A distributed track fusion tracking method based on set membership filtering

Abstract

The present invention discloses a distributed track fusion tracking method based on set membership filtering, comprising the following steps: establishing three UAV motion models of uniform straight line, uniform circular and uniform acceleration; calculating the observation noise boundary of each sensor at each moment based on the observation data of all sensors; using set membership filtering to generate the optimal estimate of the local state; calculating the fusion weight according to the observation noise boundary of each sensor at each moment, fusing the local estimates, and obtaining the optimal estimate of the global state. The present invention takes into account the influence of the dynamic change of the sensor observation noise boundary, and can effectively improve the accuracy of target tracking and positioning through observation noise self-learning, local set membership filtering and global track fusion.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle target tracking, and in particular to a distributed track fusion tracking method based on set membership filtering. Background Art

[0002] Drones are gradually being used to perform reconnaissance and surveillance, electronic suppression and deception, suicide saturation attacks, etc., due to their maneuverability and high cost-effectiveness. Accordingly, an effective drone countermeasure system is particularly important. In the drone countermeasure system, target tracking is the basis. Obtaining accurate and reliable track data helps to infer the intention of the drone and assess its threat level, thus guiding the formulation of correct response decisions.

[0003] There are multiple challenges in the target tracking process. First, malicious deception interference during the tracking process will cause complex noise interference, which is not suitable for description using Gaussian distribution. Secondly, the measurement noise of the sensor is generally specified based on the sensor's own variance parameters or experience, without considering that the measurement noise is caused by the sensor's own credibility and the degree of environmental interference. If the measurement noise does not change during the entire tracking process, the filtering algorithm will lack adaptability to complex environments, which will lead to increased filtering errors. Finally, when using the weighted average method for multi-sensor track fusion, the weights are generally specified based on the sensor's observation noise. Failure to consider the dynamic changes in the measurement variance will lead to inaccurate weight allocation and reduced fusion accuracy. Summary of the invention

[0004] In view of the above-mentioned difficulties faced by UAV target tracking, the present invention proposes a distributed track fusion tracking method based on set membership filtering, which solves the problems of non-Gaussian noise, large filtering error, low fusion accuracy, etc. through observation noise self-learning, local set membership filtering and global track fusion.

[0005] The technical solution to achieve the purpose of the present invention is: a distributed track fusion tracking method based on set membership filtering, comprising the following steps:

[0006] Step 1, obtain the total number of sensors, sensor observation data, drone initial state, and drone motion type;

[0007] Step 2, establish three UAV motion models: uniform linear motion, uniform circular motion, and uniform acceleration;

[0008] Step 3, based on the observation data of all sensors, calculate the observation noise boundary of each sensor at each moment;

[0009] Step 4: according to the observation noise boundary obtained in step 3, the local state is estimated by using set membership filtering;

[0010] Step 5: Calculate the fusion weights according to the observation noise boundary of each sensor at each moment, fuse the local estimates, and obtain the optimal estimate of the global state.

[0011] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned distributed track fusion tracking method based on set membership filtering when executing the program.

[0012] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned distributed track fusion tracking method based on set membership filtering.

[0013] A computer program product includes a computer program, which implements the above-mentioned distributed track fusion tracking method based on set membership filtering when executed by a processor.

[0014] By adopting the above technical solution, the present invention can achieve the following technical effects: the three established motion models of uniform linear motion, uniform circular motion and uniform acceleration can effectively describe most of the motion situations of the UAV, and at the same time, bounded non-Gaussian noise is used to describe the observation noise of the sensor to avoid the influence of non-Gaussian noise in complex environments. At the same time, the observation noise boundary of each sensor at each moment is calculated based on the observation data of all sensors. ,according to The set membership filter is used to obtain the optimal estimate of the local state, and then according to Assign fusion weights for global fusion to effectively improve target tracking accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a principle block diagram of the multi-sensor distributed track fusion tracking of the present invention.

[0016] Figure 2 This is a schematic diagram of the drone’s track after track fusion. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0018] like Figure 1 As shown, the present invention proposes a distributed track fusion tracking method based on set membership filtering, and the specific implementation steps are as follows:

[0019] Step 1: Initialization. The operator inputs the total number of sensors, sensor observation data, drone initial state, and drone motion type.

[0020] Step 2: Establish three UAV motion models: uniform linear motion, uniform circular motion, and uniform acceleration, and simulate and generate the motion trajectories observed by each sensor. Establish the following linear system:

[0021] (1)

[0022] express The state vector at time, express Time sensor The observation vector. is the state transition matrix, is the observation matrix, yes The state transition matrix at time -1, yes The observation matrix at time . is the system noise, For sensor The observation noise is bounded non-Gaussian noise and is assumed to be contained in the following ellipsoid:

[0023] (2)

[0024] in and Represent the bounded ellipsoids where the system noise and observation noise are located respectively; and Represent the dimensions of the system noise and observation noise state space respectively; and Represent the shape and size control matrices of the system noise ellipsoid and the observation noise ellipsoid respectively; , is the identity matrix, represents the measurement noise boundary; by adjusting and Generate different motion models.

