A maneuvering target fusion tracking method for full-dimensional deficient dimension disordered measurement
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINESE AERONAUTICAL RADIO ELECTRONICS RES INST
- Filing Date
- 2023-04-21
- Publication Date
- 2026-06-09
AI Technical Summary
In distributed and clustered operations, the full-dimensional and out-of-dimensional disordered measurement problems of multi-source heterogeneous sensors lead to information loss and affect the quality of fusion, especially making it difficult to track maneuvering targets in complex combat environments.
A fusion tracking method for maneuvering targets oriented towards full-dimensional and out-of-order measurements is adopted. By optimizing the filtering and updating the model through model conditional reinitialization, extended Kalman filtering, and interactive multi-model algorithm, the method can effectively process and fuse measurement data.
It improves the utilization efficiency of out-of-order measurements from multiple heterogeneous sensors, enhances the tracking capability and fusion accuracy of moving targets, and meets the real-time requirements of engineering.
Smart Images
Figure CN116774210B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of integrated airborne avionics systems, and particularly to collaborative sensing and fusion positioning technology of heterogeneous sensors such as airborne radar and optoelectronics. Specifically, it refers to a fusion tracking method for maneuvering targets oriented towards full-dimensional, out-of-dimension, and disordered measurement. Background Technology
[0002] Based on the detection characteristics of airborne sensors, heterogeneous sensors such as radar, electro-optical, and electronic warfare sensors can be divided into active and passive types. Radar is an active sensor, and its measurement data is full-dimensional, including range, azimuth, and elevation. When the radar is suppressed or jammed, range measurement becomes unusable, and the measurement data degenerates into missing-dimensional data, that is, only including azimuth and elevation. Electro-optical and electronic warfare sensors are passive sensors, and their measurement data is missing-dimensional. When electro-optical sensors enter ranging mode, the electro-optical measurement data is transformed into full-dimensional data including range information.
[0003] In current engineering applications, the probability of out-of-order measurement data is low, and such data is usually treated as erroneous and discarded, not proceeding to subsequent filtering and updating stages. However, with the emergence of new combat styles such as distributed warfare and cluster warfare, different platforms collaborate to complete detection tasks by loading single-configuration sensors; that is, each platform loads only one type of sensor and completes the detection task through distributed collaboration. In this context, communication latency will become increasingly prominent, and the problem of out-of-order full-dimensional and missing-dimensional measurement data reported by sensors from various platforms will be unavoidable. Current processing methods in engineering applications are prone to causing significant information loss, thus affecting the fusion quality. Furthermore, considering the increasingly complex trend of current adversarial styles, the complex and ever-changing nature of target movement has become inevitable. Therefore, how to achieve fusion tracking of moving targets based on out-of-order full-dimensional and missing-dimensional measurements from multi-source heterogeneous sensors, while meeting real-time engineering requirements, has become an urgent problem to be solved. Summary of the Invention
[0004] The purpose of this invention is to provide a fusion tracking method for maneuvering targets based on full-dimensional, missing-dimensional, and disordered measurements, in order to solve the problem of full-dimensional, missing-dimensional, and disordered measurements based on multi-source heterogeneous sensors.
[0005] To achieve the above objectives, the present invention employs the following technical solution:
[0006] A fusion tracking method for maneuvering targets based on full-dimensional missing-dimensional disordered measurements includes:
[0007] The first step is to reinitialize the model conditions; the motion models used include uniform linear motion model and current statistical model; after reinitializing the model conditions, the target state mixture estimate and covariance mixture estimate are obtained.
[0008] The second step is to determine whether the received full-dimensional or missing-dimensional measurement data is out of order; if it is in order, proceed to the third step; if it is out of order, proceed to the fourth step.
[0009] The third step is to perform model conditional filtering based on the full-dimensional or missing-dimensional measurement data in the sequential measurement data, and obtain the motion model state estimate, covariance estimate and motion model likelihood probability, then proceed to the fifth step.
