, used to obtain the coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value;

[0131] The state estimation result acquisition module 6 is used to obtain the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and output the state;

[0132] The iterative calculation module 7 is used to return to the prediction value calculation module after outputting the state, perform iteration, and calculate the state estimation at the subsequent time. The technical solution of the present invention is further described below in conjunction with specific embodiments.

[0133] Example 1

[0134] The purpose of the present invention is to propose a new method to estimate the state without measurement value, and at the same time break through the limitation of traditional methods that the measurement equation must be in a linear system, expand the application scope of this type of method, and enable it to be applied to linear and nonlinear, Gaussian and non-Gaussian systems.

[0135] The target tracking method based on the smooth variable structure filter provided by the present invention comprises the following steps:

[0136] S1, according to the tracking target system, establish a discrete linear system or a nonlinear system, select a suitable target tracking motion model for the system, and initialize the system;

[0137] S2, predict the state of the target at the next moment through the state equation, obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement based on the measurement equation and the predicted value of the state;

[0138] S3, input the sensor's measurement data at the next moment, and obtain new information based on the measurement data and the predicted value of the measurement;

[0139] S4, using the new information, calculates the measurement gain term according to the smooth variable structure filter;

[0140] S5, obtaining the coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value;

[0141] S6, obtain the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term. Output state, which includes the position, velocity, acceleration, etc. of the target;

[0142] Return to S2, iterate, and calculate the state estimate at subsequent times.

[0143] The present invention can be embodied in the following steps:

[0144] S1 establishes discrete system equations according to the target tracking system, establishes appropriate state equations (1) and measurement equations (2) according to the target tracking motion model, and initializes the system.

[0145] The established system can be expressed as:

[0146] x k+1 =f(x k )+w k (1)

[0147] z k+1 =h(x k+1 )+υ k+1 (2)

[0148] The parameters in the formula are defined as follows: k represents the state at time k, f represents the relationship from the state at time k to the next k+1 time, w k represents process noise, h represents the corresponding relationship between state and measurement, z k+1 represents the measurement at time k+1, υ k+1 represents the measurement noise;

[0149] S2: Predict the next moment through the state equation and the estimated value of the target at the current moment, obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement based on the measurement equation and the predicted value of the state.

[0150] The predicted value of the next moment state and the predicted value of the measurement It can be calculated using the following formulas (3) and (4):

[0151]

[0152]

[0153] S3: Input the sensor's measurement data at the next moment, and obtain the new information based on the measurement data and the predicted value of the measurement. k+1|k It can be calculated by formula (5):

[0154]

[0155] S4: Based on the new information, the gain term corresponding to the measurement is obtained according to the basic theory of smooth variable structure filter. The gain term corresponding to the measurement can be calculated by formula (6):

[0156]

[0157] The parameters in the formula are defined as follows: |e| represents the absolute value of e, γ represents the convergence rate, and its value is generally 0≤γ<1. represents the multiplication of corresponding elements of the matrix; sat represents the saturation function, which can be expressed by formula (7):

[0158]

[0159] ψ represents the width of the smooth bounded layer;

[0160] S5: Obtain the coefficients corresponding to each state and measurement gain term through the covariance of state and measurement and the covariance of measurement prediction value. It is necessary to first obtain the covariance P of state and measurement through the system equation xz , and the covariance P of the measured predicted value zz , and then find the coefficient C corresponding to the state and measurement gain terms k+1 .

[0161] The final state estimation value obtained in S6 is It can be calculated by formula (8):

[0162]

[0163] Return to S2, iterate, and estimate the state at subsequent moments until the end.

[0164] Example 2

[0165] When the target tracking system is a linear system, the established linear system equation can be expressed as:

[0166] x k+1 =Fx k +w k (9)

[0167] z k+1 =Hx k+1 +υ k+1 (10)

[0168] The parameters in the formula are defined as follows: k represents the state at time k, F represents the relationship from the state k to the next time k+1, and H represents the correspondence between the state and the measurement. The estimated value of the state of the target at the current moment is predicted by the state equation to obtain the predicted value of the state at the next moment, and the predicted value of the measurement is calculated based on the measurement equation and the predicted value of the state.

[0169] The predicted value of the state and the predicted value of the measurement They can be calculated using the following formulas:

[0170]

[0171]

[0172] State prediction value The covariance P k+1|k It can be calculated according to formula (13):

[0173] P k+1|k =FP k|k F T +Q k (13)

[0174] F T is the transpose of the matrix F, P k|k represents the covariance of the state at time k, Q k represents the process noise w k The covariance of .

