Maneuvering target tracking method based on cascaded multistage adaptive fusion

Abstract

The invention discloses a maneuvering target tracking method based on cascaded multistage adaptive fusion, and relates to the technical field of target tracking and information fusion, in particular to a maneuvering target tracking method. The objective of the invention is to solve the problem that the existing method is difficult to effectively distinguish the real source of an error when the maneuvering uncertainty and the measurement uncertainty are deeply coupled, so that the maneuvering target tracking precision is poor. The method comprises the following steps of: 1, outputting state estimation and a state prediction value by a strong tracking filter STF, and outputting a covariance matrix of the state estimation and a covariance matrix of the state prediction value by the strong tracking filter STF; performing CKF output state estimation and a covariance matrix of the CKF output state estimation; 2, generating equilibrium state estimation; 3, obtaining enhanced state estimation; 4, calculating an error cross covariance matrix; calculating an optimal weight; and performing optimal recombination on the state estimation output by the strong tracking filter STF and the enhanced state estimation to obtain a final output state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target tracking and information fusion, and particularly relates to a method for tracking a maneuvering target. BACKGROUND

[0002] High maneuvering target tracking is a key core technology in the fields of modern radar system, autonomous navigation and intelligent control, and is also an important symbol for measuring the technical level of a country in the fields of advanced manufacturing, airspace management and intelligent transportation. In the field of civil aviation, accurate flight path tracking of an aircraft is the basis for ensuring air traffic safety and efficiency. In space activities, reliable monitoring of space flying bodies and debris is the fundamental prerequisite for ensuring the safety of on-orbit facilities and avoiding catastrophic collisions. Therefore, how to achieve robust and high-precision estimation of the motion state of a target under the condition that the target presents complex motion patterns such as high speed and strong maneuvering is a technical problem that needs to be solved urgently and has great significance in the field of target tracking.

[0003] The challenge of this problem mainly comes from two levels of deep coupling uncertainty. First, the kinematic uncertainty of the target itself. Non-cooperative targets usually do not follow the preset motion model, and their acceleration may jump sharply in a very short time, showing strong nonlinearity and non-stationarity, which puts high requirements on the dynamic response ability and model adaptability of the tracking algorithm. Secondly, the measurement uncertainty of the detection environment. The sensor is easily affected by complex electromagnetic environment, sea clutter, multipath effect and other factors in the process of obtaining target measurement information, resulting in significant non-Gaussian 'heavy-tailed' distribution characteristics of the measurement noise, and frequent occurrence of outliers, which seriously pollutes the real target information.

[0004] To cope with the above-mentioned double challenges, the academic and industrial circles have proposed various tracking methods, among which multi-model parallel fusion is a mainstream technical route. This kind of method usually runs multiple independent filters for different motion patterns in parallel, and then performs a one-time weighted average of their outputs through a fusion center. However, this single-level fusion paradigm has a fundamental limitation: it tries to solve the two kinds of uncertainty with different natures in one step, which is difficult to effectively distinguish the real source of error when the maneuvering uncertainty and the measurement uncertainty are deeply coupled. For example, a large residual error caused by the real maneuvering of the target may be misjudged as an abnormal measurement and be excessively suppressed, resulting in tracking lag; on the contrary, an abnormal measurement caused by strong noise may be misjudged as a maneuvering and cause filter divergence. This ambiguity of error source fundamentally limits the further improvement of the tracking performance of traditional fusion methods.

[0005] Therefore, the structural defects of the existing parallel fusion method make its performance reach a bottleneck when facing complex coupling scenarios. The current technical field urgently needs an innovative fusion methodology at the architecture level. This new methodology needs to abandon the planar processing mode of a single fusion layer and instead build a cascading processing logic with depth and hierarchy. Through this logic, complex tracking problems can be decomposed into a series of well-defined subtasks, focusing on separating and suppressing specific sources of uncertainty at different processing stages, thereby performing step-wise refinement on state estimation, with a view to fundamentally breaking through the performance bottleneck of existing fusion methods. SUMMARY

[0006] The purpose of the present application is to solve the problem that the existing method is difficult to effectively distinguish the true source of error when the maneuvering uncertainty and the measurement uncertainty are deeply coupled, resulting in poor tracking accuracy of the maneuvering target, and proposes a maneuvering target tracking method based on cascading multi-level adaptive fusion.

