A target tracking method based on IMM-STUKF information fusion algorithm

Abstract

The present invention discloses a target tracking method based on the IMM-STUKF information fusion algorithm. The strong tracking filter is combined with the UKF to propose an improved STUKF filtering algorithm. The system has better adaptability and robustness by introducing a fading factor, and the influence of system model uncertainty on the estimation error is effectively reduced. Then the interactive multi-model algorithm IMM is introduced, and the IMM-STUKF filtering algorithm is proposed for target tracking, which further improves the tracking accuracy of the target under various maneuvers.

Description

Technical Field

[0001] The invention relates to the field of target tracking, in particular to a target tracking method based on an IMM-STUKF information fusion algorithm. Background Art

[0002] At present, Kalman filtering has been widely used as a filtering algorithm with excellent performance in the field of target tracking, but it also has certain shortcomings. The implementation of the filter needs to be based on a certain target tracking model, and Kalman filtering can only track a single maneuvering model. If the target only performs a single maneuver, it can keep tracking the target. If faced with actual mission scenarios, the targets often have the characteristics of diverse maneuvers and strong model uncertainty (model simplification, excessive system noise, and model parameter changes), it is difficult for Kalman filtering to achieve the actual target state estimation effect. At the same time, Kalman filtering also requires that the system must be a linear system. For nonlinear systems or when the noise model does not conform to the Gaussian distribution, Kalman filtering is not applicable and cannot provide accurate state estimates. Summary of the invention

[0003] The technical problem solved by the present invention is: in view of the characteristics of diverse target maneuvers and strong model uncertainty, namely model simplification, excessive system noise and model parameter changes, a target tracking method based on the IMM-STUKF information fusion algorithm is provided to obtain a more accurate target state estimation value and improve the tracking accuracy of the target.

[0004] The technical solution of the present invention is: a target tracking method based on the IMM-STUKF information fusion algorithm, comprising:

[0005] The strong tracking filter STF is combined with the unscented Kalman filter UKF, and a suboptimal fading factor is introduced into the unscented Kalman filter to make the filter residual sequences orthogonal to each other, thus obtaining the STUKF filtering algorithm.

[0006] The STUKF filtering algorithm and the interactive multi-model algorithm IMM are fused, and the IMM-STUKF information fusion algorithm is used to track the target to obtain the target state estimation, reduce the influence of uncertainty on the estimation error, and improve the tracking accuracy of the target under various maneuvers.

[0007] Furthermore, a suboptimal fading factor is introduced into the unscented Kalman filter UKF to keep the filter residual sequences orthogonal to each other. The specific method is as follows:

[0008] Introducing the fading factor λ into the unscented Kalman filter UKF k , redefine the original state prediction covariance, observation prediction covariance, state and observation mutual covariance in the unscented Kalman filter, and the updated expressions are as follows:

[0009]

[0010] Where P k|k-1 is the state prediction covariance matrix, P zz,k|k-1 is the observation-prediction covariance matrix, P xz,k|k-1 is the state and observation covariance matrix, n is the state dimension, W i p is the weighted coefficient of the variance of each sigma point after transformation, χ i,k / k-1 is the predicted sigma point, is the predicted value of the state variable, Q k is the process noise covariance matrix, is the predicted measured sigma point, is the predicted value of the observed quantity, R k is the observation noise covariance matrix;

[0011] In the formula, the fading factor λ k The expression is as follows:

[0012]

[0013] In the formula, tr[·] represents the trace of the matrix, N k ,M k The definition is as follows:

[0014] N k =C 0,k -βR k

[0015]

[0016] Where β is the weakening factor, which is the same as the definition of UKF filter; C 0,k is the covariance matrix of the residuals.

[0017] Furthermore, the covariance matrix C of the residual 0,k , calculated by the following formula:

[0018]

[0019] In the formula, ε k is the residual at time k, and ρ is the forgetting factor.

[0020] Furthermore, the Kalman filter in the interactive multiple model algorithm IMM is replaced by the STUKF filtering algorithm to obtain the IMM-STUKF information fusion algorithm.

