Weak computing condition track real-time cooperative fusion method

Abstract

This invention relates to the field of multi-target tracking and multi-sensor collaborative sensing technology, and discloses a real-time collaborative fusion method for tracks under weak computational conditions. The method includes: obtaining the track consistency cost of pairwise pairing of all tracks in the platform within an initialization time window, and constructing a cross-platform track-track cost matrix; performing global optimal matching on the cross-platform track-track cost matrix, and solidifying the correspondence between tracks across different platforms into a track mapping lookup table; for each local track at the current time step, determining whether there is a valid local measurement; if so, filtering and updating the local track using the local measurement; if not, determining the corresponding cross-platform track based on the track mapping lookup table and updating it; if both local measurement and collaborative information are unavailable, predicting and maintaining the local track based on the target motion model, thus establishing a track association and fusion mechanism with low communication overhead and low computational redundancy.

Description

Technical Field

[0001] This invention relates to the field of multi-target tracking and multi-sensor collaborative sensing technology, and in particular to a real-time collaborative fusion method for flight tracks under weak computational conditions. Background Technology

[0002] Multi-target tracking and trajectory estimation are key technologies in multi-sensor perception systems, widely used in scenarios such as traffic monitoring, airspace target surveillance, regional security, and multi-platform collaborative perception. In these applications, the system typically needs to perform continuous state estimation for multiple moving targets in complex environments while maintaining the consistency and stability of trajectory numbers.

[0003] In practical engineering conditions, due to limitations in platform size, power consumption, and computing resources, sensor platforms often struggle to continuously obtain complete and stable target observation information. Factors such as occlusion, interference, limited field of view, and changes in platform attitude can easily lead to intermittent data loss or quality degradation. When effective measurements are insufficient, existing methods typically rely on target motion models for prediction updates. However, without measurement constraints, state uncertainty accumulates, leading to decreased track accuracy and even track drift or interruption. Furthermore, traditional high-precision data association algorithms, such as Multi-Hypothesis Tracking (MHT) and Joint Probabilistic Data Interconnection (JPDA), usually require maintaining large hypothesis trees or calculating complex joint probabilities. As the number of targets and clutter density increases, the computational load grows exponentially, making it difficult to run in real-time on edge platforms with extremely limited onboard computing resources, such as missiles and UAVs. Simultaneously, centralized fusion architectures require transmitting all raw measurements to a central node, placing high demands on communication bandwidth and imposing excessive computational load on the central node. Failure of the central node can paralyze the entire system. Summary of the Invention

[0004] To address the aforementioned technical issues, this invention provides a real-time collaborative fusion method for flight tracks under weak computational conditions. It establishes a flight track association and fusion mechanism with low communication overhead and low computational redundancy to ensure real-time collaborative perception capabilities in complex dynamic environments, and has significant engineering application value.

[0005] This invention provides a real-time collaborative fusion method for flight tracks under weak computational conditions, comprising the following steps: Establish the target kinematic model and the measurement models of each platform. Each platform independently performs single-frame measurement based on global nearest neighbor and association matching and state update of local tracks to form a local track set. Within the preset initialization time window, obtain the local track sets formed by multiple platforms independently tracking, obtain the track consistency cost of pairing all tracks between any two platforms and within their respective local track sets within the initialization time window, and construct a cross-platform track-track cost matrix based on the track consistency cost. Perform global optimal matching on the cross-platform track-track cost matrix to determine the correspondence between tracks on different platforms, and solidify the correspondence into a track mapping lookup table; For each local track at the current time step, determine whether there is a valid local measurement; If it exists, the local track is filtered and updated using local measurements; if it does not exist, the cross-platform track corresponding to the local track is determined according to the track mapping lookup table, and after obtaining its status information as cooperative information, the local track is compensated and updated using the cooperative information; if neither local measurements nor cooperative information are available, the local track is predicted and maintained based on the target motion model, and the track status updated at the current time step is output.

[0006] Optionally, the track consistency cost between any two platforms within the platform, and between all tracks in their respective local track sets, within the initialization time window, is obtained, including the following steps: Pair the tracks in the local track sets of any two platforms across platforms to construct candidate track pairs; For each candidate track pair, obtain its state residuals and composite covariance at each time point within the initialization time window, and calculate the Mahalanobis distance between tracks at the corresponding time points. The Mahalanobis distance between tracks at each time point within the initialization time window is statistically processed to obtain the track consistency cost of the candidate track pair.