[0007] The specific process of the maneuvering target tracking method based on cascading multi-level adaptive fusion is as follows:

[0008] Step 1: input the target measurement data received by the radar at time k into the strong tracking filter STF and the cubature Kalman filter CKF respectively;

[0009] The strong tracking filter STF outputs the state estimate and the predicted value of the state , and the covariance matrix of and the covariance matrix of and ;

[0010] The cubature Kalman filter CKF outputs the state estimate and the covariance matrix of ;

[0011] The target is a UAV;

[0012] The state is the azimuth and distance of the target relative to the radar;

[0013] Step 2: through a dynamic weighting mechanism, the state estimate output by the strong tracking filter STF and the state estimate output by the cubature Kalman filter CKF are fused to generate a balanced state estimate ;

[0014] Step 3: the balanced state estimate is subjected to strong tracking correction to obtain an enhanced state estimate ;​​

[0015] Step 4: Calculate the state estimation of the strong tracking filter STF output and the enhanced state estimation between the error cross-covariance matrix ; Calculate the optimal weight based on the error cross-covariance matrix Calculate the optimal weight and the optimal weight ; Calculate the optimal weight based on the optimal weight recombine the state estimation of the strong tracking filter STF output and the enhanced state estimation to obtain the final output state .

[0016] The beneficial effects of the present invention are:

[0017] 1. Realize phased suppression of coupled uncertainties:

[0018] The traditional parallel fusion method tries to solve both target maneuvering and measurement noise uncertainties in one step, resulting in ambiguous error sources and limited performance. The present invention, through innovative cascading processing logic, decomposes the complex tracking problem into a series of well-defined subtasks: the first level fusion (MMFF) mainly deals with the uncertainty caused by target maneuvering mode switching; the second level fusion (STEF) focuses on suppressing system response delay; the third level fusion (AMFF) recombines all information under the criterion of global optimization. This hierarchical processing paradigm fundamentally realizes effective separation and step-by-step suppression of uncertainties from different sources, solving the problem of coupled error sources in the background technology.

[0019] 2. Gradual optimization and refinement of state estimation:

[0020] In the cascading method of the present invention, information flow is passed and optimized step by step, and the output of each level is corrected and refined based on the previous level. The state estimation of the target is gradually refined. Experimental results directly verify this point, from the basic filtering layer to MMFF, then to STEF and finally to AMFF, the tracking error shows a clear step-by-step decreasing trend. The final estimation result is significantly better than the output of any single link in the method in terms of accuracy and robustness, effectively distinguishing the true source of error when maneuvering uncertainty and measurement uncertainty are deeply coupled, improving the tracking accuracy of maneuvering targets.

[0021] 3. Significant reduction in tracking error, improvement in key performance indicators:

[0022] The excellent performance of the application is quantitatively supported by simulation experiments. In a typical severe test scene of strong maneuvering and mixed noise coupling, compared with the high-performance strong tracking filter (STF) as a benchmark, the state estimation of the final output of the application:

[0023] In terms of positioning accuracy, the average position mean square error (ARMSE) is reduced by 41.8%.

[0024] In terms of dynamic characteristic capture, the average speed mean square error (ARMSE) is reduced by as high as 62.6%.

[0025] In terms of robustness, the error distribution is more concentrated and there are fewer outliers, indicating that the method has higher consistency and reliability in the face of random disturbances.