[0021] The present invention also provides a target tracking system based on the IMM-STUKF information fusion algorithm, comprising:

[0022] The STUKF filtering module is used to combine the strong tracking filter STF with the unscented Kalman filter UKF, introduce a suboptimal fading factor into the unscented Kalman filter, and keep the filter residual sequences orthogonal to each other, thus obtaining the STUKF filtering algorithm;

[0023] IMM-STUKF information fusion module, used to call the STUKF filter module and fuse it with the interactive multi-model algorithm IMM;

[0024] The target tracking module is used to track the target using the IMM-STUKF information fusion module to obtain the target state estimation, reduce the impact of uncertainty on the estimation error, and improve the tracking accuracy of the target under various maneuvers.

[0025] Further, in the STUKF filtering module,

[0026] Introducing the fading factor λ into the unscented Kalman filter UKF k , redefine the original state prediction covariance, observation prediction covariance, state and observation mutual covariance in the unscented Kalman filter, and the updated expressions are as follows:

[0027]

[0028] Where P k|k-1 is the state prediction covariance matrix, P zz,k|k-1 is the observation-prediction covariance matrix, P xz,k|k-1 is the state and observation covariance matrix, n is the state dimension, W i p is the weighted coefficient of the variance of each sigma point after transformation, χ i,k / k-1 is the predicted sigma point, is the predicted value of the state variable, Q k is the process noise covariance matrix, is the predicted measured sigma point, is the predicted value of the observed quantity, R k is the observation noise covariance matrix;

[0029] In the formula, the fading factor λ k The expression is as follows:

[0030]

[0031] In the formula, tr[·] represents the trace of the matrix, N k ,M k The definition is as follows:

[0032] N k =C 0,k -βR k

[0033]

[0034] Where β is the weakening factor, which is the same as the definition of UKF filter; C 0,k is the covariance matrix of the residuals.

[0035] Further, in the STUKF filtering module,

[0036] The covariance matrix C of the residual 0,k , calculated by the following formula:

[0037]

[0038] In the formula, ε k is the residual at time k, and ρ is the forgetting factor.

[0039] Furthermore, in the IMM-STUKF information fusion module, the fusion method is: replacing the Kalman filter in the interactive multiple model algorithm IMM with the STUKF filtering algorithm to obtain the IMM-STUKF information fusion algorithm.

[0040] The present invention also provides a computer program product, which, when executed by a processor, implements the steps of the above method.

[0041] The present invention also provides a computer-readable storage medium, and when the computer program in the medium is executed by a processor, the steps of the above-mentioned method are implemented.

[0042] The advantages of the present invention compared with the prior art are:

[0043] (1) The present invention combines the strong tracking filter with the UKF to propose an improved STUKF filtering algorithm. By introducing the fading factor, the system has better adaptability and robustness, and effectively reduces the impact of system model uncertainty on the estimation error. Then the interactive multi-model algorithm IMM is introduced to obtain the IMM-STUKF filtering algorithm for target tracking, which further improves the tracking accuracy of the target under various maneuvers.

[0044] (2) The present invention aims to solve the problem that the UKF algorithm needs to be predicted by the original system model and has a high degree of dependence on the model. By taking advantage of the strong tracking filter's strong robustness to the uncertainty of the system model, the adaptability of the strong tracking theory is combined with the nonlinear high approximation of the UKF algorithm, and the STUKF filtering algorithm is obtained. This ensures that there is still strong tracking capability when the system model is uncertain, thereby improving the target tracking accuracy.

[0045] (3) To address the problem that the above-mentioned algorithm can only track a single maneuvering model, the present invention introduces an interactive multi-model algorithm to obtain the IMM-STUKF filtering algorithm, automatically adjusts the model probability according to the target motion state, and simultaneously calculates multiple filtering models to obtain the target state estimation value, thereby improving the tracking capability of the target under multiple maneuvering modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A comparison chart of EKF and UKF for nonlinear system state estimation methods;

[0047] Figure 2 It is the flow chart of IMM algorithm;

[0048] Figure 3 The motion trajectory of the two aircraft cooperative tasks under the dual infrared detection system in the embodiment of the present invention and the target tracking trajectory under the fusion algorithm of the present invention;

[0049] Figure 4 It is a probability change curve diagram of each model in the IMM-STUKF algorithm under the dual infrared detection system in the embodiment of the present invention;

[0050] Figure 5 This is a comparison diagram of the root mean square error of different algorithms in the dual infrared detection system in an embodiment of the present invention;

[0051] Figure 6 A comparison diagram of the root mean square error of speeds of different algorithms in a dual infrared detection system in an embodiment of the present invention;

[0052] Figure 7 The motion trajectory of the two aircraft cooperative tasks under the infrared + radar system in the embodiment of the present invention and the target tracking trajectory under the fusion algorithm of the present invention;

[0053] Figure 8 This is a comparison diagram of the root mean square error of different algorithms in the infrared + radar detection system in an embodiment of the present invention;

[0054] Fig. 9 This is a comparison chart of the root mean square error of speed of different algorithms in the infrared + radar detection system in an embodiment of the present invention.

