An interactive multi-mode track filtering method based on error backtracking and pseudo-measurement completion

Abstract

The application relates to the technical field of target tracking, and discloses an interactive multi-mode track filtering method based on error tracing and pseudo-measurement completion. The method comprises the following steps: inputting measurement data without sensor relative system deviation as an observation value of an interactive multi-mode filter; filtering the measurement data without sensor relative system deviation by using multiple different motion models; evaluating the probability of each motion model fitting the real track of a target in real time and taking the probability as a weight; weighting and fusing the filtering results of the corresponding motion models by using the weight; and outputting the weighted fusion result. Based on the weighted fusion result, the track of the target is predicted by using the interactive multi-mode filter, the track prediction result is used for completing the dimension-lacking measurement vector, the completed vector is input into the interactive multi-mode filter, and the weighted fusion result at the latest moment is obtained. The application improves the track filtering precision.

Description

Technical Field

[0001] This application relates to the field of target tracking technology, and in particular to an interactive multi-mode track filtering method based on error tracing and pseudo-measurement completion. Background Technology

[0002] Multi-sensor target composite tracking has always been a key research area in information fusion. To achieve high-precision and stable target tracking, it is necessary to establish a motion model that matches the actual motion of the target, an error level that matches the input and observation features, and an observation equation adapted to the measurement dimensions. The Interactive Multi-Mode (IMM) filtering algorithm maps the possible motion states of the target to multiple models, continuously adjusting the probabilities of each model (summing to 1) based on the target's historical trajectory. The final parameter estimate is a weighted sum of the filtered estimates from each model. Generally, IMM exhibits good compatibility between tracking performance and stability.

[0003] In engineering applications, observation noise is often difficult to measure accurately. The main reasons are: (1) the limitations of the measurement principle, some targets have many echo points and frequent jump phenomena, resulting in very unstable errors; (2) some target tracks are processed by coordinate transformation and filtering, which causes the error to be non-white noise and distorted (such as the wrong motion model used for filtering); (3) factors such as vibration generated during the flight of the aircraft platform cause the sensors to gradually produce a certain amount of system deviation. The degree of influence of multiple sensors may be different. If the observation values ​​affected by different system deviations are input into the same filter, it will have disastrous consequences, such as filter divergence.

[0004] Traditional differential-dimensional sensor composite tracking filtering algorithms, such as collaborative tracking using radar 3D measurements and optical radar vector measurements, often require constructing different measurement equations for different sensor measurements. When one measurement arrives, because the state variables are consistent, the same prediction step can be performed first. However, during measurement updates, state updates and Jacobian matrix solutions are performed based on each measurement and its corresponding measurement matrix. Often, when 3D measurements are missing for extended periods, relying solely on these missing measurements leads to a rapid and significant amplification of the filtered track error, especially when the position component is missing. Summary of the Invention

[0005] In view of this, this application provides an interactive multi-mode trajectory filtering method based on error source tracing and pseudo-measurement completion, which is not affected by sensor measurement dimension loss, measurement error and target maneuvering, and has high filtering accuracy.

[0006] This application discloses an interactive multi-mode track filtering method based on error source tracing and pseudo-measurement completion, which includes: Step 1: Receive time-series measurement data of the target trajectory from multiple sensors, perform error source analysis on the measurement data to obtain measurement data without sensor-relative system bias and the random noise level of the measurement data, and use the measurement data without sensor-relative system bias as the input observation value of the interactive multimode filter; the random noise level of the measurement data is used to calculate the noise covariance matrix of the observation value of the interactive multimode filter. The measurement data includes distance, azimuth, and elevation angles. Step 2: The measurement data without sensor-dependent system bias are filtered using multiple different motion models configured by the interactive multimode filter. The measurement noise parameter is represented by the noise covariance matrix. The probability of each motion model fitting the target's true trajectory is evaluated online by the deviation between the trajectory prediction results of each motion model and the actual measurement results of the updated measurement, and this deviation is used as a weight. The filtering results of the corresponding motion models are then weighted and fused using the weights, and the weighted fusion result is output as the trajectory filtering result output by the interactive multimode filter. Step 3: Based on the weighted fusion result, the target trajectory is predicted using the interactive multimode filter. The predicted time is the measurement time corresponding to the latest measurement data. The missing dimension measurement vector is completed using the trajectory prediction result. The completed vector is then input into the interactive multimode filter to obtain the weighted fusion result of the latest time. The missing dimension measurement vector refers to a vector that is missing one or two of the range, azimuth, and pitch angles.