[0026] 4. System-level performance gain is achieved by cascading fusion logic:

[0027] The core advantage of the application is not only the improvement of a specific module, but also the innovative cascading processing logic itself. Through careful organization and step-by-step processing of information flow, system-level "1+1>2" synergistic gain is achieved. The performance of the final output (AMFF) exceeds that of the benchmark adaptive strong tracking filter as input, which is already very advanced, which strongly proves the superiority of the fusion processing logic proposed in the application, and provides a new and excellent system solution for solving target tracking problems in complex coupling environments. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The overall structure diagram of the cascading multi-stage adaptive fusion method proposed for the application;

[0029] Figure 2 The target maneuvering behavior and complex environment scene diagram;

[0030] Figure 3 The X-direction tracking error comparison curve of the application method and single filter in the two-dimensional plane rectangular coordinate system, showing the step-by-step performance improvement;

[0031] Figure 4 The Y-direction tracking error comparison curve of the application method and single filter in the two-dimensional plane rectangular coordinate system, showing the step-by-step performance improvement. DETAILED DESCRIPTION

[0032] Specific implementation one: the specific process of the maneuvering target tracking method based on the cascading multi-stage adaptive fusion of the embodiment is:

[0033] The basic filtering layer provides the raw information input to the entire fusion method by running at least two complementary base filters in parallel. Specifically, it outputs a high-maneuverability response state estimation sequence for rapid response to target maneuvers and a high-stability state estimation sequence for ensuring accuracy during smooth motion.

[0034] The first-level fusion module (MMFF) receives two outputs from the basic filtering layer. Its core is to perform a preliminary dynamic-static trade-off and fusion of the high-mobility response estimate and the high-stability estimate through a dynamic weighting mechanism based on the innovation norm, generating a robust and balanced intermediate state.

[0035] The second-level fusion module (STEF) receives the output of the first-level fusion module and uses it as prior information. Its core is to use the latest measurement information and introduce a second-level fading factor to perform an online strong tracking correction. This aims to actively compensate for the tracking delay that may be caused by maneuvering and improve the dynamic response agility of the system.

[0036] The third-level fusion module (AMFF) receives the high-mobility response estimate from the basic filtering layer and the enhanced post-estimate from the second-level fusion module. Its core is based on the optimal fusion theory that takes into account the error cross-covariance, calculates the optimal weight matrix, and performs a final weighted combination of the two input information streams to obtain the theoretically optimal final state estimate of the system.

[0037] Step 1: Run the basic filtering layer: At time k, input the target measurement data (echo data) received by the radar into the strong tracking filter STF and the capacitive Kalman filter CKF respectively;

[0038] Strong tracking filter STF output agile state estimation And the predicted value of agile state ,as well as covariance matrix and covariance matrix ;

[0039] Robust state estimation of the output of a volumetric Kalman filter (CKF) and covariance matrix ;

[0040] The input to the strong tracking filter is the radar echo, which is obtained first. Time prediction value Then, based on the measured values, corrections (also called updates) are obtained. The estimated value of time This "prediction-update" process iterates repeatedly to continue obtaining... and the predicted value and the estimated value at the next time. The final result is the estimated value at each time.

[0041] The strong tracking filter STF is a high maneuverability filter;

[0042] The cubature Kalman filter CKF is a high stability filter;

[0043] The target is a UAV;

[0044] The state is the target's azimuth and range relative to the radar;

[0045] Step 2: Perform the first level fusion (MMFF): fuse the agile state estimate output by the strong tracking filter STF and the robust state estimate output by the cubature Kalman filter CKF through a dynamic weighting mechanism to generate a balanced state estimate .

[0046] Step 3: Perform the second level fusion (STEF): perform a strong tracking correction on the balanced state estimate to obtain an enhanced state estimate with high dynamic response capability .

[0047] Step 4: Perform the third level fusion (AMFF): calculate the error cross-covariance matrix between the agile state estimate output by the strong tracking filter STF and the enhanced state estimate ; calculate the optimal weight based on the error cross-covariance matrix and the optimal weight ; perform optimal recombination on the agile state estimate output by the strong tracking filter STF and the enhanced state estimate based on the optimal weight to obtain the final output state .