[0055] Fig.10 The main flow chart of the method of the present invention is shown in FIG. DETAILED DESCRIPTION

[0056] In order to better understand the technical solution of the present invention, the specific implementation mode of the present invention is described in detail below with reference to the accompanying drawings.

[0057] The process of the present invention is as follows Fig.10As shown in the figure, the strong tracking filter STK is first combined with the unscented Kalman filter UKF, and a STUKF filtering algorithm is proposed. By introducing the fading factor, the system has better adaptability and robustness, and effectively reduces the influence of system model uncertainty on the estimation error.

[0058] Then, the interactive multi-model algorithm IMM is introduced to improve the tracking accuracy of the target under various maneuvers. The IMM-STUKF information fusion algorithm is proposed. The IMM-STUKF information fusion algorithm is used to track the target and obtain the target state estimation.

[0059] Finally, this embodiment carries out simulation verification and analysis on the algorithm proposed by the present invention.

[0060] The specific implementation plan is as follows:

[0061] (1) STUKF filter algorithm design

[0062] At present, a variety of improved nonlinear filtering algorithms have been proposed based on Kalman filtering. In practical applications, there are many common solutions for nonlinear system filtering. Among them, the extended Kalman filter (EKF) is to linearize it, and the unscented Kalman filter (UKF) does not need to linearize the nonlinear function, but approximates the probability density distribution of the nonlinear function through the unscented transformation (UT) method. Its basic idea is to select some sample points that can represent the statistical characteristics of the mean and variance of the state variable, and perform nonlinear transformation on them through the state equation and observation equation respectively, and use the transformed samples to calculate their mean and variance to achieve the prediction of the target state. Figure 1 Comparison of EKF and UKF state estimation methods for nonlinear systems.

[0063] According to UT transformation, the specific formula of UKF filtering algorithm is as follows:

[0064] First, 2n+1 sigma points are constructed to approximate the random distribution characteristics of the state variables. n is the state dimension of the model. The sigma points are selected using the following formula:

[0065]

[0066] Where λ = α 2 (n+k)-n represents the proportional factor, α determines the distribution of sigma points around the mean, and when the dimension of the state variable is greater than 1, k=3-n. (n+λ)P x The square root of the matrix can be obtained by Cholesky decomposition, Represents the i-th row or i-th column of a matrix.

[0067] Assuming f is a nonlinear function of the state variable x, the function point set y after the nonlinear transformation of the sigma point is i It is expressed as:

[0068] y i =f(x i )(i=1,2,...,2n) (2)

[0069] The mean and variance of the new sample point y after transformation can be calculated using the following formula:

[0070]

[0071] Among them, W i m , W i p Respectively represent the weighted coefficients of the mean and variance of each sigma point after transformation, so it satisfies the condition β is a parameter related to the prior distribution of the state, and is generally taken as 2 in the case of Gaussian distribution.

[0072] Applying UT transform to filtering algorithm to solve the nonlinear problem of the system is UKF. The specific calculation steps are as follows:

[0073] 1) Parameter initialization, calculate the mean and variance of the initial state vector:

[0074]

[0075] 2) When k∈{1,2,3,...}, 2n+1 sigma points are generated:

[0076]

[0077] 3) Time update

[0078] Substitute each sigma point into the state equation to calculate and obtain the predicted value of the state variable and the covariance matrix P k|k-1 as follows:

[0079]

[0080] 4) Measurement update

[0081] Regenerate 2n+1 sigma points using state prediction values.

[0082]

[0083] Substitute the new sigma point into the observation equation and calculate the predicted value of the observation Variance matrix P zz,k|k-1and the covariance matrix P of the state and observation xz,k|k-1 .