[0025] (1) Establish a uniform linear motion model. The state space of the uniform linear motion of the UAV is: , followed by Coordinates and speed in the corresponding direction. The state transfer and observation process are as follows:

[0026] (3)

[0027] in, represents the sampling interval;

[0028] (2) Establish a uniform circular motion model. The state space of the UAV’s uniform circular motion is: , representing Coordinates and speed in the corresponding direction, Represents angular velocity. The state transfer and observation process are as follows:

[0029] (4)

[0030] (3) Establish a uniform acceleration motion model. The state space of the UAV’s uniform acceleration motion is: , representing Coordinates and speed and acceleration in corresponding directions. The state transfer and observation process are as follows:

[0031] (5)

[0032] Different sensors have different detection accuracy due to the influence of the environment they are in. The observation noise boundary value Randomly select a value in [10,50]. According to the observation equation The simulation generates the motion trajectory observed by each sensor.

[0033] Step 3: Calculate the observation noise boundary of each sensor at each moment based on the observation data of all sensors . The sensors obtain the same target observations, based on The observation data of each sensor at the moment Compute the mean of all observations:

[0034] (6)

[0035] Calculate the Sensor No. The estimated distribution of measurement variance at sampling time is:

[0036] (7)

[0037] For sensor forward The arithmetic mean of the estimated variance of measurements at time points is:

[0038] (8)

[0039] calculate Moment Estimates of the variance of sensor measurements:

[0040] (9)

[0041] is the learning factor. When the external interference factor is large, should be chosen larger to make full use of prior information; when the external interference factor is small, It should be chosen to be small so that the estimation of measurement variance can capture and utilize new information in the current measurement as much as possible;

[0042] Take each sensor exist The measurement noise bound at time:

[0043] (10)

[0044] in Represents the trace of the matrix.

[0045] Step 4: Based on the result from step 3 , set membership filtering is used for local state estimation. If the ellipsoid It is known that under the condition that the noise is unknown but bounded, the state equation can be obtained make ,Right now:

[0046] (11)

[0047] here Represents an ellipsoid The ellipsoid obtained after linear transfer of the state transfer matrix is: represents the vector sum;

[0048] Find a place that contains The outer bounding ellipsoid , and satisfy , so that ,here:

[0049] (12)

[0050] in:

[0051] (13)

[0052] In the formula, and Represents the ellipsoid and The center of the ball, and Represents the ellipsoid and The shape and size of the control matrix, is a scalar to be sought;

[0053] By making The volume of the minimum is obtained to obtain the scalar in formula (13) ,Right now:

[0054] (14)

[0055] in , is the eigenvalue of the corresponding matrix, is the dimension of the state vector;

[0056] Assume that the collective state obtained by the measurement equation and the measurement noise vector is , this set can be represented as follows:

[0057] (15)

[0058] In the formula, represents the measurement noise boundary value;

[0059] Compute the minimum outer bounding ellipsoid and The linear combination of .exist time, , the recursive algorithm is as follows:

[0060] (16)

[0061] In the formula Represents the observation error, gain matrix:

[0062] (17)

[0063] Where:

[0064] (18)

[0065] yes The largest singular value of

[0066] Repeat the above steps until filtering is completed for all moments.

[0067] Step 5: Based on the observation noise boundary of each sensor at each moment Calculate the fusion weights, fuse the local estimates, and obtain the optimal estimate of the global state. Fusion is performed in a scalar weighted manner. According to formula (10), Each sensor at the time The measurement noise boundary And calculate the fusion weight as follows:

[0068] (19)

[0069] The global optimal state estimate is calculated by weighted average according to the weights as follows:

[0070] (20)

[0071] Step 6: Output , and obtain accurate and reliable track data of the UAV.

[0072] The experimental results of the present invention are as follows Figure 2 The hardware environment is as follows: CPU i7-12700H @ 2.30 GHz, memory 24G, and the observation tracks of 5 sensors are fused in total. Figure 2 The effect of drone target tracking is demonstrated. Figure 2 The left half of the figure shows the trajectory obtained by our sensors under three motion modes with noise, while the right half shows the trajectory after distributed track fusion based on set membership filtering. Figure 2 It can be seen that after track fusion, the target tracking accuracy can be effectively improved.

[0073] Bounded non-Gaussian noise is used to describe the sensor's observation noise to avoid the impact of non-Gaussian noise in complex environments. The observation noise boundary of each sensor at each moment is calculated based on the observation data of all sensors. ,according to The set membership filter is used to obtain the optimal estimate of the local state, and then according to Assigning fusion weights for global fusion can effectively improve target tracking accuracy.