[0007] Furthermore, the step of performing error source tracing statistics on the measurement data to obtain measurement data without sensor-relative system bias and the random noise level of the measurement data includes: The measurement data is subjected to relative system bias elimination to obtain measurement data without sensor relative system bias; The random noise level of the measurement data is estimated.

[0008] Furthermore, the process of eliminating relative system bias in the measurement data to obtain measurement data free from sensor relative system bias includes: In multi-sensor measurement, coordinate transformation is used to convert the measurement data of all sensors on the target to the body spherical coordinate system. The measurement deviations of the first sensor and the highest priority sensor in the multi-sensor system for the target are calculated in the body spherical coordinate system. The measurement deviations include distance deviation, azimuth deviation, and pitch deviation. The first sensor is all the sensors in the multi-sensor system except for the highest priority sensor. The body spherical coordinate system is a spherical coordinate system with the aircraft as the origin, and the target position is described by distance, azimuth, and pitch. The system bias estimate is used to register and subtract the measurement data corresponding to the first sensor to obtain measurement data without sensor-related system bias.

[0009] Furthermore, the process of obtaining the estimated system deviation is as follows: Assuming that the relative system bias of all sensors in a multi-sensor system is stable, the system bias estimate can be obtained using the following formula:

[0010] in, , , They represent the first The systematic bias estimates of the distance, azimuth, and elevation angles at the time of the shot. They represent the first Distance deviation, azimuth deviation, and pitch deviation at the time of shooting. , , They represent the first Systematic bias estimates of distance, azimuth, and elevation angles at the time of shooting; The measurement data without sensor-related system bias can be obtained using the following formula:

[0011] in, They represent the first The distance, azimuth, and pitch angles at the moment of shooting are not relative to the system bias of the sensor. , , They represent the first The distance, azimuth, and pitch angle corresponding to the first sensor at the moment of shooting.

[0012] Further, the step of estimating the random noise level of the measurement data to obtain the random noise level of the measurement data includes: Step 101: Use the seven-point moving average method to perform moving average processing on all historical measurement data of the single sensor in the multi-sensor system to obtain the average value at the current moment. Use the average value at the current moment as the benchmark value for calculating the random jitter level of the measurement data.

[0013] in, , , They represent the first The moving averages of distance, azimuth, and elevation angles at the time of shooting. , They represent the first The original measurements of distance, azimuth, and elevation angles at the time of shooting. It is a positive integer, and its value ranges from 1 to 7; Step 102: Select the one closest to the current time. Within the shooting time window, calculate the random noise levels for distance, azimuth, and elevation angles:

[0014] in, , , They represent the first Random noise levels at the time of shooting, distance, azimuth, and elevation angles; They represent the first The original measurements of distance, azimuth, and pitch angles taken by the sensor at the moment of shooting. The total number of frames used for the moving average is generally selected as 20-50 frames; Step 103: When the next frame arrives, the time window for processing measurement data is pushed forward by one frame. Steps 101 and 102 are repeated until the random noise levels of distance, azimuth, and elevation angles of all single-sensor historical measurement data in the multi-sensor system are obtained.

[0015] Furthermore, the various motion models include a uniformly accelerated linear motion model, a uniform circular motion model, and a Singer motion model; Before performing filtering, the measurement data acquired by multiple sensors are uniformly converted to the body spherical coordinate system; The existing extended Kalman filter method is used to perform innovation estimation, filtering and anti-divergence processing under each motion model. The existing interactive multimode filter algorithm is used to calculate the probability update of each motion model. The track filtering results of each motion model are weighted and fused based on the updated probability to obtain the weighted fusion result. The weighted fusion result includes a new state estimate and covariance matrix, which is used as the track filtering result output by the interactive multimode filter in the previous step.