[0084]

[0085] Correct the predicted value through the observed value and update the estimated value of the state And the variance matrix P k .

[0086]

[0087] The UKF algorithm does not need to linearize the system when estimating the target state, but uses the original system model to perform UT transformation for direct prediction. While improving the accuracy, it also increases the algorithm's dependence on the model. At the same time, the target's maneuvering range is large and the maneuvering mode is complex, so it is very difficult to accurately model it. In addition, the measurement noise is large, and the uncertainty of the system model will cause the filtering accuracy of the state estimation using UKF to decrease, and even cause the system to diverge. Therefore, the present invention proposes a strong tracking UKF filtering algorithm in combination with a strong tracking filter. By introducing a suboptimal fading factor method through a strong tracking filter, the filter residual sequence is forced to remain orthogonal to each other, ensuring that it still has a strong tracking ability when the system model is uncertain, realizing the combination of the STF and UKF algorithms, improving the filtering accuracy of the algorithm, and making the system have better adaptability and robustness. The specific implementation method is as follows:

[0088] When the filter works normally, the output residual sequence is orthogonal, but in actual applications, the uncertainty of the system model will cause the filter output residual sequence to be non-orthogonal, and the filtering error will increase. STF adjusts the gain matrix K in real time by introducing a fading factor. k , so that the following two conditions are met:

[0089]

[0090] In the formula, min represents the minimum expected value on the left side of the equal sign. is the estimated value of the target state at time k, ε k is the residual at time k. Formula (10) represents the performance index of the filter, which minimizes the variance of the state estimation error. Formula (11) indicates that the residuals of the system at different times always remain orthogonal.

[0091] In order to realize the combination of STF and UKF and ensure that the condition of formula (11) holds, it is necessary to introduce a fading factor λ into the UKF algorithm. k , redefine the state prediction covariance, observation prediction covariance, state and observation mutual covariance, and the updated expressions are as follows:

[0092]

[0093] The fading factor λ in the formula k The expression is as follows:

[0094]

[0095] In the formula, tr[·] represents the trace of the matrix, N k ,M k The definition is as follows:

[0096] N k =C 0,k -βR k (16)

[0097]

[0098] The β in the formula is the weakening factor, which is the same as the definition of UKF filter. 0,k is the covariance matrix of the residuals, which can be estimated by the following formula:

[0099]

[0100] The ρ in the formula is the forgetting factor, which is usually taken as 0.95.

[0101] The UKF filtering algorithm is improved by using equations (12) to (18) and the fading factor λ is introduced. k It inherits the advantage of STF's strong robustness when the system model is uncertain, combines the adaptability of strong tracking theory with the nonlinear high approximation of UKF algorithm, and obtains the improved STUKF filtering algorithm, which reduces the estimation error of the filter and improves the tracking accuracy of the target.

[0102] (2) IMM-STUKF information fusion algorithm design

[0103] Only a single maneuvering model can be tracked. If the target only maneuvers in a single way, the target can be tracked. However, in actual scenarios, the target can maneuver in multiple ways. At this time, the above single model tracking algorithm cannot meet the actual needs. The interactive multiple model (IMM) algorithm is based on multiple Kalman filter models that simultaneously calculate the estimated value of the target state, assign a different maneuvering model to each filter, and the probability of each model will change with the change of the target maneuvering mode. Finally, the estimation results of different models are weighted and fused as the final estimate. This algorithm can well solve the target tracking problem under multiple maneuvering modes.

[0104] The interactive multi-model algorithm process is as follows Figure 2 As shown, assuming an IMM algorithm with r maneuver models, the calculation steps are as follows:

[0105] 1) Data input interaction

[0106] Calculate the mixing probability of model i and model j at the previous moment:

[0107]

[0108] P in the formula ij is the transition probability from model i to model j, is the normalized coefficient, expressed as:

[0109]

[0110] Then each model estimates the state and covariance after input interaction:

[0111]

[0112] 2) Kalman filter

[0113] Substitute the state estimate and covariance estimate after the previous interaction into each model for Kalman filter calculation, and update the state estimate of each model and covariance estimate P j (kk).

[0114] 3) Update model probability

[0115] Calculate the maximum likelihood function for each model:

[0116]

[0117] In the formula

[0118]

[0119] Then the model probability of each model is:

[0120]

[0121] Where c is a normalization constant, expressed as:

[0122]

[0123] 4) Output interaction

[0124] The model probability is used to perform weighted fusion of the estimated values ​​of each model to obtain the combined state estimate and covariance estimate.