[0048] Specific implementation method two: the difference between this implementation method and the specific implementation method one is that in the step 2, the first level fusion (MMFF) is performed: fuse the agile state estimate output by the strong tracking filter STF and the robust state estimate output by the cubature Kalman filter CKF through a dynamic weighting mechanism to generate a balanced state estimate ; the specific process is as follows:

[0049] Step 21: calculate a dynamic threshold related to the current target measurement noise covariance matrix ; indicates as:

[0050] (1)

[0051] in, This represents the preset scaling factor, which is usually set to 2; Represents the current measurement noise covariance matrix traces; This represents the target measurement noise covariance matrix received by the radar at time k.

[0052] Step 22: Calculate the agile state estimate of the strong tracking filter STF output. The innovation norm; denoted as:

[0053] (2)

[0054] in, This represents the measured value of the target measurement data received by the radar at time k; The measurement matrix represents the operation of mapping the target state vector x from the state space to the measurement space. This represents the predicted value of the agile state output by the strong tracking filter (STF) at time k. This indicates a search for the norm. Let k represent the innovation norm at time k;

[0055] Step 23: By comparing the innovation norm with the dynamic threshold, determine the appropriate state estimation method for agile state estimation. Dynamic weights ; indicates as:

[0056] (3)

[0057] in, This indicates the preset maneuver weights. This represents the preset stationary weights. To ensure greater trust in the agile filter during maneuvers, it needs to satisfy... ;

[0058] Step 24: Based on state estimation Dynamic weights Agile state estimation of the output of the strong tracking filter STF Robust state estimation of the output of the capacitive Kalman filter (CKF) The equilibrium state is estimated by fusion. ; indicates as:

[0059] (4)

[0060] The other steps and parameters are the same as in Specific Implementation Method 1.

[0061] Specific implementation three: the difference between this implementation and specific implementation one or two is that: the step 3 performs second level fusion (STEF) on the balanced state estimation Perform strong tracking correction to obtain enhanced state estimation with high dynamic response capability The specific process is:

[0062] Step 31: based on the balanced state estimation Calculate the innovation vector

[0063] Step 32: based on the innovation vector, obtain the actual covariance matrix, and based on the actual covariance matrix and the theoretical covariance matrix, calculate the second level fading factor

[0064] Step 33: based on the second level fading factor, amplify the covariance matrix of the balanced state estimation Obtain the amplified covariance matrix

[0065] Step 34: based on the balanced state estimation And the amplified covariance matrix Obtain enhanced state estimation with high dynamic response capability And the corrected posterior error covariance matrix .

[0066] The other steps and parameters are the same as specific implementation one or two.

[0067] Specific implementation four: the difference between this implementation and one of specific implementations one to three is that: in step 31, based on the balanced state estimation Calculate the innovation vector; represented as:

[0068] (5)

[0069] Wherein, Indicates the innovation vector at time k.

[0070] The other steps and parameters are the same as one of specific implementations one to three.

[0071] Specific implementation five: the difference between this implementation and one of specific implementations one to four is that: in step 32, based on the innovation vector, obtain the actual covariance matrix, and based on the actual covariance matrix and the theoretical covariance matrix, calculate the second level fading factor; represented as:

[0072] (6)

[0073] Wherein, Indicates the balanced state estimation ​covariance matrix of the state estimation denotes the covariance matrix of the state estimation denotes the transpose of the covariance matrix of the state estimation denotes the covariance matrix of the state estimation denotes the transpose of the covariance matrix of the state estimation denotes the second-order fading factor at time k.

[0074] denotes the actual covariance matrix of the state estimation denotes the theoretical covariance matrix of the state estimation

[0075] The other steps and parameters are the same as those in the first to fourth embodiments.

[0076] The sixth embodiment is different from the first to fifth embodiments in that the step 33 amplifies the covariance matrix of the state estimation based on the second-order fading factor, to obtain an amplified covariance matrix; denoted as:

[0077] (7)

[0078] wherein, denotes the amplified covariance matrix of the state estimation at time k.

[0079] The other steps and parameters are the same as those in the first to fifth embodiments.

[0080] The seventh embodiment is different from the first to sixth embodiments in that the step 34 obtains an enhanced state estimation and a corrected posterior error covariance matrix based on the state estimation and the amplified covariance matrix ; the specific process is as follows:

[0081] Step 341: calculate the Kalman gain based on the amplified covariance matrix ; denoted as:

[0082]

[0083] wherein, denotes the Kalman gain at time k.