[0016] Further, step 3 includes: Step 301: Use multiple different motion models in the interactive multimode filter to predict the target trajectory for the next step, and obtain the trajectory prediction value of each motion model at the prediction time; Step 302: Transform the trajectory prediction values ​​of each motion model to the airframe spherical coordinate system to form the corresponding prediction vector; where, the first... The prediction vector includes the one with the first The trajectory prediction value of the first motion model corresponds to the first... The distance of the moment of shooting Azimuth and pitch angle ; Step 303: Compare the predicted vector with the missing dimension measurement vector to obtain the difference. Based on the difference, obtain the weights corresponding to each motion model. The difference includes at least one of the following: distance difference, azimuth difference, and pitch difference. Step 304: Based on the weights corresponding to each motion model and the trajectory prediction values ​​of each motion model at the prediction time, calculate the pseudo-measurement data, which is used to complete the missing dimension measurement vector. Obtain the completed vector and use it as the input of the observation value at the prediction time of the interactive multimode filter.

[0017] Further, step 301 includes: Using the three motion models in the interactive multimode filter, the target trajectory is predicted for the next step:

[0018] in, Indicates the first Shooting time, i.e., the next step The trajectory prediction values ​​of each motion model Indicates the first The target trajectory motion state of the motion model starts from the first... Shooting time to the number The state transition matrix at each time step. Indicates the first The filtered value obtained by weighted fusion of the filtering results of the three motion models at each frame time is the weighted fusion result. The trajectory prediction values ​​of each motion model Transform to the aircraft spherical coordinate system to obtain the distance, azimuth, and pitch angles corresponding to the predicted trajectory, denoted as follows: , and .

[0019] Further, step 303 includes: When the missing dimension measurement vector lacks distance, the azimuth and elevation angles predicted by the three motion models are subtracted from the azimuth and elevation angles measured by the actual sensors, respectively, to obtain the azimuth angle difference. and pitch angle difference ; The weights of the three motion models are calculated using the following formulas:

[0020] in, For the first The weights of each motion model, , The first Shooting Time The difference between the azimuth and elevation angles predicted by the model and the azimuth and elevation angles measured by the actual sensors. Indicates the first Shooting Time The difference between the azimuth angle predicted by the model and the azimuth angle measured by the actual sensor. Indicates the first Shooting Time The difference between the pitch angle predicted by the model and the pitch angle measured by the actual sensor. When the missing dimension measurement vector lacks range and pitch angle, the difference is obtained by subtracting the azimuth angle predicted by the three motion models from the azimuth angle measured by the actual sensor. The formulas for calculating the weights of the three motion models are as follows: .

[0021] Further, step 304 includes: Step 3041: Calculate the weighted predicted value using the following formula:

[0022] in, For the first The weighted predicted value based on the time of capture; Step 3042: Calculate the weighted predicted values. After performing a coordinate transformation to the aircraft's spherical coordinate system, pseudo-measurement data are obtained, namely the distance, azimuth, and pitch angles corresponding to the weighted predicted values, which are denoted as follows: , , ; Step 3043: Select and The missing distance and azimuth values ​​corresponding to the missing dimension measurement vector are used to replace the missing distance and azimuth values ​​in the missing dimension measurement vector, thus completing the missing distance and azimuth values. This completed vector is then used as the first value of the interactive multimode filter. Input the observed values ​​at the time of shooting.

[0023] Due to the adoption of the above technical solution, this application has the following advantages: 1. This application is not affected by missing dimensions in sensor measurements: by constructing pseudo-measurements to replace missing distances or pitch angles, the input of the newly captured time measurement filtering observations is still full-dimensional.

[0024] 2. This application is not affected by sensor measurement error distortion: By using error tracing methods, the deviation between other sensor measurements and the highest priority sensor measurement is eliminated, preventing the phenomenon of "the more it is integrated, the worse it becomes"; the actual noise level of multi-sensor measurements is estimated online to obtain an accurate filtered observation noise matrix, thereby improving filtering accuracy and target tracking accuracy.

[0025] 3. This application is not affected by unknown target maneuvers: By adopting an interactive multimode filter framework, three different typical maneuver motion models are introduced. Based on historical observations, the filter automatically adjusts the probability of each motion model to adapt to the real-time maneuver trajectory of the target. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0027] Figure 1 This is a flowchart illustrating an interactive multi-mode track filtering method based on error tracing and pseudo-measurement completion, according to an embodiment of this application. Detailed Implementation

[0028] The present application will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only some, not all, of the embodiments of the present application. All other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of the present application.