[0125]

[0126] The IMM algorithm can automatically adjust according to the motion state of the target, assign different model probabilities to each model, and then perform weighted fusion of the estimated values, which enables it to maintain high tracking accuracy when facing targets that can maneuver in multiple ways, and is more in line with actual target tracking problems. Therefore, the present invention will introduce an interactive multi-model algorithm and replace the Kalman filter in the above step 2) with the STUKF filter algorithm proposed in the first section to obtain an improved IMM-STUKF information fusion algorithm.

[0127] Most of the current aircraft perform tasks through cluster collaboration, which greatly improves the system's capabilities. However, the detection information of a single detection system is very limited and the accuracy is insufficient. Therefore, by fusing the observation data of multiple detection systems, the data obtained is more accurate and the success rate of the mission is improved. Therefore, information fusion plays a key role in the process of collaborative detection. The common structures of current information fusion algorithms can be divided into distributed fusion and centralized fusion.

[0128] The present invention selects a centralized information fusion method, and performs centralized information fusion processing on the observation information of multiple detection systems obtained through the IMM-STUKF filtering algorithm, which effectively reduces the influence of system model uncertainty and various target maneuvers on target state estimation, improves the accuracy of target tracking, and realizes collaborative detection of multiple aircraft.

[0129] (3) Simulation test analysis

[0130] In order to further verify the feasibility of the target tracking method proposed in the present invention, simulation experiments were carried out on the target tracking capabilities of the dual infrared detection system and the infrared + radar detection system. Through 50 Monte Carlo simulations, the root mean square error of position and velocity was used as the target state estimation accuracy indicator.

[0131] 1) Simulation verification of dual infrared detection system

[0132] Assume that there are two aircrafts and both are equipped with infrared detection systems to track the target, so only angle information can be obtained. The angle measurement error of the infrared detection system is 0.05°. Then the observation equation of the cooperative detection system is:

[0133]

[0134] In order to reflect the algorithm capability proposed in the present invention, the IMM-STUKF filtering algorithm proposed in the present invention and the three filtering algorithms of CA-UKF, CT-UKF and IMM-UKF are compared in simulation experiments. CV, CT and CA are selected as the model sets of the IMM algorithm, which are called Model 1, Model 2 and Model 3 respectively. The target is located at an altitude of 25 kilometers and is flying at a fixed altitude. The initial position is (49000, 0), the speed is 6 Mach, the trajectory deviation angle is π, and the maximum normal overload is 2.5g. Assuming that the detection distance of the aircraft and the target detection system is close, when the aircraft enters the terminal guidance, the target will also find the aircraft. At this time, the target will maneuver to avoid, and the most common maneuvering method of the target is acceleration and turning. Therefore, it is assumed that the target moves at a uniform speed within 0 to 3s, performs a left turn maneuver with the maximum normal overload within 3 to 9s, and performs a right turn maneuver with the maximum normal overload within 9 to 15s. After 15 seconds, it performs uniform acceleration until the mission is completed. The initial position of aircraft 1 is (0, 3000), and the initial position of aircraft 2 is (0, -3000). The speed of both aircraft is Mach 3, the trajectory deviation angle is 0, and the maximum normal overload is 10g. The simulation period is set to 0.1ms.

[0135] The simulation results are as follows:

[0136] Figure 3 The motion trajectory of the two aircraft collaborative mission and the target tracking trajectory using the fusion algorithm of the present invention. It can be seen from the figure that the target state estimation value obtained by collaborative detection by the dual infrared detection system can well realize target tracking and make the aircraft mission successful.

