Unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA

Abstract

The invention discloses an unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA, and belongs to the technical field of unmanned aerial vehicle monitoring and tracking. According to the method, the problem that continuous and stable tracking, high-precision positioning and target identity recognition are difficult to consider in a complex urban environment by an existing single detection means is solved, and the continuous position finding capability of a radar and the passive identity recognition capability of TDOA are deeply coupled by constructing an intelligent data fusion framework; and by adopting a hybrid association and optimal fusion algorithm, stable and accurate continuous tracking of the unmanned aerial vehicle target with identity information is realized, and the situation awareness and decision-making capability of an anti-unmanned aerial vehicle system is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle monitoring and tracking, in particular to an unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA. BACKGROUND

[0002] In the modern complex urban electromagnetic and physical environment, a single unmanned aerial vehicle detection sensor has inherent shortcomings. Pulse Doppler radar or phased array radar can provide stable, continuous, and high refresh rate three-dimensional tracks (range, azimuth, and elevation), and has direct speed measurement capability for targets, which is the cornerstone of realizing continuous tracking. However, its active electromagnetic wave emission characteristics make it susceptible to interference, and in a dense target environment, track confusion is easy to occur. More importantly, the radar echo cannot carry the identity information of the target.

[0003] Complementary to this is the TDOA radio detection system. This system passively intercepts the communication radio frequency signals such as image transmission and remote control between the unmanned aerial vehicle and the remote controller through multiple receiving stations deployed on the ground, and realizes passive positioning of the unmanned aerial vehicle by calculating the time difference of the signals arriving at different stations. Its core advantages are: identity recognition potential: by analyzing the fingerprint characteristics (such as specific frequency bands, modulation methods) of the radio frequency signals or directly decoding the serial numbers (such as DJI unmanned aerial vehicle SN code) in the data packets, a unique identity (ID) can be assigned to the target. Passive concealment: it does not actively emit electromagnetic waves and is difficult to be detected and counteracted. Countermeasures: the detection capability for low-altitude, slow-speed, and small-size targets is usually better than that of radar.

[0004] However, the positioning accuracy of the TDOA system is severely dependent on the geometric configuration between the base station and the target (i.e. the precision attenuation factor, DOP), resulting in uneven (anisotropic) error ellipses in all directions, and when the signal is intermittently transmitted, the data stream is discontinuous; therefore, it does not meet the existing requirements, and for this we propose an unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA. SUMMARY

[0005] The purpose of the present application is to provide an unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA, which generates a unified target track with high-precision position, clear identity information, and continuous smooth trajectory by utilizing the continuous tracking capability of radar and the identity recognition characteristics of TDOA, thereby solving the problems raised in the above background technology.

[0006] To achieve the above purpose, the present application provides the following technical scheme: an unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA, comprising the following steps:

[0007] Receiving radar raw tracks and TDOA raw tracks and their covariance matrices with a fixed system processing period For time reference, time align the radar data and TDOA data, and output the time-aligned radar target list and TDOA target list ;

[0008] Based on the radar target list and TDOA target list, a hybrid association strategy combining statistical test and neural network is adopted to determine whether the radar track and TDOA track are from the same UAV, including coarse association, fine association and double threshold decision;

[0009] Based on the associated target pairs in the association matrix, optimal fusion and track splicing processing are performed;

[0010] In each processing cycle Output the fusion target list , wherein each target includes a unique identifier, a high-precision three-dimensional position, a position uncertainty estimate, a target type / identity, a tracking state and a data timestamp.

[0011] Further, with a fixed system processing cycle For time reference, time align the radar data and TDOA data, and the calculation process is as follows:

[0012] The radar original track is , and the TDOA original track is ;

[0013] For radar data, use the uniform speed model for extrapolation / interpolation, as shown below:

[0014]

[0015] wherein, represents the estimated radar target three-dimensional position at the unified reference time ; represents the original three-dimensional position coordinates of the target actually detected and reported by the radar at ; represents the target speed vector used for calculation; represents the time interval from the radar original observation time to the reference time ;

[0016] The same processing is performed on the position of the TDOA data, as shown below:

[0017]

[0018] wherein, represents the unified reference time the estimated TDOA target 3D position; the actual TDOA system position at the time of the original 3D position calculation; the target velocity vector used for TDOA data extrapolation; the time interval from the TDOA original positioning time to the reference time ;

[0019] the TDOA covariance matrix , if the time difference is less than a preset threshold , the original matrix is used, otherwise the inflated conservative covariance matrix is used.