[0029] In multi-sensor collaborative tracking scenarios, the main technical challenges include differences in measurement dimensions between dissimilar sensors, the influence of measurement system bias / random errors, and the unknown target motion model. Based on this, see [reference needed]. Figure 1 This application proposes an embodiment of an interactive multi-mode track filtering method based on error source tracing and pseudo-measurement completion, which includes: Step 1: Receive time-series measurement data of the target trajectory from multiple sensors, perform error source analysis on the measurement data to obtain measurement data without sensor-relative system bias and the random noise level of the measurement data, and use the measurement data without sensor-relative system bias as the input observation value of the interactive multimode filter; the random noise level of the measurement data is used to calculate the noise covariance matrix of the observation value of the interactive multimode filter. The measurement data includes distance, azimuth, and elevation angles. Step 2: The measurement data without sensor-dependent system bias are filtered using multiple different motion models configured by the interactive multimode filter. The measurement noise parameter is represented by the noise covariance matrix. The probability of each motion model fitting the target's true trajectory is evaluated online by the deviation between the trajectory prediction results of each motion model (predicting the same time as the updated measurement) and the actual measurement results of the updated measurement, and this deviation is used as a weight. The filtering results of the corresponding motion models are then weighted and fused using the weights, and the weighted fusion result is output as the trajectory filtering result of the interactive multimode filter. Step 3: Based on the weighted fusion result, the target trajectory is predicted using the interactive multimode filter. The predicted time is the measurement time corresponding to the latest measurement data. The missing dimension measurement vector is completed using the trajectory prediction result. The completed vector is then input into the interactive multimode filter to obtain the weighted fusion result of the latest time. The missing dimension measurement vector refers to a vector that is missing one or two of the range, azimuth, and pitch angles.

[0030] Optionally, the step of performing error source tracing statistics on the measurement data to obtain measurement data without sensor-relative system bias and the random noise level of the measurement data includes: The measurement data is subjected to relative system bias elimination to obtain measurement data without sensor relative system bias; The random noise level of the measurement data is estimated.

[0031] Optionally, the step of eliminating relative system bias in the measurement data to obtain measurement data without sensor relative system bias includes: In multi-sensor measurement, coordinate transformation is used to convert the measurement data of all sensors on the target to the body spherical coordinate system. The measurement deviations of the first sensor and the highest priority sensor in the multi-sensor system for the target are calculated in the body spherical coordinate system. The measurement deviations include distance deviation, azimuth deviation, and pitch deviation. The first sensor is all the sensors in the multi-sensor system except for the highest priority sensor. The body spherical coordinate system is a spherical coordinate system with the aircraft as the origin, and the target position is described by distance, azimuth, and pitch. The system bias estimate is used to register and subtract the measurement data corresponding to the first sensor to obtain measurement data without sensor-related system bias.

[0032] Optionally, the process of obtaining the system deviation estimate is as follows: Assuming that the relative system bias of all sensors in a multi-sensor system is stable, the system bias estimate can be obtained using the following formula:

[0033] in, , , They represent the first The systematic bias estimates of the distance, azimuth, and elevation angles at the time of the shot. They represent the first Distance deviation, azimuth deviation, and pitch deviation at the time of shooting. , , They represent the first Systematic bias estimates of distance, azimuth, and elevation angles at the time of shooting; The measurement data without sensor-related system bias can be obtained using the following formula:

[0034] in, They represent the first The distance, azimuth, and pitch angles at the moment of shooting are not relative to the system bias of the sensor. , , They represent the first The distance, azimuth, and pitch angle corresponding to the first sensor at the moment of shooting.

[0035] Optionally, the step of estimating the random noise level of the measurement data to obtain the random noise level of the measurement data includes: Step 101: Use the seven-point moving average method to perform moving average processing on all historical measurement data of the single sensor in the multi-sensor system to obtain the average value at the current moment. Use the average value at the current moment as the benchmark value for calculating the random jitter level of the measurement data.

[0036] in, , , They represent the first The moving averages of distance, azimuth, and elevation angles at the time of shooting. , They represent the first The original measurements of distance, azimuth, and elevation angles at the time of shooting. It is a positive integer, and its value ranges from 1 to 7; Step 102: Select the one closest to the current time. Within the shooting time window, calculate the random noise levels for distance, azimuth, and elevation angles:

[0037] in, , , They represent the first Random noise levels at the time of shooting, distance, azimuth, and elevation angles; They represent the first The original measurements of distance, azimuth, and pitch angles taken by the sensor at the moment of shooting. The total number of frames used for the moving average is generally selected as 20-50 frames; Step 103: When the next frame arrives, the time window for processing measurement data is pushed forward by one frame. Steps 101 and 102 are repeated until the random noise levels of distance, azimuth, and elevation angles of all single-sensor historical measurement data in the multi-sensor system are obtained.