[0007] Optionally, the statistical processing involves calculating an average value, and the track consistency cost is obtained by averaging the Mahalanobis distances within the time window: , In the formula, For candidate track pairs and Cost of track consistency between different routes; The length of the window; For the track With track Mahalanobis distance between them.

[0008] Optionally, the correspondence can be solidified into a track mapping lookup table, including the following steps: Introducing binary assignment variables The assigned variables satisfy the feasibility condition of one-to-one pairing; Using binary assignment The reported track consistency costs are linearly weighted to construct a global cost function for cross-platform track matching. ; global cost function To optimize the objective, a global nearest neighbor algorithm is used to solve the cross-platform track-track cost matrix. Under the constraints, the objective function is obtained. The global optimal matching decision matrix that yields the minimum value ; Through continuous Extract the globally optimal matching decision matrix within each initialization time window. track index pairs with a median value of 1 The frequency of occurrence of each track index pair is counted, and index pairs that meet the stability threshold requirements are solidified into cross-platform track number mapping relationships and stored in the track mapping lookup table.

[0009] Optional, binary assignment variable Platform for quantitative description With the platform The track matching relationship between them, where: , To ensure that the track matching results satisfy the one-to-one constraint, a binary assignment variable is used. The following feasibility conditions must be met: , , In the formula, and They are respectively traversing the platform With the platform The index number of the track set; Candidate track pairs; For the platform Total number of tracks; For the platform Total number of flight tracks.

[0010] Optional, global cost function This can be expressed by the following formula: , In the formula, Cost matrix Located in the middle line, number The matrix elements of the column represent the physical meaning of the track consistency cost.

[0011] Optional, globally optimal matching decision matrix We obtain it from the following formula: , In the formula, Represents a set of binary assignment variables To optimize the object, while satisfying the one-to-one matching constraint, a search is performed to find the global cost function. The optimal matching combination that achieves the minimum value; The optimal matching decision matrix is ​​obtained by solving for the elements of a binary variable.

[0012] Optionally, index pairs that meet the stability threshold requirements can be fixed into cross-platform track number mapping relationships and stored in a track mapping lookup table, including the following steps: For any cross-platform track index pair Define its stability index for: , In the formula, To initialize the total number of time windows; where, the indicator variable Defined as: , When the stability index meets Then the track index is considered to be With track index The mapping relationship between the numbers is stable within a time window, and this mapping relationship is solidified into a cross-platform track number mapping relationship, with a stability threshold. The setting is based on the system's fault tolerance capability for erroneous associations; Output a fixed cross-platform track number mapping relationship, which serves as the basis for cross-platform collaboration during system operation, and enter the online collaboration phase.

[0013] Optionally, the local track can be updated using collaborative information, including the following steps: If no valid local measurement exists at the current moment, the state estimate of the corresponding cooperative platform's matching target at the current moment is obtained based on the track mapping lookup table. and the corresponding covariance matrix And project the collaborative target state onto this platform. In the local coordinate system, the transformed state estimate is obtained. and covariance matrix Construct this platform Collaborative pseudo-measurement and its measurement error covariance : , , Introducing a collaborative measurement mapping matrix This is used to map the target state to the collaborative measurement space; Constructing the State – Collaborative Measurement Cross-Covariance Matrix and covariance of measurement residuals : , , In the formula, To predict the covariance matrix; The collaborative filter gain is derived. : ; Perform cross-platform collaborative track state and its state covariance matrix update: , , In the formula, for The target track is updated in real time. The state; for Mean of time-state prediction; for Track updated in real time The state covariance matrix.

[0014] Optionally, after outputting the track status updated at the current time step, the output may also include: During the subsequent online operation phase, each platform directly retrieves the track number mapping pairs between heterogeneous platforms in real time by searching the track mapping lookup table, and skips the cross-platform global matching and association calculation. Each platform only needs to independently perform a single-frame association determination between local measurements and local tracks at each time step: if a valid local measurement is determined, then the local measurement is used to perform filtering updates; if a valid local measurement is determined, then track updates based on cooperative information or prediction maintenance based on the target kinematics model are performed according to the cooperative platform information corresponding to the track mapping lookup table index.