[0020] Further, the TDOA covariance matrix is a positioning error covariance matrix estimated in real time based on the current geometry.

[0021] Further, the coarse association calculates, for each pair of potential associated targets , an innovation vector, an innovation covariance matrix and a Mahalanobis distance square for fast screening, as shown below:

[0022] Calculate the innovation vector:

[0023]

[0024] wherein, represents the position difference in three-dimensional space between the radar target and the TDOA target ;

[0025] Calculate the innovation covariance matrix:

[0026]

[0027] wherein, is the error covariance matrix of radar measurement, which is a diagonal matrix, representing the measurement error of the radar in each coordinate axis; is the error covariance matrix of radar measurement; measures the uncertainty of the innovation vector;

[0028] Calculate the normalized Mahalanobis distance square:

[0029]

[0030] used to measure the statistical significance of the difference after considering the uncertainty ;

[0031] if , then the target pair is added into the entering candidate set , where is the chi-square test threshold, for three-dimensional space, at 95% confidence level ≈7.815;

[0032] The fine association filters the candidate target pairs after the coarse association, constructs an enhanced feature vector, and inputs the pre-trained BP neural network to output the association confidence score, as shown below:

[0033] For each target pair in the candidate set , an enhanced feature vector is constructed.

[0034]

[0035] wherein represents a one-dimensional vector combining multiple types of information together as the input of the pre-trained BP neural network, for the network to comprehensively determine the association possibility; represents the three-dimensional coordinate components of the radar target ; represents the three-dimensional coordinate components of the TDOA target ; is the logarithm of the maximum eigenvalue of the innovation covariance matrix;

[0036] represents the maximum eigenvalue of the matrix, and the feature vector is input into the pre-trained BP neural network to obtain the association score , and ;

[0037] wherein the BP neural network is trained through a large amount of historical data, for learning the complex association mode beyond the Mahalanobis distance, and the BP neural network outputs an association confidence score between 0 and 1;

[0038] The double-threshold decision makes statistics on the number of times that the association score of the radar target and the TDOA target exceeds the first threshold in consecutive processing periods. ;

[0039] If , it is determined as the final association, and the association matrix is output, wherein represents that the radar target is successfully associated with the TDOA target .

[0040] Further, the innovation vector represents the spatial difference between the radar observation position and the TDOA observation position.

[0041] Further, the enhanced feature vector includes the log of the sum of the Mahalanobis distance and the maximum eigenvalue of the innovation covariance matrix.

[0042] Further, the optimal fusion employs a Kalman gain-based fusion algorithm to calculate the fusion gain matrix, the optimal fused position estimate, and the fused error covariance matrix. The track splicing process switches to radar-only tracking when the TDOA signal is lost and switches back to fusion tracking mode when the TDOA signal is restored. The target state in the stable fusion stage is output, including the position and covariance .

[0043] Further, the fusion gain matrix, the optimal fused position estimate, and the fused error covariance matrix are calculated as follows:

[0044] The fusion gain matrix is calculated as follows:

[0045]

[0046] wherein, represents the gain matrix; and represent the error covariance matrices of TDOA and radar, respectively;

[0047] The fusion gain matrix determines the weight distribution of the two observations on the final fusion result.

[0048] The optimal fused position estimate is calculated as follows:

[0049]

[0050] wherein, represents the optimal fused position estimate; represents the innovation vector;

[0051] The result of the radar position is weighted and corrected by the gain matrix .

[0052] If the error of the radar is much smaller than the error of the TDOA, the gain matrix will be small, and the final result will be more biased towards the radar, and vice versa.

[0053] The optimal fused position is calculated as follows:

[0054]

[0055] wherein, represents the optimal position estimation after fusion; represents the position vector of radar observation; represents the position vector of TDOA observation;

[0056] The position of TDOA is taken as the basis, and the gain matrix is used to weight and correct the difference between the radar position and the TDOA position, so as to obtain a more optimal position estimation;

[0057] The error covariance matrix after fusion is calculated as follows:

[0058]

[0059] wherein, represents the error covariance matrix corresponding to the optimal position estimation after fusion; represents a unit matrix; represents the remaining gain.