[0137] Figure 4 This is the probability change curve of each model in the IMM-STUKF algorithm. In the first 3 seconds, when the target is moving at a constant speed, the probability of model 1 is about 0.8. When the target changes to a turning maneuver after 3 seconds, the probability of model 2 increases from 0.1 to 0.9. After 15 seconds, the target accelerates and the probability of model 3 gradually increases. The IMM-STUKF algorithm can quickly switch the models in the model set, select the correct model and assign a higher model probability to estimate the target state, thereby achieving target tracking in multiple maneuvering modes and improving the target tracking capability;

[0138] Figure 5 and Figure 6It can be found that the target state estimation error of the single model filters CA-UKF and CT-UKF is small when the maneuvering models are consistent, but the position error and velocity error are significantly larger than the IMM algorithm during the entire movement of the target, which further shows that the IMM algorithm has higher tracking accuracy when facing multiple maneuvers of the target; by comparing IMM-UKF and IMM-STUKF, it is found that the target state estimation accuracy of the improved IMM-STUKF is significantly better than that of IMM-UKF, and the overall curve is relatively smooth, without a large mutation due to changes in the target's motion mode.

[0139] In order to further illustrate the effect of the four algorithms on target tracking, the average value of the position and velocity root mean square error is calculated over time. The results are shown in the following table:

[0140] Table 1 Comparison of the position and velocity root mean square error of four algorithms

[0141]

[0142] According to the data in Table 1, it can be shown that when the present invention adopts the IMM-STUKF algorithm to estimate the target state, compared with CA-UKF, CT-UKF, and IMM-UKF under the same conditions, the position root mean square error is reduced by 80.92%, 71.11%, and 54.68%, respectively, and the speed root mean square error is reduced by 85.15%, 70.92%, and 61%, respectively. Through data comparison, it can be more clearly seen that the target state estimation error of the IMM-STUKF algorithm is smaller than that of other algorithms. Therefore, the centralized information fusion algorithm based on IMM-STUKF proposed by the present invention can effectively improve the target tracking accuracy, and the dual infrared detection system cooperative detection system using the information fusion algorithm can also complete the high-precision tracking of the target.

[0143] 2) Infrared + radar detection system collaborative detection

[0144] Since the infrared detection system can detect the angle information of the target, but cannot obtain the distance information of the target, while the radar detection system can detect the distance information of the target. Therefore, using the infrared and radar detection systems for collaborative detection can make full use of the high-precision angle information of the infrared detection system and the high-precision distance information of the radar detection system, and complement the measurement data of the two detection systems, thereby reducing the error of the detection system in estimating the target state and improving the target tracking accuracy. Therefore, we further assume that the two aircraft are equipped with infrared and radar detection systems respectively to conduct simulation experiments, and compare and analyze the collaborative detection method with the dual infrared detection system.

[0145] Assuming that the distance measurement error of the radar detection system is 20m and the angle measurement error is 0.1°, the observation equation of the system is:

[0146]

[0147] The rest of the simulation parameters are the same as those in the collaborative detection experiment of the dual infrared detection system. The simulation results are as follows:

[0148] from Figure 7 It can be seen that the coordinated detection method of infrared + radar can also achieve accurate tracking of the target. Figure 8 and Fig. 9 It can be found that by introducing the radar detection system for cooperative detection, the target distance information has a high measurement accuracy, and the target state estimation error is significantly reduced, especially the position root mean square error is reduced by 62.68%, and the speed root mean square error is also reduced by 27.56%. Therefore, the use of the infrared + radar cooperative detection method can further improve the target tracking accuracy of the algorithm of the present invention, especially for the target position estimation, making up for the disadvantage that the infrared detection system cannot detect distance information, so that the aircraft can obtain more accurate target information and improve the probability of success of the aircraft mission.

[0149] Through simulation analysis, it is verified that the use of IMM-STUKF information fusion algorithm for target tracking can reduce estimation errors, improve tracking accuracy, and have good tracking performance. At the same time, two collaborative detection methods, dual infrared and infrared + radar detection systems, are proposed. It is found that the collaborative detection method of radar + infrared detection system has more accurate state estimation values ​​and can be better applied to target tracking.

[0150] It is to be understood that the present invention is described by way of embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and embodiments that can fall within the scope of the claims of this application all fall within the scope protected by the present invention.

[0151] The contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

Claims

1. A target tracking method based on IMM-STUKF information fusion algorithm, characterized in that: include: The strong tracking filter STF is combined with the unscented Kalman filter UKF, and a suboptimal fading factor is introduced into the unscented Kalman filter to make the filter residual sequences orthogonal to each other, thus obtaining the STUKF filtering algorithm. The STUKF filtering algorithm and the interactive multi-model algorithm IMM are fused, and the IMM-STUKF information fusion algorithm is used to track the target to obtain the target state estimation, reduce the influence of uncertainty on the estimation error, and improve the tracking accuracy of the target under various maneuvers.