[0010] The fourth step involves using full-dimensional or missing-dimensional measurement data from the disordered measurement data, combined with the information covariance and predicted covariance obtained from the update cycle, as well as the measurement model used, to complete the model conditional filtering based on disordered measurements, and obtain the state estimate, covariance estimate, and model likelihood probability of each motion model. Proceed to the fifth step.
[0011] Fifth step, calculate the model probability at the current time.
[0012] The sixth step is estimation fusion, which fuses the target state estimate and covariance estimate at the current moment as the target tracking result.
[0013] Furthermore, the sixth step also includes:
[0014] Using the model probability at the current moment as the weight, a convex combination method is used to fuse the innovation covariance and prediction covariance of each model, which are then used as intermediate quantities in the fourth step of the next cycle.
[0015] Furthermore, in the first step:
[0016] In the k-th update cycle, the probability of the i-th model is calculated as follows: The corresponding target state estimates and covariance estimates are respectively and Therefore, the conditional mixture probability is:
[0017]
[0018] Where the number of models r = 2, Π ij The prior probability represents the transition probability from the i-th model to the j-th model; the mixture estimate of the target state and the mixture estimate of the covariance after reinitialization of the j-th model are:
[0019]
[0020]
[0021] The superscript T indicates matrix transpose, and the same applies below.
[0022] Furthermore, the second step includes:
[0023] During this update cycle, new measurement data with a reporting time of d is received. The measurement data includes two types: full-dimensional and dimension-deficient, which are respectively represented as and Judge the relationship between time d and time k. If d≥k, it is sequential measurement data and go to the third step; if d<k, it is out-of-order measurement data and go to the fourth step.
[0024] The third step includes:
[0025] I. State prediction
[0026] (3.1) For the uniform linear motion model, that is, i = 1:
[0027]
[0028]
[0029] Among them, and respectively represent the state prediction matrix and covariance prediction matrix from time k to time d, represents the target motion state transition matrix from time k to time d, represents the process noise covariance from time k to time d, respectively represent the target state mixed estimation and covariance mixed estimation of the i-th model;
[0030] (3.2) For the current statistical model, that is, i = 2:
[0031]
[0032]
[0033] Among them, U k,d , are the state transition matrix, information input matrix and mean of maneuvering acceleration in the current statistical model, and take the acceleration value in ; respectively represent the target state mixed estimation and covariance mixed estimation of the i-th model;
[0034] II. State update
[0035] (3.3) If the full-dimensional measurement data at time d is received For the motion model i (i = 1, 2), considering the requirements of engineering applications for the amount of calculation, the extended Kalman filter form is adopted, that is:
[0036]
[0037]
[0038]
[0039]
[0040] in, and Let K represent the state estimate and covariance estimate of the target at time d, respectively. d Indicates gain. and Let R represent the new interest and the new interest covariance, respectively; d,rap H represents the error covariance of full-dimensional measurement data. d,rap To extend the Jacobian matrix of the Kalman filter, let If the coordinates corresponding to the position state are represented as (x, y, z), then:
[0041]
[0042] in, 0 3×6 Represents a 3x6 matrix containing all zeros;
[0043] (3.4) If the missing dimension measurement data at time d is received For motion model i (i = 1, 2), the extended Kalman filter is also used, i.e.:
[0044]
[0045]
[0046]
[0047]
[0048] in, and Let K represent the target state estimate and covariance estimate at time d, respectively. d Indicates gain. and Let R represent the new interest and the new interest covariance, respectively; d,ap Let the error covariance of the missing-dimensional measurement data be represented by the following expression. The coordinates of the position state are represented as (x, y, z), H d,ap for:
[0049]
[0050] in,
[0051] III. Model Likelihood Probability Calculation
[0052] The likelihood probability of motion model i is calculated as follows:
[0053]
[0054] Furthermore, the model probability at the current time t is calculated as follows:
[0055]
[0056] Where t = max(k, d).
[0057] Furthermore, the target tracking estimation result in step six is expressed as follows:
[0058]
[0059]
[0060] Where t = max(k, d).