[0060] Further, the error covariance matrix after fusion is calculated in Joseph form, which is used to ensure the symmetry and positive definiteness of the covariance matrix.

[0061] Further, the target type / identity is derived from TDOA analysis.

[0062] Compared with the prior art, the beneficial effects of the present application are:

[0063] 1. The present application realizes significant performance improvement by designing a deep fusion architecture and intelligent association mechanism of radar and TDOA, which is not a simple data superposition, but makes full use of the stable, continuous and high refresh rate spatial coordinates provided by radar and the passive positioning information with identity provided by TDOA system, so that the system after fusion has the core advantages of both.

[0064] 2. The present application first performs fast screening based on Mahalanobis distance coarse association, and excludes a large number of unmatched target pairs, and then constructs an enhanced feature vector containing geometric configuration information, and uses neural network to learn complex and nonlinear association mode for fine association judgment. This double filtering mechanism greatly reduces the probability of false association and missed association, especially in a multi-target and dense environment, and provides a clean and reliable data basis for subsequent information fusion, which is a key prerequisite for realizing high-performance fusion tracking.

[0065] 3、The application adopts the optimal fusion algorithm based on Kalman gain and the adaptive covariance management strategy, realizes the essential optimization of positioning accuracy, in the information fusion stage, the algorithm is not a simple average of the radar and TDOA position, but according to the real-time, dynamic error covariance of the two, the optimal weight is calculated, so that the fusion result is the optimal estimation in the statistical sense, which not only significantly improves the positioning accuracy at a single moment, especially compensates the large error of TDOA in a certain direction, but also the output fusion covariance matrix more truly reflects the uncertainty of the current positioning result, thereby providing a crucial reliability measure for subsequent data processing and application decision BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 The flow chart of the unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA of the application. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0068] In order to solve the problem that the existing single detection means (such as radar or TDOA) is difficult to balance continuous stable tracking, high-precision positioning and target identity recognition in a complex urban environment, please refer to Figure 1 The embodiment provides the following technical solutions:

[0069] The unmanned aerial vehicle target intelligent fusion and tracking method based on radar and TDOA comprises the following steps:

[0070] Data preprocessing and time alignment step: receiving radar original track and TDOA original track and their covariance matrix , with a fixed system processing period as a time reference, time aligning the radar data and TDOA data, and outputting the radar target list and TDOA target list after time alignment ; ;

[0071] Among them, the covariance matrix of TDOA is the positioning error covariance matrix estimated in real time based on the current geometric configuration

[0072] Target association judgment step: based on the radar target list and TDOA target list, a hybrid association strategy combining statistical test and neural network is used to judge whether the radar track and TDOA track are from the same unmanned aerial vehicle, including coarse association, fine association and double threshold decision;

[0073] Information fusion and track management step: based on the associated target pairs in the association matrix, optimal fusion and track splicing processing are carried out;

[0074] Final system output step: in each processing period Output fusion target list , , wherein each target includes a unique identifier , a high-precision three-dimensional position , a position uncertainty estimate , a target type / identity , a tracking state and a data timestamp .

[0075] The technical effects of the above technical solutions are: the stable, high refresh rate three-dimensional track and direct speed measurement capability provided by the radar are used as the cornerstone of continuous tracking, and the passive positioning, identity recognition potential and superior low-altitude slow target detection capability of the TDOA system are introduced to form a powerful functional complement. Through rigorous data preprocessing and time alignment, the comparability and fusion basis of heterogeneous asynchronous data under the unified time reference are ensured; the hybrid association strategy combining statistical test and neural network has both computational efficiency and pattern learning ability, significantly improving the accuracy and robustness of correctly associating radar and TDOA tracks in dense target environment, overcoming the association challenges brought by track confusion and anisotropic positioning error; the optimal fusion algorithm based on Kalman gain and the flexible track management mechanism not only generate optimal position estimates superior to any independent source, but also seamlessly switch to radar tracking mode when TDOA signals are intermittently interrupted, and after signal recovery, re-associate and return to fusion tracking, thereby ensuring the continuity and stability of the track; finally, the system outputs a fusion target list integrating unique identity, high-precision position, uncertainty estimate, target type and tracking state, etc. Multi-dimensional information, providing users with a panoramic situation awareness far beyond a single sensor, complete and reliable information, greatly improving the detection, identification, tracking and decision-making efficiency of the anti-unmanned aerial vehicle system.