[0175] Predicted value of measurement The covariance P zz It can be calculated by formula (14):

[0176] P zz =HP k+1|k H T (14)

[0177] The covariance P between the state and the measurement xz It can be calculated by formula (15):

[0178] P xz =P k+1|k H T (15)

[0179] Input the sensor's measurement data at the next moment, and obtain the new information based on the measurement data and the predicted value of the measurement. k+1|k It can be calculated by formula (16):

[0180]

[0181] Through the new information, the gain term corresponding to the measurement is calculated according to the smooth variable structure filter.

[0182] The corresponding gain term G is measured k+1 , which can be calculated by formula (17):

[0183]

[0184] The parameters in the formula are defined as follows: |e| represents the absolute value of e, γ represents the convergence rate, and its value is generally 0≤γ<1. represents the multiplication of the corresponding elements of the matrix, sat represents the saturation function, which is specifically the following formula (18):

[0185]

[0186] ψ represents the width of the smooth bounded layer;

[0187] Through the covariance of the state and the predicted covariance of the measurement, the coefficients corresponding to each state and the measurement gain term are obtained. It is necessary to first obtain the covariance P of the state measurement through the system equation xz , and the covariance P of the measured predicted value zz , and then find the coefficient C corresponding to the state and measurement gain terms k+1 .

[0188] C k+1 =P xz *(P zz ) -1 (19)

[0189] The final state estimate is It can be calculated by formula (20):

[0190]

[0191] calculate The covariance P k+1|k+1 and the a posteriori innovation e k+1|k+1 , output state, which includes the position, velocity, acceleration, etc. of the target, and iterate to update the next moment until the end.

[0192] Example 3

[0193] When the target tracking system of the present invention is a nonlinear system, the model of the established nonlinear system can be expressed as:

[0194] x k+1 =f(x k )+w k (twenty one)

[0195] z k+1 =h(x k+1 )+υ k+1 (twenty two)

[0196] Perform linear approximation on the nonlinear system, such as finding the first-order Taylor truncation of the system, and finding the Jacobian matrices of f and h as follows:

[0197]

[0198]

[0199] Covariance P of the state prediction value k+1|k It can be calculated according to the following formula (25):

[0200]

[0201] Then the covariance P between the state and the measurement is xz and the covariance P of the measured predicted valuezz It can be calculated by the following formulas (26) and (27) respectively:

[0202]

[0203]

[0204] C corresponding to the state and measurement k+1 It can be calculated by the following formula (28):

[0205] C k+1 =P xz *(P zz ) -1 (28)

[0206] The corresponding gain term G is measured k+1 , which can be calculated by formula (29):

[0207]

[0208] The final state estimate is It can be calculated from (30):

[0209]

[0210] calculate The covariance P k+1|k+1 and the a posteriori innovation e k+1|k+1 , output state, which includes the position, velocity, acceleration, etc. of the target, and iterate to update the next moment until the end.

[0211] Example 4

[0212] When the target tracking system of the present invention is a nonlinear system, the established nonlinear system can be expressed as:

[0213] x k+1 =f(x k )+w k (31)

[0214] z k+1 =h(x k+1 )+υ k+1 (32)

[0215] Perform unscented transformation on nonlinear systems, according to the state and P k|k Get 2n+1 sampling points, sampling point X i,k|k And its corresponding weight W i It can be obtained by the following formulas (33) and (34):

[0216]

[0217]

[0218] Where κ is a scale parameter, is (n+κ)P k|k The i-th row or i-th column of the RMS matrix, n is the dimension of the state; the sampling point X i,k|k The predicted value of the corresponding state It can be calculated by formula (35):

[0219]

[0220] State prediction estimate and the covariance P of the state prediction value k+1|k It can be obtained from formulas (36) and (37):

[0221]

[0222]

[0223] Predicted value of measurement The predicted value of the sampling point And the new information of measurement k+1|k From (38), (39) and (40), we can obtain:

[0224]

[0225]

[0226]

[0227] Then the covariance P between the state and the measurement is xz and the covariance P of the measured predicted value zz It can be calculated by the following formulas (41) and (42) respectively:

[0228]

[0229]

[0230] Then the coefficient C corresponding to the state and measurement k+1 It can be calculated by formula (43):

[0231] C k+1 =P xz *(P zz ) -1 (43)

[0232] The corresponding gain term G is measured k+1, which can be calculated by formula (44):

[0233]

[0234] The final state estimate is It can be calculated by formula (45):

[0235]

[0236] calculate The covariance P k+1|k+1 and the a posteriori innovation e k+1|k+1 , output state, which includes the position, velocity, acceleration, etc. of the target, and iterate to update the next moment until the end.

[0237] The following is an evaluation of the performance of the target tracking method based on the smooth variable structure filter in the present invention by using specific example 1 of embodiment 1 in a linear system in combination with the line, to demonstrate the performance superiority of the method of the present invention.