[0038] Optionally, the plurality of different motion models include a uniformly accelerated linear motion model, a uniform circular motion model, and a Singer motion model; Before performing filtering, the measurement data acquired by multiple sensors are uniformly converted to the body spherical coordinate system; The existing extended Kalman filter method is used to perform innovation estimation, filtering and anti-divergence processing under each motion model. The existing interactive multimode filter algorithm is used to calculate the probability update of each motion model. The track filtering results of each motion model are weighted and fused based on the updated probability to obtain the weighted fusion result. The weighted fusion result includes a new state estimate and covariance matrix, which is used as the track filtering result output by the interactive multimode filter in the previous step.

[0039] Optionally, step 3 includes: Step 301: Use multiple different motion models in the interactive multimode filter to predict the target trajectory for the next step, and obtain the trajectory prediction value of each motion model at the prediction time; Step 302: Transform the trajectory prediction values ​​of each motion model to the airframe spherical coordinate system to form the corresponding prediction vector; where, the first... The prediction vector includes the one with the first The trajectory prediction value of the first motion model corresponds to the first... The distance of the moment of shooting Azimuth and pitch angle ; Step 303: Compare the predicted vector with the missing dimension measurement vector to obtain the difference. Based on the difference, obtain the weights corresponding to each motion model. The difference includes at least one of the following: distance difference, azimuth difference, and pitch difference. Step 304: Based on the weights corresponding to each motion model and the trajectory prediction values ​​of each motion model at the prediction time, calculate the pseudo-measurement data, which is used to complete the missing dimension measurement vector. Obtain the completed vector and use it as the input of the observation value at the prediction time of the interactive multimode filter.

[0040] Optionally, step 301 includes: Using the three motion models in the interactive multimode filter, the target trajectory is predicted for the next step:

[0041] in, Indicates the first Shooting time, i.e., the next step The trajectory prediction values ​​of each motion model Indicates the first The target trajectory motion state of the motion model starts from the first... Shooting time to the number The state transition matrix at each time step. Indicates the first The filtered value obtained by weighted fusion of the filtering results of the three motion models at each frame time is the weighted fusion result. The trajectory prediction values ​​of each motion model Transform to the aircraft spherical coordinate system to obtain the distance, azimuth, and pitch angles corresponding to the predicted trajectory, denoted as follows: , and .

[0042] Optionally, step 303 includes: When the missing dimension measurement vector lacks distance, the azimuth and elevation angles predicted by the three motion models are subtracted from the azimuth and elevation angles measured by the actual sensors, respectively, to obtain the azimuth angle difference. and pitch angle difference ; The weights of the three motion models are calculated using the following formulas:

[0043] in, For the first The weights of each motion model, , The first Shooting Time The difference between the azimuth and elevation angles predicted by the model and the azimuth and elevation angles measured by the actual sensors. Indicates the first Shooting Time The difference between the azimuth angle predicted by the model and the azimuth angle measured by the actual sensor. Indicates the first Shooting Time The difference between the pitch angle predicted by the model and the pitch angle measured by the actual sensor. When the missing dimension measurement vector lacks range and pitch angle, the difference is obtained by subtracting the azimuth angle predicted by the three motion models from the azimuth angle measured by the actual sensor. The formulas for calculating the weights of the three motion models are as follows: .

[0044] Optionally, step 304 includes: Step 3041: Calculate the weighted predicted value using the following formula:

[0045] in, For the first The weighted predicted value based on the time of capture; Step 3042: Calculate the weighted predicted values. After performing a coordinate transformation to the aircraft's spherical coordinate system, pseudo-measurement data are obtained, namely the distance, azimuth, and pitch angles corresponding to the weighted predicted values, which are denoted as follows: , , ; Step 3043: Select and The missing distance and azimuth values ​​corresponding to the missing dimension measurement vector are used to replace the missing distance and azimuth values ​​in the missing dimension measurement vector, thus completing the missing distance and azimuth values. This completed vector is then used as the first value of the interactive multimode filter. Input the observed values ​​at the time of shooting.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.