[0076] Taking a fixed system processing period as the time reference, the radar data and TDOA data are time-aligned, and the calculation process is as follows:

[0077] The radar raw track is , TDOA raw track is ;

[0078] For radar data, uniform velocity model is used for extrapolation / interpolation, as shown in the following formula:

[0079]

[0080] Wherein, represents the estimated radar target three-dimensional position at the unified reference time ; represents the actual radar target three-dimensional position coordinates detected and reported at ; represents the target velocity vector used for extrapolation; represents the time interval from the radar raw observation time to the reference time ;

[0081] The position of TDOA data is processed in the same way, as shown in the following formula:

[0082]

[0083] Wherein, represents the estimated TDOA target three-dimensional position at the unified reference time ; represents the actual TDOA system target three-dimensional position coordinates calculated at ; represents the target velocity vector used for TDOA data extrapolation; represents the time interval from the TDOA raw positioning time to the reference time ;

[0084] For the covariance matrix of TDOA , if the time difference is less than the preset threshold , the original matrix is used, otherwise the inflated conservative covariance matrix is used.

[0085] The technical effects of the above technical solutions are: the radar data and the TDOA position data are extrapolated or interpolated by using a uniform speed model conforming to the target motion law, instead of simple timestamp matching, so that the position calculation error caused by data transmission delay or processing period difference is significantly reduced, the accuracy of spatial alignment is ensured, and more importantly, the innovative processing of the TDOA covariance matrix can maintain the accuracy by using the original matrix when the time difference is small, and actively use the inflated conservative covariance matrix when the time difference exceeds the threshold, so that the positioning error based on the geometric configuration calculated at the old time can be reduced, the uncertainty is artificially increased to reflect the time-dependent decay, thereby avoiding the misassociation or fusion performance degradation caused by excessive trust in an expired precision index in the subsequent association and fusion links; the radar and TDOA target lists output to the subsequent process are not only synchronized in time, but also reasonably and consistently evaluated in terms of data reliability through the above-mentioned complete time alignment mechanism, thereby fundamentally improving the robustness and reliability of the entire fusion tracking system.

[0086] The coarse association calculates the innovation vector, innovation covariance matrix and Mahalanobis distance square for each pair of potential associated targets , and performs fast screening, as shown in the following:

[0087] The innovation vector is calculated as follows:

[0088]

[0089] wherein, represents the position difference between the radar target and the TDOA target in the three-dimensional space;

[0090] The innovation covariance matrix is calculated as follows:

[0091]

[0092] wherein, is the error covariance matrix of the radar measurement, which is a diagonal matrix, representing the measurement error of the radar in each coordinate axis; is the error covariance matrix of the radar measurement; measures the uncertainty of the innovation vector;

[0093] The normalized Mahalanobis distance square is calculated as follows:

[0094]

[0095] The innovation vector represents the spatial difference between the radar observation position and the TDOA observation position, and is used to measure the statistical significance of the difference after considering the uncertainty .

[0096] If , the target pair is added into the candidate set , where is the chi-square test threshold, for three-dimensional space (degree of freedom = 3), at 95% confidence level ≈7.815;

[0097] The candidate target pairs after coarse association screening are fine associated, an enhanced feature vector is constructed, and a pre-trained BP neural network is inputted to output an association confidence score, as shown below:

[0098] For each target pair in the candidate set , an enhanced feature vector is constructed.

[0099]

[0100] wherein represents a one-dimensional vector combining multiple information together as the input of the pre-trained BP neural network for the network to comprehensively judge the association possibility; represents the three-dimensional coordinate components of the radar target ; represents the three-dimensional coordinate components of the TDOA target ; is the logarithm of the maximum eigenvalue of the innovation covariance matrix;

[0101] represents the maximum eigenvalue of the matrix, and the feature vector is inputted into the pre-trained BP neural network to obtain the association degree score , and ;

[0102] wherein the BP neural network is trained by a large amount of historical data (including correct association and error association samples) to learn the complex association mode beyond the Mahalanobis distance (for example, in a specific area, a certain motion mode is more related to a specific ID of the unmanned aerial vehicle), and the BP neural network outputs an association confidence score between 0 and 1;

[0103] Double-threshold decision, in continuous processing periods, the number of times that the association score of the radar target and the TDOA target is greater than the first threshold is counted.