[0061] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, it implements the steps of the maneuvering target fusion tracking method for full-dimensional missing-dimensional disordered measurement.
[0062] Compared with the prior art, the present invention has the following technical features:
[0063] This invention achieves effective processing of out-of-order measurement data through one-step lag filtering of non-sequential measurements, optimizes the filtering update model to estimate and update multi-source full-dimensional and missing-dimensional data, and combines an interactive multi-model algorithm to complete the fusion tracking of maneuvering targets. Ultimately, it improves the utilization efficiency of out-of-order measurements from multiple heterogeneous sensors within a unified framework, thereby enhancing the tracking capability of maneuvering targets. Attached Figure Description
[0064] Figure 1 This is a comparison chart of target fusion tracking results (Scenario 1);
[0065] Figure 2 This is a comparison chart of target fusion tracking estimation accuracy (Scenario 1);
[0066] Figure 3A This is a diagram showing the full-dimensional state tracking results of the target (Scenario 2);
[0067] Figure 3B This is a diagram showing the target azimuth and elevation tracking results (Scenario 2). Detailed Implementation
[0068] The present invention will be further described below with reference to the accompanying drawings; this example is implemented based on the content of the present invention, and detailed implementation methods and specific operation processes are given, but the protection scope of the present invention is not limited to the following embodiments.
[0069] Heterogeneous sensors report full-dimensional and missing-dimensional measurement data, and the reported data exhibits out-of-order behavior within each update cycle. To address this issue, this invention presents a fusion tracking method for maneuvering targets based on full-dimensional, missing-dimensional, and out-of-order measurements. To address the out-of-order measurement data within the update cycle, a one-step lag filtering method for non-sequential measurements is introduced. Considering the full-dimensional and missing-dimensional characteristics of the measurement data, the filtering update model is optimized to achieve coverage of measurement data from various sensor configurations within a unified framework. For complex target maneuvers, a smaller, engineering-applicable model set is constructed, and an interactive multi-model algorithm is introduced. Ultimately, this achieves efficient utilization of out-of-order measurements from heterogeneous sensors, improving the tracking capability for maneuvering targets.
[0070] Specifically, it includes the following steps:
[0071] Step 1: Model condition re-initialization; The motion models used in this scheme include the uniform linear motion model and the current statistical model; After model condition re-initialization, the target state mixture estimate of the j-th model is obtained. Combined estimation with covariance j = 1, 2; details are as follows:
[0072] In the k-th update cycle (corresponding to time k), the calculated probability of the i-th model is expressed as: The corresponding target state estimates and covariance estimates are respectively and Therefore, the conditional mixture probability is:
[0073]
[0074] Where the number of models r = 2, Π ij The prior probability represents the transition probability from the i-th model to the j-th model; the mixture estimate of the target state and the mixture estimate of the covariance after reinitialization of the j-th model are:
[0075]
[0076]
[0077] The superscript T indicates matrix transpose, and the same applies below.
[0078] Step 2: Based on the received full-dimensional measurement data Or missing dimension measurement data Determine if the multi-source measurement data is out of order. If it is in order, proceed to step three; if it is out of order, proceed to step four. Details are as follows:
[0079] During this update period, new measurement data with a reporting time of d is received. The measurement data includes two types: full-dimensional and dimension-deficient, which are represented as and Judge the relationship between time d and time k. If d ≥ k, it is sequential measurement data, and go to the third step; if d < k, it is out-of-order measurement data, and go to the fourth step.
[0080] Third step: Based on the full-dimensional measurement data or dimension-deficient measurement data in the sequential measurement (d ≥ k) data, complete the model condition filtering to obtain the i-th model state estimate covariance estimate and the likelihood probability of the motion model i = 1, 2, go to the fifth step; the specific implementation is as follows:
[0081] (I) State prediction
[0082] (3.1) For the uniform linear motion model, that is, i = 1:
[0083]
[0084]
[0085] Among them, and respectively represent the state prediction matrix and covariance prediction matrix from time k to time d, represents the target motion state transition matrix from time k to time d, represents the process noise covariance from time k to time d, respectively represent the target state mixed estimate and covariance mixed estimate of the i-th model.