[0238] Example 1: When tracking a maneuvering target, the Kalman filter (KF) method, the method based on a smooth variable structure filter with a low-order Luenberger estimator (method 1), the method based on a smooth variable structure filter with a pseudo-measurement value (method 2) and the method proposed in the present invention are used to perform processing, and the performance of the four methods is compared. In all methods, the state model uses a linear uniform velocity model F as shown below (46):

[0239]

[0240] In the above formula, T represents the sampling rate, the simulation environment is, and the initial state of the aircraft is x 0 =[-25000m,30m / s,-10000m,20m / s] T The target first flies at a constant speed for 60 seconds, then turns at a constant speed, flies at a turning rate of 6° / s for 30 seconds, then flies at a constant speed for 60 seconds, then maneuvers at a turning rate of -3° / s for 50 seconds, and finally flies at a constant speed for 100 seconds until it reaches the destination. The flight trajectory is as follows Figure 4 shown.

[0241] The radar provides position measurement, and the measurement error after coordinate system conversion obeys a Gaussian distribution with a mean of 0 and a standard deviation of 100m; the radar sampling rate T is 1s. The simulation performed 500 Monte Carlo operations, and gave the initial state covariance matrix P and system noise matrix Q of different methods, the measurement noise matrix R, the smooth boundary layer width ψ, the convergence rate γ and other parameters, as shown in Table 1.

[0242] Table 1 Variable definition table

[0243]

[0244] Figure 4 The trajectory of a simulation result is displayed, including the real trajectory, the measurement point, and the trajectory obtained after processing with different filters. Figure 4 It shows that all four methods can smooth the trajectory, and the method proposed in the present invention can obtain a more accurate trajectory. Figures 5-6 The absolute error results of the position and velocity estimation of the method proposed in the present invention and other methods are shown respectively. Table 2 is the cumulative root mean square error ARMSE (500 Monte Carlo) between the estimated results and the true state in position or velocity.

[0245] Table 2 Estimated cumulative root mean square error ARMSE (500 Monte Carlo).

[0246]

[0247] Combination Figures 4 to 6 and all experimental results in Table 2. The proposed method shows the best performance in terms of accuracy (i.e., ARMSE). It is worth noting that all filter algorithm modeling uses only the uniform velocity model, while in the simulation, the target flight trajectory is uniform motion and coordinated turning motion. When the filter model is the same as the motion model, i.e., uniform velocity, the Kalman filter has a good performance. However, when the motion changes from uniform velocity to coordinated turning motion, modeling errors occur, which significantly degrades the performance of the Kalman filter, such as Figures 4 to 6 As shown. Since the Kalman filter requires the filter model to be consistent with the actual motion model, when the models do not match, the accuracy of the Kalman filter's estimated speed and position will affect each other and decrease at the same time. Compared with the case without modeling error, the method proposed in the present invention reduces the accuracy of estimated speed in the presence of modeling error, but the reduction in position estimation accuracy is not as rapid as the position error in the Kalman filter. In addition, from Figure 5 and Figure 6 It can be seen that when the maneuver ends, for example, after 90 seconds and 200 seconds, the Kalman filter takes a long time to recover to a good filtering state, while the recovery time of the method proposed in the present invention is shorter. As shown in Table 2, the position ARMSE of the Kalman filter is at least twice that of the proposed method. In addition, the speed accuracy of the method proposed in the present invention is improved by 32% and 20% on the x-axis and y-axis respectively compared with the Kalman filter. The accuracy of the traditional methods 1 and 2 is lower than that of the method described in the present invention. In summary, the comprehensive performance of the method proposed in the present invention is the best among the above methods.