[0104] If , the final association is determined, and the association matrix is outputted, wherein,​ representing radar targets and TDOA targets correlation is successful.

[0105] The technical effects of the above technical solutions are: through the three-level progressive hybrid correlation criterion composed of coarse correlation, fine correlation and double-threshold decision, efficient, accurate and robust determination of the correlation relationship between radar and TDOA targets in a complex scene is realized; in the coarse correlation stage, innovation statistics and Mahalanobis distance are used for fast calculation and screening, a large number of irrelevant target pairs are effectively eliminated, the subsequent calculation complexity is greatly reduced, and the real-time performance of the system is ensured; in the fine correlation stage, an enhanced feature vector is constructed and a pre-trained BP neural network is introduced, which can deeply mine and learn the complex nonlinear correlation mode between data, and more finely identify the candidate target pairs screened in the coarse stage, effectively overcoming the limitations of relying solely on geometric distance statistics in the face of anisotropic positioning errors and dense target environments, and significantly improving the accuracy of the correlation determination; the final double-threshold decision mechanism examines the correlation confidence history in multiple consecutive processing periods, introduces a time dimension of continuous verification, effectively avoids false correlation and missed correlation caused by single-frame data fluctuations or transient interference, and greatly enhances the stability and reliability of the entire correlation determination process; the above design ensures that the final output correlation matrix can truly and stably reflect the corresponding relationship between radar targets and TDOA targets, thereby laying a solid and reliable foundation for subsequent information fusion.

[0106] The optimal fusion adopts a fusion algorithm based on Kalman gain to calculate a fusion gain matrix, an optimal position estimate after fusion, and an error covariance matrix after fusion. When the TDOA signal is lost, the track splicing processing switches to tracking relying only on the radar. When the TDOA signal is restored, the target correlation determination step is re-executed. After successful correlation, the fusion tracking mode is switched back, and the target state in the stable fusion stage is output, including the position and covariance ;

[0107] The fusion gain matrix, the optimal position estimate after fusion, and the error covariance matrix after fusion are calculated as follows:

[0108] The fusion gain matrix is calculated as follows:

[0109]

[0110] wherein, represents a gain matrix; and represent error covariance matrices of TDOA and radar, respectively;

[0111] Wherein, fusion gain matrix determines the weight distribution of two observations to the final fusion result, which is optimal under linear Gaussian assumption;

[0112] Calculate the optimal position estimation after fusion:

[0113]

[0114] Wherein, represents the optimal position estimation after fusion; represents the innovation vector;

[0115] The result of weighted correction of radar position is obtained by gain matrix The core role of gain matrix is to allocate weight, which determines how much to believe the observation of radar and how much to believe the observation of TDOA when fusing the final result;

[0116] If the error of radar is much smaller than the error of TDOA, gain matrix will be small, and the final result will be more biased towards radar, and vice versa;

[0117] Calculate the optimal position estimation after fusion:

[0118]

[0119] Wherein, represents the optimal position estimation after fusion; represents the position vector of radar observation; represents the position vector of TDOA observation;

[0120] The difference between radar position and TDOA position is weighted corrected by gain matrix to obtain a better position estimation, and this algorithm is optimal under linear Gaussian assumption;

[0121] Calculate the error covariance matrix after fusion:

[0122]

[0123] Wherein, represents the error covariance matrix corresponding to the optimal position estimation after fusion; represents the unit matrix; represents the remaining gain;

[0124] The error covariance matrix after fusion is calculated in Joseph form to ensure the symmetry and positive definiteness of the covariance matrix.