2. The target tracking method based on the IMM-STUKF information fusion algorithm according to claim 1, characterized in that: A suboptimal fading factor is introduced into the unscented Kalman filter UKF to keep the filter residual sequences orthogonal to each other. The specific method is as follows: Introducing the fading factor λ into the unscented Kalman filter UKF k , redefine the original state prediction covariance, observation prediction covariance, state and observation mutual covariance in the unscented Kalman filter, and the updated expressions are as follows: Where P k|k-1 is the state prediction covariance matrix, P zz,k|k-1 is the observation-prediction covariance matrix, P xz,k|k-1 is the state and observation covariance matrix, n is the state dimension, is the weighted coefficient of the variance of each sigma point after transformation, χ i,k / k-1 is the predicted sigma point, is the predicted value of the state variable, Q k is the process noise covariance matrix, is the predicted measured sigma point, is the predicted value of the observed quantity, R k is the observation noise covariance matrix; In the formula, the fading factor λ k The expression is as follows: In the formula, tr[·] represents the trace of the matrix, N k ,M k The definition is as follows: N k =C 0,k -βR k Where β is the weakening factor, which is the same as the definition of UKF filter; C 0,k is the covariance matrix of the residuals.

3. The target tracking method based on the IMM-STUKF information fusion algorithm according to claim 2 is characterized in that: The covariance matrix C of the residual 0,k , calculated by the following formula: In the formula, ε k is the residual at time k, and ρ is the forgetting factor.

4. The target tracking method based on the IMM-STUKF information fusion algorithm according to claim 1, characterized in that: The Kalman filter in the interactive multiple model algorithm IMM is replaced by the STUKF filtering algorithm to obtain the IMM-STUKF information fusion algorithm.

5. A target tracking system based on the IMM-STUKF information fusion algorithm, characterized in that: include: The STUKF filtering module is used to combine the strong tracking filter STF with the unscented Kalman filter UKF, introduce a suboptimal fading factor into the unscented Kalman filter, and keep the filter residual sequences orthogonal to each other, thus obtaining the STUKF filtering algorithm; IMM-STUKF information fusion module, used to call the STUKF filter module and fuse it with the interactive multi-model algorithm IMM; The target tracking module is used to track the target using the IMM-STUKF information fusion module to obtain the target state estimation, reduce the impact of uncertainty on the estimation error, and improve the tracking accuracy of the target under various maneuvers.

6. The target tracking system based on the IMM-STUKF information fusion algorithm according to claim 5, characterized in that: In the STUKF filtering module, Introducing the fading factor λ into the unscented Kalman filter UKF k , redefine the original state prediction covariance, observation prediction covariance, state and observation mutual covariance in the unscented Kalman filter, and the updated expressions are as follows: Where P k|k-1 is the state prediction covariance matrix, P zz,k|k-1 is the observation-prediction covariance matrix, P xz,k|k-1 is the state and observation covariance matrix, n is the state dimension, is the weighted coefficient of the variance of each sigma point after transformation, χ i,k / k-1 is the predicted sigma point, is the predicted value of the state variable, Q k is the process noise covariance matrix, is the predicted measured sigma point, is the predicted value of the observed quantity, R k is the observation noise covariance matrix; In the formula, the fading factor λ k The expression is as follows: In the formula, tr[·] represents the trace of the matrix, N k ,M k The definition is as follows: N k =C 0,k -βR k Where β is the weakening factor, which is the same as the definition of UKF filter; C 0,k is the covariance matrix of the residuals.

7. The target tracking system based on the IMM-STUKF information fusion algorithm according to claim 6 is characterized in that: In the STUKF filtering module, The covariance matrix C of the residual 0,k , calculated by the following formula: In the formula, ε k is the residual at time k, and ρ is the forgetting factor.

8. The target tracking method based on the IMM-STUKF information fusion algorithm according to claim 5, characterized in that: In the IMM-STUKF information fusion module, the fusion method is: replacing the Kalman filter in the interactive multiple model algorithm IMM with the STUKF filtering algorithm to obtain the IMM-STUKF information fusion algorithm.

9. A computer program product, characterized in that When the computer program product is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium, characterized in that: When the computer program in the medium is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images