[0086] (3.2) For the current statistical model, that is, i = 2:
[0087]
[0088]
[0089] Among them, U k,d , are the state transition matrix, information input matrix, and mean value of maneuvering acceleration in the current statistical model. Take the acceleration value in ; respectively represent the target state mixed estimate and covariance mixed estimate of the i-th model.
[0090] (II) State update
[0091] (3.3) If full-dimensional measurement data at time d is received For motion model i (i = 1, 2), considering the computational requirements of engineering applications, an extended Kalman filter is adopted, i.e.:
[0092]
[0093]
[0094]
[0095]
[0096] in, and Let K represent the state estimate and covariance estimate of the target at time d, respectively. d Indicates gain. and Let R represent the new interest and the new interest covariance, respectively; d,rap H represents the error covariance of full-dimensional measurement data. d,rap To extend the Jacobian matrix of the Kalman filter, let If the coordinates corresponding to the position state are represented as (x, y, z), then:
[0097]
[0098] in, 0 3×6 This represents a 3x6 matrix consisting entirely of zeros.
[0099] (3.4) If the missing dimension measurement data at time d is received For motion model i (i = 1, 2), the extended Kalman filter is also used, i.e.:
[0100]
[0101]
[0102]
[0103]
[0104] in, and Let K represent the target state estimate and covariance estimate at time d, respectively. d Indicates gain. and Let R represent the new interest and the new interest covariance, respectively; d,ap Let the error covariance of the missing-dimensional measurement data be represented by the following expression. The coordinate values of the position state are represented as (x, y, z), H d,ap is:
[0105]
[0106] where
[0107] (3) Model likelihood probability calculation
[0108] Calculate the likelihood probability of motion model i as:
[0109]
[0110] Go to the fifth step.
[0111] Step 4: Based on the full-dimensional measurement data or the dimension-reduced measurement data in the out-of-order measurement (d < k) data and the innovation covariance S obtained by combining the update period k the prediction covariance P k and the measurement model H used k|k-1 complete the model conditional filtering based on the out-of-order measurement, and obtain the state estimation of each motion model k covariance estimation and the model likelihood probability i = 1, 2, go to the fifth step. i = 1, 2, go to the fifth step.
[0112] where the model conditional filtering based on the out-of-order measurement is specifically:
[0113] Let S k and P k|k-1 respectively represent the innovation covariance and the covariance prediction estimate calculated in the k-th update period, and H k represents the corresponding measurement model (such as the correspondence model between distance, azimuth, pitch and target position), then:
[0114] (1) State forward prediction
[0115] (4.1) Forward prediction of the target state from time k to time d, for motion model i:
[0116]
[0117] where when i = 1, is the state transition matrix in the uniform linear motion model, when i = 2, U d,k 、 are the state transition matrix, information input matrix and mean of maneuvering acceleration in the current statistical model.
[0118] (4.2) Forward prediction of the target state covariance from time k to time d, for model i:
[0119]
[0120] in, The process noise covariance of model i is represented by intermediate quantities. and They are respectively:
[0121]
[0122]
[0123] (II) Status Update
[0124] (4.3) Calculate the innovation covariance of the forward-predicted motion model i for full-dimensional measurement data. With missing dimension measurement data
[0125]
[0126] The covariance matrix between the target state and the measurement data at time k is:
[0127]
[0128] (4.4) Calculate the target state estimate at the current time. for:
[0129]
[0130] Current time error covariance estimation for:
[0131]
[0132] (III) Model Likelihood Probability Calculation
[0133] The likelihood probability of motion model i is calculated as follows:
[0134]
[0135] Among them, new information The calculation is as follows:
[0136]
[0137] Proceed to step five.