[0125] The technical effects of the above technical solutions are: through the optimal fusion algorithm based on Kalman gain, the observation data and uncertainty information (covariance matrix) of the two heterogeneous sensors of radar and TDOA can be fully utilized, the fusion gain matrix is calculated for weighted optimization, so that the optimal position estimation with significantly higher accuracy than any independent sensor is obtained, and a more accurate error covariance matrix is calculated at the same time, which quantitatively reflects the credibility of the fused position. At the same time, through the intelligent track splicing processing mechanism, the system has strong fault tolerance and self-adaptation capability. When the TDOA signal is lost due to shielding or interference, it can seamlessly degrade to pure radar tracking mode to ensure uninterrupted core track. Once the TDOA signal is restored, it can verify the target identity through re-association and quickly switch back to the high-precision fusion tracking mode. This process effectively overcomes the inherent shortcomings of intermittent TDOA data streams, ensuring that a continuous, stable and high-precision position and uncertainty evaluation are output throughout the tracking period, and finally achieving robust, continuous and reliable tracking of unmanned aerial vehicle targets in complex operational environments.

[0126] Working principle: First, the raw data of radar and TDOA are time-aligned, which can lay a unified time reference for subsequent processing. Then, an innovative hybrid association strategy is used for target association judgment. The hybrid association strategy combines the statistical test based on Mahalanobis distance for fast rough screening and the intelligent fine association based on BP neural network, and confirms the association relationship in continuous multiple cycles through a double-threshold decision mechanism, thereby significantly improving the accuracy and reliability of association judgment. After confirming the association, enter the information fusion and track management stage. The optimal fusion algorithm based on Kalman gain is used to synthesize the high refresh rate and stable three-dimensional point track of radar and the identity information and passive positioning advantages provided by TDOA to calculate the optimal position estimation and error covariance with higher accuracy than any single sensor. At the same time, this stage has intelligent track splicing capability, which can seamlessly switch between fusion tracking and pure radar tracking mode according to the availability of TDOA signal, ensuring the continuity and stability of the track. Finally, a fusion target list containing unique ID, high-precision three-dimensional position, uncertainty estimation, target identity and tracking state is periodically output. Based on the above steps, the inherent defects of radar lacking identity recognition ability and being susceptible to interference, and TDOA positioning error anisotropy and discontinuous data are overcome, forming a complementary solution, and finally achieving high-precision, continuous and stable tracking of unmanned aerial vehicle targets in complex urban environments.

[0127] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for intelligent fusion and tracking of UAV targets based on radar and TDOA, characterized in that, Includes the following steps: Received radar raw spot and TDOA raw spot and covariance matrix With a fixed system processing cycle Using this as a time reference, the radar data and TDOA data are time-aligned, and a time-aligned radar target list is output. and TDOA target list ; Based on the radar target list and the TDOA target list, a hybrid association strategy combining statistical testing and neural networks is adopted to determine whether radar points and TDOA points originate from the same UAV, including coarse association, fine association and dual threshold decision. Based on the associated target pairs in the association matrix, perform optimal fusion and track stitching processing; In each processing cycle Output fusion target list Each target includes a unique identifier, high-precision 3D location, location uncertainty estimate, target type / identity, tracking status, and data timestamp.

2. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA as described in claim 1, characterized in that, With a fixed system processing cycle Using this as a time reference, the radar data and TDOA data are time-aligned. The calculation process is as follows: The original radar dot pattern is The original TDOA point is ; For radar data, extrapolation / interpolation is performed using a uniform velocity model, as shown in the following equation: ； in, Indicates a unified reference time. Above, the estimated three-dimensional position of the radar target; Indicates that the radar is actually in Continuously detect and report the target's original three-dimensional position coordinates; This represents the target velocity vector used for calculation; Indicates the time from the initial radar observation. to the reference time The time interval; The location of TDOA data is processed in the same way, as shown in the following formula: ； in, Represents a unified reference time Above, the estimated three-dimensional position of the TDOA target; This indicates that the TDOA system is actually in The original three-dimensional position coordinates of the target obtained through calculation at any time; This represents the target velocity vector used for extrapolation from TDOA data; Indicates the time from the original TDOA positioning. to the reference time The time interval; For the covariance matrix of TDOA If time difference Less than the preset threshold If the result is positive, use the original matrix; otherwise, use the expanded conservative covariance matrix.

3. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 1, characterized in that, The covariance matrix of the TDOA is the positioning error covariance matrix estimated in real time based on the current geometric configuration.

4. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 1, characterized in that, The coarse association applies to each pair of potential association targets. The innovation vector, innovation covariance matrix, and squared Mahalanobis distance are calculated for fast filtering, as shown below: Calculate the innovation vector: ； in, Indicates radar target With TDOA goals Differences in position in three-dimensional space; Calculate the new information covariance matrix: ； in, It is the error covariance matrix of radar measurement, which is a diagonal matrix and represents the measurement error of the radar on each coordinate axis; It is the error covariance matrix of radar measurements; The uncertainty of the innovation vector was measured; Calculate the normalized squared Mahalanobis distance: ； Used to measure differences When considering uncertainty Statistical significance after; like Then the target is eligible to enter the candidate set. ,in, As the chi-square test threshold, for three-dimensional space, at a 95% confidence level... ≈7.815; The refined association constructs an enhanced feature vector for the candidate target pairs after the coarse association screening, and inputs it into a pre-trained BP neural network to output an association confidence score, as shown below: For candidate set For each pair of targets, construct an enhanced feature vector. ; ； in, This represents a one-dimensional vector that combines multiple pieces of information, which serves as the input to a pre-trained BP neural network, allowing the network to comprehensively determine the likelihood of association. Indicates radar target The three-dimensional coordinate components; Indicates TDOA objectives The three-dimensional coordinate components; The logarithm of the largest eigenvalue of the new information covariance matrix; The largest eigenvalue of the matrix is ​​represented by this eigenvector. This eigenvector is then input into a pre-trained backpropagation (BP) neural network to obtain the correlation score. ,and ; Among them, the BP neural network is trained with a large amount of historical data to learn complex association patterns that go beyond Mahalanobis distance, and the BP neural network outputs an association confidence score between 0 and 1. The dual-threshold decision, in continuous Within each processing cycle, statistical radar targets are counted. With TDOA goals correlation score Exceeding the first threshold Number of times ; like If so, it is determined to be the final association, and the association matrix is ​​output. ,in, Indicates radar target With TDOA goals Association successful.

5. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 4, characterized in that, The innovation vector represents the spatial difference between the radar observation position and the TDOA observation position.

6. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 1, characterized in that, The enhanced eigenvectors include the squared Mahalanobis distance and the logarithm of the largest eigenvalue of the new covariance matrix.

7. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 1, characterized in that, The optimal fusion employs a Kalman gain-based fusion algorithm to calculate the fusion gain matrix, the optimal position estimate after fusion, and the error covariance matrix after fusion. In the track stitching process, when the TDOA signal is lost, tracking is switched to radar-only tracking. When the TDOA signal is recovered, the target association judgment step is re-executed. Upon successful association, the system switches back to fusion tracking mode and outputs the target state during the stable fusion phase, including position. Covariance .

8. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 7, characterized in that, The fusion gain matrix, the optimal position estimate after fusion, and the error covariance matrix after fusion are calculated as follows: Calculate the fusion gain matrix: ； in, Represents the gain matrix; and Let represent the error covariance matrices of TDOA and radar, respectively; Among them, the fusion gain matrix determines the weight allocation of the two observations to the final fusion result; Calculate the optimal location estimate after fusion: ； in, This represents the optimal location estimate after fusion; Represents the innovation vector; Through the gain matrix The result of weighted correction of radar position; If the radar error is much smaller than the error gain matrix of TDOA It will be very small, and the final result will be more biased towards radar, and vice versa; Calculate the optimal position after fusion: ； in, This represents the optimal location estimate after fusion; Represents the position vector observed by the radar; Represents the position vector of the TDOA observation; Based on the position of TDOA, use the gain matrix By weighting and correcting the difference between the radar position and the TDOA position, a better position estimate is obtained; Calculate the fused error covariance matrix: ； in, This represents the error covariance matrix corresponding to the optimal position estimate after fusion; Represents the identity matrix; This indicates the remaining gain.

9. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 8, characterized in that, The fused error covariance matrix is ​​calculated using the Joseph form to ensure the symmetry and positive definiteness of the covariance matrix.

10. The method for intelligent fusion and tracking of UAV targets based on radar and TDOA according to claim 1, characterized in that, The target type / identity is derived from TDOA parsing.

Images