[0074] The present invention provides a distributed track fusion tracking method based on set membership filtering. There are many methods and ways to implement the technical solution. The above is only a preferred implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A distributed track fusion tracking method based on set membership filtering, characterized in that: The steps include: Step 1, obtain the total number of sensors, sensor observation data, drone initial state, and drone motion type; Step 2: Establish three UAV motion models: uniform linear motion, uniform circular motion, and uniform acceleration motion. Specifically, they include: Step 2-1, establish the following linear system: （1） express The state vector at time, express Time sensor The observation vector of yes The state transition matrix at time -1, yes The observation matrix at time, is the system noise, For sensor The observation noise is bounded non-Gaussian noise and is assumed to be contained in the following ellipsoid: （2） in and Represent the bounded ellipsoids where the system noise and observation noise are located respectively; and Represent the dimensions of the system noise and observation noise state space respectively; and Represent the shape and size control matrices of the system noise ellipsoid and the observation noise ellipsoid respectively; , is the identity matrix, represents the measurement noise boundary; by adjusting and Generate different motion models; Step 2-2, establish a uniform linear motion model; the state space of the uniform linear motion of the UAV is: , followed by Coordinates and speed in the corresponding direction; the state transfer and observation process are as follows: （3） in, represents the sampling interval; Step 2-3, establish a uniform circular motion model; the state space of the UAV uniform circular motion is , representing Coordinates and speed in the corresponding direction, Represents angular velocity; the state transfer and observation process are as follows: （4） Step 2-4, establish a uniform acceleration motion model; the state space of the uniform acceleration motion of the drone is , representing Coordinates and speed and acceleration in corresponding directions; state transfer and observation process are as follows: （5） Step 2-5: Different sensors have different detection accuracy due to the environment they are in. Set the sensor The observation noise boundary value Randomly select values ​​in [10,50]; according to the observation equation The simulation generates the motion trajectory observed by each sensor; Step 3, based on the observation data of all sensors, calculate the observation noise boundary of each sensor at each moment, specifically including: Step 3-1, The sensors obtain the same target observations, based on The observation data of each sensor at the moment Compute the mean of all observations: （6） Step 3-2, calculate the Sensor No. The estimated distribution of measurement variance at sampling time is: （7） Step 3-3, sensor forward The arithmetic mean of the estimated variance of measurements at time points is: （8） Step 3-4, calculate Moment Estimates of the variance of sensor measurements: （9） is the learning factor; Step 3-5, take each sensor exist The measurement noise bound at time: （10） in represents the trace of the matrix; Step 4: according to the observation noise boundary obtained in step 3, the local state is estimated by using set membership filtering; Step 5: Calculate the fusion weights according to the observation noise boundary of each sensor at each moment, fuse the local estimates, and obtain the optimal estimate of the global state.

2. The distributed track fusion tracking method based on set membership filtering according to claim 1 is characterized in that: Step 4 includes: Step 4-1, if the ellipsoid It is known that under the condition that the noise is unknown but bounded, the state equation can be obtained make ,Right now: （11） here Represents an ellipsoid The ellipsoid obtained after linear transfer of the state transfer matrix is: represents the vector sum; Step 4-2, find a The outer bounding ellipsoid , and satisfy , so that ,here: （12） in: （13） In the formula, and Represents the ellipsoid and The center of the ball, and Represents the ellipsoid and The shape and size of the control matrix, is a scalar to be sought; Step 4-3, by using The volume of the minimum is obtained to obtain the scalar in formula (13) ,Right now: （14） in , is the eigenvalue of the corresponding matrix, is the dimension of the state vector; Step 4-4, assume that the collective state obtained by the measurement equation and the measurement noise vector is , this set can be represented as follows: （15） In the formula, Represents the measurement noise boundary value; Step 4-5, calculate the minimum outer bounding ellipsoid and The linear combination of ;exist time, , the recursive algorithm is as follows: （16） In the formula Represents the observation error, gain matrix: （17） Where: （18） yes The largest singular value of Step 4-6, repeat steps 4-1 to 4-5 until filtering is completed for all moments.

3. The distributed track fusion tracking method based on set membership filtering according to claim 2 is characterized in that: Step 5 includes: Step 5-1, perform fusion in a scalar weighted manner, and obtain the following equation (10): Each sensor at the time The measurement noise boundary The fusion weights are calculated as follows: （19） Step 5-2, calculate the global optimal state estimate by weighted average according to the weights as follows: （20）。 4. The distributed track fusion tracking method based on set membership filtering according to claim 3 is characterized in that: The method also includes step 6, specifically: output , get the UAV track data.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the distributed track fusion tracking method based on set membership filtering as described in any one of claims 1 to 4 is implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the distributed track fusion tracking method based on set membership filtering as described in any one of claims 1 to 4 is implemented.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the distributed track fusion tracking method based on set membership filtering described in any one of claims 1 to 4 is implemented.

Images