[0015] The technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art: This invention provides a real-time collaborative fusion method for tracks under weak computational conditions. Each platform independently executes single-frame measurement and local track association matching based on global nearest neighbors. Cross-platform association is completed only during the initialization phase through statistical matching over multiple consecutive time windows, and the stable correspondence is solidified into a track mapping lookup table. Subsequent time steps do not repeat complex association calculations. This avoids the exponential computational growth of traditional high-precision algorithms, can be executed in real-time on embedded platforms, and simplifies cross-platform track association to statistical solidification during the initialization phase and extremely low-overhead reading during the online phase. It avoids multi-platform combined optimization for each frame, significantly reducing the online computational pressure during multi-sensor collaboration. This method replaces online joint probability calculation with table lookup, allowing cross-platform collaborative updates to directly read the corresponding track status without repeating complex track-track association calculations.

[0016] If no local measurement is available at the current time step, the corresponding cross-platform track is determined first using the track mapping lookup table, and its state information is obtained as cooperative information for compensation updates. Only when both local measurement and cooperative information are unavailable does it degenerate into prediction maintenance based on the motion model. This utilizes cross-platform observation complementarity to compensate for intermittent measurement gaps on a single platform. Even if the platform cannot detect the target due to occlusion, limited field of view, etc., it can still use the stable track of the target from other platforms for filtering and updating through the fixed mapping relationship, effectively suppressing the accumulation of state uncertainty. This avoids the accuracy degradation and track interruption caused by relying solely on the motion model. Traditional methods rely solely on model prediction when measurements are missing, leading to rapid error divergence. This invention maintains track continuity and accuracy and extends track lifetime through cooperative information injection. Simultaneously, it considers system robustness; cooperative information and local measurement form redundancy, ensuring that temporary failure of any platform will not cause global track breakage.

[0017] Each platform independently performs local tracking, forming a local track set. Local tracks are exchanged only during the initialization phase to construct the cost matrix; subsequently, only the status information of the corresponding tracks (not the original measurements) is transmitted, and collaborative information is only acquired on demand when local measurements are missing. This significantly reduces communication bandwidth requirements, eliminating the need to continuously transmit original measurements to a central node; only the associated track status is transmitted, resulting in a much smaller data volume than the original measurements. Decentralization enhances system resilience; there is no single central node, and each platform operates independently yet supports each other through mapping tables. The failure of any node does not affect the track maintenance and collaborative updates of the remaining platforms. It meets the resource constraints of distributed platforms (missiles, UAVs), prioritizing local processing with collaborative processing as a secondary measure, and conforms to practical engineering conditions with limited size, power consumption, and computing resources.

[0018] By performing globally optimal matching based on track consistency costs within the initialization time window and solidifying the matching results into a track mapping lookup table, this mapping relationship remains unchanged in subsequent time steps. This avoids the problem of inconsistent track numbers for the same target across different platforms due to independent tracking by each platform. After solidifying the mapping table, even if subsequent measurements are intermittent, the track reference relationship for the same target remains stable across platforms, providing a reliable association basis for collaborative fusion and situational awareness unification, which meets the core requirement of track identity consistency in multi-target tracking. Attached Figure Description

[0019] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a real-time collaborative fusion method for weak computational conditions of flight tracks provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of multi-sensor collaborative detection of a three-dimensional scene in an embodiment of the present invention; Figure 3 This is a timing diagram of the target azimuth angle measurement received by sensor 1, including target echo and clutter; Figure 4 The target azimuth measurement timing diagram received by sensor 2 includes target echo and clutter; Figure 5 This is a timing diagram of the target elevation angle measurement received by sensor 1, including target echo and clutter. Figure 6 The timing diagram of the target pitch angle measurement received by sensor 2 includes target echo and clutter. Figure 7 The cross-platform track association and matching conclusion diagram includes track numbers between two sensor platforms; Figure 8 This is a target tracking effect diagram for Sensor 1, including the actual trajectories and estimated tracks of three targets; Figure 9 This is a target tracking effect diagram for Sensor 2, including the actual trajectories and estimated tracks of three targets; Figure 10 This is a comparison chart of the real-time performance and computational complexity of algorithms on a high-performance computing platform, including a comparison of the single-step time of traditional centralized fusion and the distributed collaborative method of this invention; Figure 11 This is a verification diagram of weak computing adaptability based on RK3588 embedded edge nodes, which includes the system's hard real-time benchmark and the time consumption and computing power margin of this invention under weak computing conversion conditions. Detailed Implementation