[0248] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When the use is implemented in whole or in part in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) mode) to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0249] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A target tracking method, characterized in that: The target tracking method comprises the following steps: Step 1: According to the tracking target system, a discrete linear system or a nonlinear system is established, a suitable target tracking motion model is selected for the system, and the system is initialized; Step 2: predict the state of the target at the next moment through the state equation to obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement based on the measurement equation and the predicted value of the state; Step 3: Input the sensor's measurement data at the next moment, and obtain new information based on the measurement data and the predicted value of the measurement; Step 4, using the new information, calculate the measurement gain term according to the smooth variable structure filter; Step 5, obtaining the coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value; Step 6, calculating the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and outputting the state; Step 7: After outputting the state, return to step 2, iterate, and estimate the state at subsequent moments; In step 1, the discrete system equation is: x k+1 =f(x k )+w k ; z k+1 =h(x k+1 )+υ k+1 ; In the formula, x k represents the state at time k, f represents the transition relationship from state k to the next time k+1, and w k represents process noise, h represents the corresponding relationship between state and measurement, z k+1 represents the measurement at time k+1, υ k+1 represents the measurement noise; In step 2, the predicted value of the state and the predicted value of the measurement Calculate using the following formulas: In step 3, the measured new information e k+1|k Calculated by the following formula: In the formula, z k+1 represents the measurement value measured by the system through the sensor at time k+1; In step 4, the gain term G corresponding to the measurement k+1 Calculated by the following formula: In the formula, |e| means the absolute value of e; γ means the convergence rate, which is 0≤γ<1; represents the multiplication of corresponding elements of the matrix; sat represents the saturation function; ψ represents the width of the smooth bounded layer; In step 5, the covariance P of the state and measurement is obtained through the system equation xz and the covariance P of the measured predicted value zz , and then find the coefficient C corresponding to the state and measurement gain terms k+1 for: C k+1 =P xz *(P zz ) -1 ; In step 6, the final state estimate is Calculated by the following formula: In step 5, obtaining the coefficients corresponding to the state and measurement gain terms includes: There are different methods for different systems. When the system equation in step 1 is a linear equation, the covariance P of the state prediction value in step 2 k+1|k Calculated according to the following formula: P k+1|k =FP k|k F T +Q k ; In the formula, the state transfer matrix F is the specific value of f, F T is the transpose of the matrix F, P k|k represents the covariance of the state at time k, Q k represents the process noise w k The covariance of Then the covariance P between the state and the measurement is xz Calculated by the following formula: P xz =P k+1|k H T ; Where H is the conversion matrix between state and measurement, which is the specific value of h in the linear system. T is the transpose of H; then the covariance P of the measured predicted value zz Calculated by the following formula: P zz =HP k+1|k H T ; Then the coefficient C corresponding to the state and measurement gain term is k+1 Calculated by the following formula: C k+1 =P xz *(P zz ) -1 。 2. The target tracking method according to claim 1, characterized in that: In step 5, obtaining the coefficients corresponding to the state and measurement gain terms includes: When the system equation in step 1 is a nonlinear equation, the system equation needs to be processed when calculating the covariance, such as performing linear approximation on the nonlinear system and calculating the first-order Taylor expansion of the system equation. The Jacobian matrices of f and h are calculated using the following formula: The covariance P of the state prediction value in step 2 k+1|k Calculated according to the following formula: P k+1|k =F k P k|k F k T +Q k ; Then the covariance P between the state and the measurement is xz and the covariance P of the measured predicted value zz Calculated according to the following formulas: Then the coefficient C corresponding to the state and measurement gain term is k+1 Calculated by the following formula: C k+1 =P xz *(P zz ) -1 。 3. The target tracking method according to claim 1, characterized in that: In step 5, obtaining the coefficients corresponding to the state and measurement gain terms includes: When the system equation in step 1 is a nonlinear equation, the method of untraceable transformation is used to calculate its mean and covariance; According to the status and P k|k Get 2n+1 sampling points, sampling point X i,k|k And its corresponding weight W i The following formulas are used to calculate: Where κ is a scale parameter, is (n+κ)P k|k The i-th row or i-th column of the RMS matrix, n is the dimension of the state, and i is the sampling point; Sampling point X i,k|k The predicted value of the corresponding state Calculated by the following formula: Then the state prediction estimate in step 2 is and the state prediction covariance P k+1|k It is obtained by the following formula: Then the predicted value of the measurement in step 2 is Predicted value through sampling points It is obtained by the following formula: In the formula, It is the sampling point The corresponding measured predicted value, W i is its weight; Then the covariance P between the state and the measurement is xz and the covariance P of the measured predicted value zz Calculate using the following formulas: Then the coefficient C corresponding to the state and measurement gain term is k+1 Calculated by the following formula: C k+1 =P xz *(P zz ) -1 。 4. A target tracking system for implementing the target tracking method according to any one of claims 1 to 3, characterized in that: The target tracking system comprises: The system initialization module is used to establish a discrete linear system or a nonlinear system according to the target tracking system, select a suitable target tracking motion model for the system, and initialize the system; The prediction value calculation module is used to predict the state of the target at the next moment through the state equation, obtain the predicted value of the state at the next moment, and calculate the predicted value of the measurement according to the measurement equation and the predicted value of the state; The measurement new information obtaining module is used to obtain new information by inputting the measurement data of the sensor at the next moment and according to the measurement data and the predicted value of the measurement; A measurement gain item obtaining module is used to obtain the measurement gain item according to the smooth variable structure filter through the new information; A coefficient calculation module, used to obtain the coefficients corresponding to the state and measurement gain terms through the covariance of the state and measurement and the covariance of the measurement prediction value; A state estimation result acquisition module is used to calculate the final state estimation result according to the predicted value of the state, the measurement gain term and the coefficient corresponding to the state and the measurement gain term, and output the state; The iterative calculation module is used to return to the prediction value calculation module after outputting the state, iterate, and estimate the state at the subsequent time.

5. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the target tracking system as claimed in claim 4.

6. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the target tracking method according to any one of claims 1 to 3.

7. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the target tracking method according to any one of claims 1 to 3.

Images