[0138] Step 5: Calculate the model probability at the current time t (t = max(k, d)):
[0139]
[0140] Step 6: Estimate and fuse; output the state estimate of the target at the current time t. Fusion estimation of P with covariance t|t As the result of target tracking estimation:
[0141]
[0142]
[0143] Where t = max(k, d);
[0144] Will As weights, the innovation covariance and prediction covariance of each motion model are synthesized using a convex combination method, which serves as an intermediate quantity in the fourth step of the next loop.
[0145] This invention achieves effective processing of out-of-order measurement data through one-step lag filtering of non-sequential measurements; it optimizes the filtering update model to achieve estimation and update of multi-source full-dimensional and missing-dimensional data; and it constructs a smaller, engineering-applicable model set and combines it with an interactive multi-model algorithm to achieve fusion tracking of maneuvering targets. Ultimately, it improves the utilization efficiency of out-of-order full-dimensional and missing-dimensional measurements from multiple heterogeneous sensors within a unified framework, thereby enhancing the tracking capability of maneuvering targets.
[0146] Test Instance
[0147] Scene 1
[0148] The simulation uses radar and photoelectric sensors to detect moving targets. The simulation duration is 300s. In the XYZ coordinate system, the initial position of the target is (50, 20, 9) km, the initial velocity is (-300, 150, 30) m / s, and the target's acceleration at (70, 120) s is (10, -12, 0) m / s². 2 A three-dimensional constant-speed turning motion, with an acceleration of (16, 28, 0) m / s² at (180, 215] s. 2 The system performs a three-dimensional constant-speed turning motion and an approximately uniform linear motion at other times. Radar detection data has a range error of 50m and an azimuth / elevation error of 5mrad, while electro-optical detection data has an azimuth / elevation error of 5mrad. In the simulation, the radar update cycle is used as the fusion update cycle; however, the electro-optical measurement data may arrive with a lag, resulting in measurement disorder.
[0149] After one Monte Carlo simulation, attached Figure 1The paper presents a comparison of the target maneuver trajectory, the results of the conventional method (discarding disordered measurements), and the fusion tracking results of this method. As shown in the figure, because both methods introduce interactive multi-model algorithms and optimize the filtering update model, they can fully utilize all-dimensional and multi-dimensional measurement data, thereby completing the fusion tracking estimation of large maneuvering targets.
[0150] After 200 Monte Carlo simulations, attached Figure 2 A comparison chart of the fusion tracking accuracy between the conventional method and the proposed method is presented. As can be seen from the chart, compared with the conventional method, the proposed method has a significantly improved fusion tracking accuracy due to the full utilization of disordered measurements.
[0151] Scene 2
[0152] This method is incorporated as a filtering estimation component into a cross-platform multi-radar photoelectric tracking fusion algorithm and simulation software.
[0153] The formation consists of 4 aircraft moving at approximately uniform speed in a straight line; there are 15 targets with initial X and Y positions ranging from 60 to 100 km and initial altitudes ranging from 7 to 13 km. The movement patterns include uniform straight-line motion, cooperative turning, and three-dimensional constant-speed turning. Each sensor has a different detection range, but the parameter settings are the same as in scenario one. Figure 3 shows the point fusion result after incorporating this method (where X, Y, and Z represent north, east, and ground directions, respectively). As can be seen from the figure, this method can be used as a filtering estimation component, supporting cross-platform multi-radar electro-optical full-dimensional, missing-dimensional, and disordered point fusion to form stable target trajectory data.