[0084] Step 342: update the state estimation based on the Kalman gain and the innovation vector, to obtain the enhanced state estimation; denoted as:

[0085]

[0086] wherein, ​This represents enhanced state estimation;

[0087] Step 343: Update the amplified covariance matrix using Kalman gain to obtain the corrected posterior error covariance matrix; represented as:

[0088]

[0089] in, Represents the identity matrix; This represents the corrected posterior error covariance matrix.

[0090] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0091] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Methods One through Seven in that: in step 4, the third-level fusion (AMFF) is performed: the agile state estimate of the strong tracking filter (STF) output is calculated. With enhanced state estimation Error cross covariance matrix between Based on the error cross-covariance matrix Calculate the optimal weight and optimal weight Based on optimal weights Agile state estimation of the output of the strong tracking filter STF and enhanced state estimation Perform optimal reorganization to obtain the final output state. The specific process is as follows:

[0092] Step 41: Calculate the agile state estimate of the strong tracking filter STF output. With enhanced state estimation Error cross covariance matrix between ; indicates as:

[0093] To ensure numerical stability and the positive definiteness of the covariance matrix, this invention draws on the Josephian covariance update concept to derive the error cross-covariance matrix. ;

[0094]

[0095] in, express Time-based Kalman gain; This represents the filtering gain of the strong tracking filter (STF) at time k. Represents the state transition matrix; Indicates to Find the transpose; express time error cross-covariance matrix; denotes time process noise covariance matrix; denotes transpose;

[0096] This formula intuitively reflects that the final posterior cross-covariance is formed by the prediction cross-covariance after measurement update through two different paths.

[0097] where k is the filter gain of the STF at time ; is denoted as:

[0098]

[0099] where, denotes the predicted value of the agile state output by the STF covariance matrix of

[0100] Step 42: based on the error cross-covariance matrix calculate the optimal weight and the optimal weight ;

[0101] Step 43: based on the optimal weight optimal reorganization of the agile state estimate and the augmented state estimate output by the STF, to obtain the final output state .

[0102] Other steps and parameters are the same as those in Embodiment 1 to Embodiment 7 to Embodiment 1.

[0103] Embodiment 9: Different from one of Embodiment 1 to Embodiment 8, in the step 42, the optimal weight and the optimal weight are calculated based on the error cross-covariance matrix ; is denoted as:

[0104] (8)

[0105] (9).

[0106] Other steps and parameters are the same as those in Embodiment 1 to Embodiment 8 to Embodiment 1.

[0107] Embodiment 10: Different from one of Embodiment 1 to Embodiment 9, in the step 43, based on the optimal weight optimal reorganization of the agile state estimate and the augmented state estimate Optimal recombination is performed to obtain a final output state ; is expressed as:

[0108] (10).

[0109] Other steps and parameters are the same as those in the first to ninth embodiments.

[0110] The high-maneuvering-response filter in the basic filtering layer is specifically a strong tracking filter (STF) which automatically sets its core parameters by minimizing long-term cumulative errors; and the high-stability filter is specifically a standard cubature Kalman filter (CKF) which has high filtering accuracy and numerical stability in a nonlinear environment. The characteristics of the two filters are complementary, providing high-quality and different information sources for subsequent cascade fusion.

[0111] The application divides a complex tracking problem into a series of serially processed subtasks through a cascade refining process of "basic filtering -> preliminary trade-off -> dynamic enhancement -> optimal recombination". The core idea of the structure is stage-by-stage uncertainty suppression, wherein the first-level fusion mainly processes the uncertainty caused by target maneuver mode switching, the second-level fusion focuses on suppressing system response delay, and the third-level fusion recombines all information under the criterion of global optimality, thereby achieving a significant performance gain at the system level and showing excellent robustness and tracking accuracy in a strongly coupled uncertainty environment.