Claims

1. An interactive multi-mode trajectory filtering method based on error source tracing and pseudo-measurement completion, characterized in that, include: Step 1: Receive time-series measurement data of the target trajectory from multiple sensors, perform error source analysis on the measurement data to obtain measurement data without sensor-relative system bias and the random noise level of the measurement data, and use the measurement data without sensor-relative system bias as the input observation value of the interactive multimode filter; the random noise level of the measurement data is used to calculate the noise covariance matrix of the observation value of the interactive multimode filter. The measurement data includes distance, azimuth, and elevation angles. Step 2: The measurement data without sensor-dependent system bias are filtered using multiple different motion models configured by the interactive multimode filter. The measurement noise parameter is represented by the noise covariance matrix. The probability of each motion model fitting the target's true trajectory is evaluated online by the deviation between the trajectory prediction results of each motion model and the actual measurement results of the updated measurement, and this deviation is used as a weight. The filtering results of the corresponding motion models are then weighted and fused using the weights, and the weighted fusion result is output as the trajectory filtering result output by the interactive multimode filter. Step 3: Based on the weighted fusion result, the target trajectory is predicted using the interactive multimode filter. The prediction time is the measurement time corresponding to the latest measurement data. The missing dimension measurement vector is completed using the trajectory prediction result. The completed vector is then input into the interactive multimode filter to obtain the weighted fusion result of the latest time. The missing dimension measurement vector refers to the vector that is missing one or two of the range, azimuth, and pitch angles. Step 3 includes: Step 301: Use multiple different motion models in the interactive multimode filter to predict the target trajectory for the next step, and obtain the trajectory prediction value of each motion model at the prediction time; Step 302: Transform the trajectory prediction values ​​of each motion model to the airframe spherical coordinate system to form the corresponding prediction vector; where, the first... The prediction vector includes the one with the first The trajectory prediction value of the first motion model corresponds to the first... The distance of the moment of shooting Azimuth and pitch angle ; Step 303: Compare the predicted vector with the missing dimension measurement vector to obtain the difference. Based on the difference, obtain the weights corresponding to each motion model. The difference includes at least one of the following: distance difference, azimuth difference, and pitch difference. Step 304: Based on the weights corresponding to each motion model and the trajectory prediction values ​​of each motion model at the prediction time, calculate the pseudo-measurement data, which is used to complete the missing dimension measurement vector, obtain the completed vector, and use it as the input of the observation value at the prediction time of the interactive multimode filter; Step 303 includes: When the missing dimension measurement vector lacks distance, the azimuth and elevation angles predicted by the three motion models are subtracted from the azimuth and elevation angles measured by the actual sensors, respectively, to obtain the azimuth angle difference. and pitch angle difference ; The weights of the three motion models are calculated using the following formulas: in, For the first The weights of each motion model, , The first Shooting Time The difference between the azimuth and elevation angles predicted by the model and the azimuth and elevation angles measured by the actual sensors. Indicates the first Shooting Time The difference between the azimuth angle predicted by the model and the azimuth angle measured by the actual sensor. Indicates the first Shooting Time The difference between the pitch angle predicted by the model and the pitch angle measured by the actual sensor. When the missing dimension measurement vector lacks range and pitch angle, the difference is obtained by subtracting the azimuth angle predicted by the three motion models from the azimuth angle measured by the actual sensor. The formulas for calculating the weights of the three motion models are as follows: ； Step 304 includes: Step 3041: Calculate the weighted predicted value using the following formula: in, For the first The weighted predicted value based on the time of capture. Indicates the first Shooting time, i.e., the next step The trajectory prediction values ​​of each motion model; Step 3042: Calculate the weighted predicted values. After performing a coordinate transformation to the aircraft's spherical coordinate system, pseudo-measurement data are obtained, namely the distance, azimuth, and pitch angles corresponding to the weighted predicted values, which are denoted as follows: , , ; Step 3043: Select and The missing distance and azimuth values ​​corresponding to the missing dimension measurement vector are used to replace the missing distance and azimuth values ​​in the missing dimension measurement vector, thus completing the missing distance and azimuth values. This completed vector is then used as the first value of the interactive multimode filter. Input the observed values ​​at the time of shooting.