[0154] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A fusion tracking method for maneuvering targets oriented towards full-dimensional missing-dimensional disordered measurement, characterized in that, include: The first step is to reinitialize the model conditions; the motion models used include the uniform linear motion model and the current statistical model. After re-initializing the model conditions, the mixture estimate of the target state and the mixture estimate of the covariance are obtained; The second step is to determine whether the received full-dimensional or missing-dimensional measurement data is out of order; if it is in order, proceed to the third step; if it is out of order, proceed to the fourth step. The third step is to perform model conditional filtering based on the full-dimensional or missing-dimensional measurement data in the sequential measurement data, and obtain the motion model state estimate, covariance estimate and motion model likelihood probability, then proceed to the fifth step. The fourth step involves using full-dimensional or missing-dimensional measurement data from the disordered measurement data, combined with the information covariance and predicted covariance obtained from the update cycle, as well as the measurement model used, to complete the model conditional filtering based on disordered measurements, and obtain the state estimate, covariance estimate, and model likelihood probability of each motion model. Proceed to the fifth step. Fifth step, calculate the model probability at the current time. The sixth step is estimation fusion, which fuses the target state estimate and covariance estimate at the current moment as the target tracking result; In the first step: In the The updated cycle is calculated to obtain the first... The probabilities of each model are represented as follows: The corresponding target state estimates and covariance estimates are respectively and Therefore, the conditional mixture probability is: Among them, the number of models , Indicates the first The model to the first j The transition probabilities of the first model, prior settings; j The target state mixture estimate and covariance mixture estimate after model reinitialization are: The superscript T indicates matrix transpose, and the same applies below.
2. The method for fusion tracking of maneuvering targets based on full-dimensional missing-dimensional disordered measurement according to claim 1, characterized in that, The sixth step also includes: Using the model probability at the current moment as the weight, a convex combination method is used to fuse the innovation covariance and prediction covariance of each model, which are then used as intermediate quantities in the fourth step of the next cycle.
3. The method for fusion tracking of maneuvering targets based on full-dimensional missing-dimensional disordered measurement according to claim 1, characterized in that, The second step includes: Within this update cycle, the time of receiving the report is... The new measurement data includes two categories: full-dimensional and missing-dimensional, represented as follows: and Determine the time. With time If the relationship, For sequential measurement data, proceed to step three; if Since the measurement data is out of order, proceed to step four.
4. The method for fusion tracking of maneuvering targets for full-dimensional missing-dimensional disordered measurement according to claim 3, characterized in that, The third step includes: I. State Prediction (3.1) For the uniform linear motion model, i.e. : in, and They represent from time 1 to 2. At the time The state prediction matrix and the covariance prediction matrix, Indicates from time At the time The target motion state transition matrix, Indicates from time At the time The process noise covariance, , They represent the first i Mixed estimation of target state and mixed estimation of covariance for each model; (3.2) For the current statistical model, i.e. : in, , , For the state transition matrix, information input matrix, and mean maneuver acceleration in the current statistical model, take... The acceleration value in the text; , They represent the first i Mixed estimation of target state and mixed estimation of covariance for each model; II. Status Update (3.3) If the received time Full-dimensional measurement data For motion models ( Considering the computational requirements of engineering applications, an extended Kalman filter is adopted, namely: in, and Representing time respectively Target state estimation and covariance estimation Indicates gain. and Let the new interest and the new interest covariance be represented respectively. This represents the error covariance of full-dimensional measurement data. To extend the Jacobian matrix of the Kalman filter, let The coordinate values corresponding to the position state are represented as follows: ,but: in, , , Represents a 3x6 matrix containing all zeros; (3.4) If the received time Dimensional missing measurement data For motion models ( Similarly, it adopts the extended Kalman filter form, that is: in, and Representing time respectively Target state estimation and covariance estimation Indicates gain. and Let the new interest and the new interest covariance be represented respectively. Let the error covariance of the missing-dimensional measurement data be represented by the following expression. The coordinate values in the position state are represented as follows: , for: in, , ; III. Model Likelihood Probability Calculation Computational motion model The likelihood probability is: 。 5. The method for fusion tracking of maneuvering targets for full-dimensional missing-dimensional disordered measurement according to claim 4, characterized in that, Calculate the current time The model probability is: in, .
6. The method for fusion tracking of maneuvering targets for full-dimensional missing-dimensional disordered measurement according to claim 5, characterized in that, The target tracking estimation result in step six is expressed as follows: in, .
7. A terminal device, comprising a processor, a memory, and a computer program stored in the memory; wherein, when the processor executes the computer program, it implements the steps of the method according to any one of claims 1-6.