[0112] In addition, in order to further intuitively verify the effectiveness and performance gain of the cascade multi-level adaptive fusion method proposed in the application in the design of the structure, the output performance of each fusion module in the method is compared. As shown in the attached Figure 2 and Figure 3 , it can be clearly observed that the performance of the algorithm is progressively improved. Starting from the basic filter (STF) as a benchmark, its performance is already excellent, but there are still fluctuations or error peaks in some stages. After the first-level fusion (MMFF), the performance curve is effectively smoothed in the stable stage through the trade-off with the stable filter, but there is still a certain delay in the response to sudden maneuvers. On this basis, the second-level fusion (STEF) significantly suppresses the error peaks in the maneuvering stage by introducing real-time measurement correction, and the robustness is further improved. Finally, the third-level fusion (AMFF) as the final output of the method, its error curve is almost at the bottom of all curves and has the smallest fluctuation throughout the tracking process, achieving the best combination of steady-state accuracy and dynamic response. This fundamentally proves the superiority of the cascade processing logic of "stage-by-stage uncertainty suppression" proposed in the application, and the system-level performance is significantly enhanced through stage-by-stage refinement of state estimation.

[0113] The beneficial effects of the present invention are verified using the following embodiments:

[0114] Example 1:

[0115] To systematically verify the effectiveness of the cascaded multi-level adaptive fusion method proposed in this invention, a highly challenging typical high-maneuverability target tracking simulation scenario was first constructed, such as... Figure 2 As shown, this scenario simulates the complex trajectory of a non-cooperative target in a two-dimensional Cartesian coordinate system. Its motion pattern includes multiple stages: 0-20 seconds of uniform linear motion, 20-40 seconds of strong acceleration in the X direction, 40-60 seconds of sharp deceleration in the Y direction, 60-80 seconds of S-shaped maneuvering, and subsequent return to uniform velocity. Simultaneously, the measurement environment is set to include mixed noise with non-Gaussian heavy-tailed characteristics to simulate complex interference in real-world detection environments. This scenario integrates the target's high maneuverability with the strong uncertainty of the external environment, posing a severe test to the comprehensive performance of any tracking algorithm. The specific implementation of this invention is aimed at... Figure 2 This presents a robust and accurate solution to the complex problem.

[0116] The solution proposed in this invention has the following overall structure: Figure 1 As shown, this is a cascaded multi-stage adaptive fusion method. This method abandons the traditional parallel fusion approach, deconstructing the complex tracking task into a serial refinement process containing four core stages. The first stage of this method is the basic filtering layer, whose core task is to provide two high-quality and complementary raw state estimation streams for the entire fusion method. In this embodiment, this layer runs a standard capacitive Kalman filter (CKF) in parallel to provide a high-precision robust estimation stream, while simultaneously running an adaptive strong tracking filter (STF) to provide a highly dynamic and agile estimation stream. These two filters simultaneously process data from... Figure 2 The measurement data of the scene shown, and their differences in characteristics, lay the foundation for the subsequent fusion stage to achieve complementary advantages.

[0117] After obtaining the two basic information streams, the method enters the second stage, namely the first-level fusion (multi-model filter fusion, MMFF). The main goal of this stage is to perform a preliminary dynamic-static tradeoff on the two outputs from the base layer, aiming to generate a robust and well-balanced intermediate state estimate. Its core mechanism is a dynamic weighting strategy based on the innovation norm of the agile filter. When the system determines that the target is in a stationary motion state (corresponding to...), the weighting strategy is applied. Figure 2 In the 0-20s or 80-100s range, robust CKF estimates are given higher weights; conversely, when the target is judged to be maneuvering (corresponding to...),... Figure 2If the target is in a high dynamic state (20-80s part of the phase), then the agile ASTF estimation will be given a higher weight. The output of this phase is a preliminary weighted fusion, balanced performance state estimation.