2. The method according to claim 1, characterized in that, The step of performing error source tracing statistics on the measurement data to obtain measurement data without sensor-relative system bias and the random noise level of the measurement data includes: The measurement data is subjected to relative system bias elimination to obtain measurement data without sensor relative system bias; The random noise level of the measurement data is estimated.

3. The method according to claim 2, characterized in that, The process of eliminating relative systematic bias in the measurement data to obtain measurement data free from sensor relative systematic bias includes: In multi-sensor measurement, coordinate transformation is used to convert the measurement data of all sensors on the target to the body spherical coordinate system. The measurement deviations of the first sensor and the highest priority sensor in the multi-sensor system for the target are calculated in the body spherical coordinate system. The measurement deviations include distance deviation, azimuth deviation, and pitch deviation. The first sensor is all the sensors in the multi-sensor system except for the highest priority sensor. The body spherical coordinate system is a spherical coordinate system with the aircraft as the origin, and the target position is described by distance, azimuth, and pitch. The system bias estimate is used to register and subtract the measurement data corresponding to the first sensor to obtain measurement data without sensor-related system bias.

4. The method according to claim 3, characterized in that, The process of obtaining the system deviation estimate is as follows: Assuming that the relative system bias of all sensors in a multi-sensor system is stable, the system bias estimate can be obtained using the following formula: in, , , They represent the first The systematic bias estimates of the distance, azimuth, and elevation angles at the time of the shot. They represent the first Distance deviation, azimuth deviation, and pitch deviation at the time of shooting. , , They represent the first Systematic bias estimates of distance, azimuth, and elevation angles at the time of shooting; The measurement data without sensor-related system bias can be obtained using the following formula: in, They represent the first The distance, azimuth, and pitch angles at the moment of shooting are not relative to the system bias of the sensor. , , They represent the first The distance, azimuth, and pitch angle corresponding to the first sensor at the moment of shooting.

5. The method according to claim 2, characterized in that, The step of estimating the random noise level of the measurement data to obtain the random noise level of the measurement data includes: Step 101: Use the seven-point moving average method to perform moving average processing on all historical measurement data of the single sensor in the multi-sensor system to obtain the average value at the current moment. Use the average value at the current moment as the benchmark value for calculating the random jitter level of the measurement data. in, , , They represent the first The moving averages of distance, azimuth, and elevation angles at the time of shooting. , They represent the first The original measurements of distance, azimuth, and elevation angles at the time of shooting. It is a positive integer; Step 102: Select the one closest to the current time. Within the shooting time window, calculate the random noise levels for distance, azimuth, and elevation angles: in, , , They represent the first Random noise levels at the time of shooting, distance, azimuth, and elevation angles; They represent the first The original measurements of distance, azimuth, and pitch angles taken by the sensor at the moment of shooting. The total number of frames used for the moving average; Step 103: When the next frame arrives, the time window for processing measurement data is pushed forward by one frame. Steps 101 and 102 are repeated until the random noise levels of distance, azimuth, and elevation angles of all single-sensor historical measurement data in the multi-sensor system are obtained.

6. The method according to claim 1, characterized in that, The various motion models include a uniformly accelerated linear motion model, a uniform circular motion model, and a Singer motion model; Before performing filtering, the measurement data acquired by multiple sensors are uniformly converted to the body spherical coordinate system; The existing extended Kalman filter method is used to perform innovation estimation, filtering and anti-divergence processing under each motion model. The existing interactive multimode filter algorithm is used to calculate the probability update of each motion model. The track filtering results of each motion model are weighted and fused based on the updated probability to obtain the weighted fusion result. The weighted fusion result includes a new state estimate and covariance matrix, which is used as the track filtering result output by the interactive multimode filter in the previous step.

7. The method according to claim 1, characterized in that, Step 301 includes: Using the three motion models in the interactive multimode filter, the target trajectory is predicted for the next step: in, Indicates the first Shooting time, i.e., the next step The trajectory prediction values ​​of each motion model Indicates the first The target trajectory motion state of the motion model starts from the first... Shooting time to the number The state transition matrix at each time step. Indicates the first The filtered value obtained by weighted fusion of the filtering results of the three motion models at each frame time is the weighted fusion result. The trajectory prediction values ​​of each motion model Transform to the aircraft spherical coordinate system to obtain the distance, azimuth, and pitch angles corresponding to the predicted trajectory, denoted as follows: , and .

Images