[0118] The next phase is the third phase, i.e. the second level fusion (strong tracking enhanced fusion, STEF), which aims to actively correct the tracking delay caused by the target's drastic maneuver. The output of the previous MMFF is regarded as a priori estimation to be enhanced, and the latest measurement information is used to perform an online strong tracking correction. When the target is in a high dynamic state as shown in the figure, this mechanism can amplify the prior error covariance online through a dynamically calculated fading factor, thereby reducing the system's trust in historical estimates and forcing the filter to trust the latest measurement information more, achieving rapid correction of the state estimation to follow the target's true trajectory. Figure 2

[0119] The last phase of this method is the third level fusion (adaptive multi-stage fusion, AMFF). This phase performs the theoretically optimal recombination of two high-quality but very different information streams to obtain the globally optimal final system output. One of the two input information streams is the STF estimation directly from the base layer, representing the global long-term stability; the other is the STEF estimation refined by the previous two stages, representing the highest local dynamic response capability. AMFF calculates the error cross-covariance between the two information streams and solves the optimal fusion equation to finally output a state estimation that theoretically achieves the best balance between tracking accuracy and robustness.

[0120] The complete method described in the present application is applied to the challenging scenario shown in Figure 2 The outstanding performance is intuitively verified by Figure 3 and Figure 4 The two figures show the tracking error comparison of each level of fusion module output in the X and Y directions of the two-dimensional plane rectangular coordinate system, respectively. The figure clearly shows an optimized trajectory that gradually improves in performance: starting from the STF filter as the benchmark, through the preliminary smoothing of MMFF, to the effective suppression of the maneuver peak by STEF, and finally to the globally optimal performance of AMFF, whose error curve is almost at the bottom of all curves throughout the tracking process, with the least fluctuation. This strongly proves the cascading processing logic of the present application, which successfully solves the tracking problem of strong maneuver and complex noise coupling through the step-by-step refinement of state estimation.

[0121] ​The present application can have other various embodiments, and those skilled in the art can make various corresponding changes and modifications according to the present application without departing from the spirit and essence of the present application, and these corresponding changes and modifications shall all belong to the protection scope of the claims of the present application.

Claims

1. A maneuvering target tracking method based on cascaded multi-level adaptive fusion, characterized in that: The specific process of the method is as follows: Step 1: At time k, input the target measurement data received by the radar into the strong tracking filter (STF) and the capacitive Kalman filter (CKF) respectively; Strong tracking filter STF output state estimation and the predicted value of the state ,as well as covariance matrix and covariance matrix ; Capacitive Kalman Filter (CKF) Output State Estimation and covariance matrix ; The target is a drone; The status is the target's azimuth and range relative to the radar; Step 2: State estimation of the output of the strong tracking filter (STF) through a dynamic trade-off mechanism. State estimation of the output of the capacitive Kalman filter (CKF) The equilibrium state is estimated by fusion. ; Step 3: Equilibrium State Estimation Perform strong tracking correction to obtain enhanced state estimates. ; Step 4: Calculate the state estimate of the strong tracking filter (STF) output. With enhanced state estimation Error cross covariance matrix between Based on the error cross-covariance matrix Calculate the optimal weight and optimal weight Based on optimal weights State estimation of the output of the strong tracking filter STF and enhanced state estimation Perform optimal reorganization to obtain the final output state. .

2. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 1, characterized in that: In step 2, a dynamic trade-off mechanism is used to estimate the state of the output of the strong tracking filter (STF). State estimation of the output of the capacitive Kalman filter (CKF) The equilibrium state is estimated by fusion. ; The specific process is as follows: Step 21: Calculate a noise covariance matrix with respect to the current target measurement. Related dynamic thresholds ; indicates as: (1) in, This indicates the preset scaling factor; Represents the current measurement noise covariance matrix traces; This represents the target measurement noise covariance matrix received by the radar at time k. Step 22: Calculate the state estimate of the strong tracking filter (STF) output. The innovation norm; denoted as: (2) in, This represents the measured value of the target measurement data received by the radar at time k; For measurement matrix; This represents the predicted state of the output of the strong tracking filter (STF) at time k. This indicates a search for the norm. Let k represent the innovation norm at time k; Step 23: Determine the state estimate by comparing the innovation norm with the dynamic threshold. Dynamic weights ; indicates as: (3) in, This indicates the preset maneuver weights. The preset stationary weights must satisfy the following conditions: ; Step 24: Based on state estimation Dynamic weights State estimation of the output of the strong tracking filter STF State estimation of the output of the capacitive Kalman filter (CKF) The equilibrium state is estimated by fusion. ; indicates as: (4)。 3. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 2, characterized in that: In step 3, the equilibrium state is estimated. Perform strong tracking correction to obtain enhanced state estimates. ; The specific process is as follows: Step 31: Equilibrium State Estimation Calculate the innovation vector; Step 32: Obtain the actual covariance matrix based on the innovation vector, and calculate the fading factor based on the actual covariance matrix and the theoretical covariance matrix; Step 33: Equilibrium state estimation based on fading factor covariance matrix The matrix is ​​then magnified to obtain the magnified covariance matrix. Step 34: Equilibrium State Estimation and the magnified covariance matrix To obtain enhanced state estimates and the corrected posterior error covariance matrix .

4. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 3, characterized in that: Step 31 is based on equilibrium state estimation Calculate the innovation vector; represented as: (5) in, Let k represent the innovation vector at time k.

5. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 4, characterized in that: In step 32, the actual covariance matrix is ​​obtained based on the innovation vector, and the fading factor is calculated based on the actual covariance matrix and the theoretical covariance matrix; expressed as: (6) in, Equilibrium state estimation The covariance matrix; Indicates to Find the transpose; Indicates to Find the transpose; This represents the fading factor at time k.

6. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 5, characterized in that: In step 33, the equilibrium state is estimated based on the fading factor. covariance matrix Magnification is performed to obtain the magnified covariance matrix, which is represented as: (7) in, Let represent the amplified covariance matrix at time k.

7. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 6, characterized in that: Step 34 is based on equilibrium state estimation and the magnified covariance matrix To obtain enhanced state estimates and the corrected posterior error covariance matrix The specific process is as follows: Step 341: Based on the magnified covariance matrix Calculate the Kalman gain; expressed as: in, Indicates the Kalman gain at time k; Step 342: Estimating the equilibrium state using Kalman gain and innovation vector Perform an update to obtain an enhanced state estimate; represented as: in, This represents enhanced state estimation; Step 343: Update the amplified covariance matrix using Kalman gain to obtain the corrected posterior error covariance matrix; represented as: in, Represents the identity matrix; This represents the corrected posterior error covariance matrix.

8. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 7, characterized in that: In step 4, the state estimate of the strong tracking filter (STF) output is calculated. With enhanced state estimation Error cross covariance matrix between Based on the error cross-covariance matrix Calculate the optimal weight and optimal weight Based on optimal weights State estimation of the output of the strong tracking filter STF and enhanced state estimation Perform optimal reorganization to obtain the final output state. The specific process is as follows: Step 41: Calculate the state estimate of the output of the strong tracking filter (STF). With enhanced state estimation Error cross covariance matrix between ; indicates as: in, express Time-based Kalman gain; This represents the filtering gain of the strong tracking filter (STF) at time k. Represents the state transition matrix; Indicates to Find the transpose; express Time error cross covariance matrix; express Time-matrix noise covariance matrix; This indicates the transpose; Where k is the filtering gain of the strong tracking filter STF. ; indicates as: in, The predicted value representing the state of the output of the strong tracking filter (STF). The covariance matrix; Step 42: Based on the error cross-covariance matrix Calculate the optimal weight and optimal weight ; Step 43: Based on optimal weights State estimation of the output of the strong tracking filter STF and enhanced state estimation Perform optimal reorganization to obtain the final output state. .

9. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 8, characterized in that: In step 42, based on the error cross-covariance matrix Calculate the optimal weight and optimal weight ; indicates as: (8) (9)。 10. The maneuvering target tracking method based on cascaded multi-level adaptive fusion according to claim 9, characterized in that: In step 43, based on optimal weight State estimation of the output of the strong tracking filter STF and enhanced state estimation Perform optimal reorganization to obtain the final output state. ; indicates as: (10)。

Images