[0021] How to avoid the computational overload caused by highly complex data association algorithms on edge platforms with limited computing resources, and how to achieve accurate compensation and continuous estimation of track status using cross-platform collaborative information when observation information is intermittently missing, is a pressing technical problem in the field of multi-target tracking. Especially in multi-platform collaborative scenarios, establishing a track association and fusion mechanism with low communication overhead and low computational redundancy to ensure real-time collaborative perception capabilities in complex dynamic environments has significant engineering application value.

[0022] To address the aforementioned problems, this invention aims to provide a real-time collaborative fusion method for tracks under weak computational constraints. This method establishes target kinematic and measurement models within each sensor platform, completes the association processing and state update of single-frame measurements and tracks, and performs consistency measurement on track states across different platforms within a short time window. It also constructs a cross-platform track similarity cost and establishes a stable track mapping lookup table. During online operation, based on the availability of local measurements and cross-platform collaborative information at the current time step, it adaptively switches between conventional filtering updates, cross-platform collaborative compensation updates, and prediction maintenance, thereby maintaining the continuity and stability of track states under incomplete observation conditions. By introducing a one-time cross-platform association based on track-level matching and a condition-triggered collaborative update mechanism, this invention ultimately achieves continuous, stable, and robust collaborative estimation of multi-target tracks under conditions of limited observation and computational resources.

[0023] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0024] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] First, this invention provides a real-time collaborative fusion method for flight paths under weak computational conditions, specifically as follows: Figure 1 As shown, it includes the following steps: Establish the target kinematic model and the measurement models of each platform. Each platform independently performs single-frame measurement based on global nearest neighbor and association matching and state update of local tracks to form a local track set. Within the preset initialization time window, obtain the local track sets formed by multiple platforms independently tracking, obtain the track consistency cost of pairing all tracks between any two platforms and within their respective local track sets within the initialization time window, and construct a cross-platform track-track cost matrix based on the track consistency cost. Perform global optimal matching on the cross-platform track-track cost matrix to determine the correspondence between tracks on different platforms, and solidify the correspondence into a track mapping lookup table; For each local track at the current time step, determine whether there is a valid local measurement; If it exists, the local track is filtered and updated using local measurements; if it does not exist, the cross-platform track corresponding to the local track is determined according to the track mapping lookup table, and after obtaining its status information as cooperative information, the local track is compensated and updated using the cooperative information; if neither local measurements nor cooperative information are available, the local track is predicted and maintained based on the target motion model, and the track status updated at the current time step is output.

[0026] Specifically, the present invention provides a real-time collaborative fusion method for tracks under weak computational conditions, including S101: establishing a target kinematics and measurement model and completing track initialization; Includes the following steps: 11) Based on the target's motion model, establish the state transition equations, assuming the state transition equations are as follows: , In the formula, for The target state at any given time. This is the state transition function. for The target state at any given time. For process noise, With a mean of 0 and a variance of The Gaussian distribution.

[0027] 12) Based on the characteristics of the measurement model and the physical characteristics of the target, establish the angle measurement equation; Assuming the relative state vectors of the target and the sensor The measurements received by the infrared sensor consist of the target's azimuth and elevation angles, and the specific form of its measurement equation is as follows: , In the formula, For measurement data, To measure the transfer function, It is the azimuth angle. The pitch angle, To measure noise, it is predetermined through sensor factory parameters, offline experimental calibration, or empirical models; With a mean of 0 and a variance of The Gaussian distribution.

[0028] 13) Accept the initial measurement, set the target position and zero velocity, construct the covariance matrix based on prior information, and complete the track initialization; S102: Each platform independently performs single-frame measurement and local track association matching based on global nearest neighbor; Includes the following steps: 21) Take the track state at the previous moment and perform a one-step prediction of the state and its covariance matrix; One of the prediction steps is as follows: 211) According to Time of the first The estimated mean of the flight path ,generate One volume point: , In the formula, For the first The flight path is Time of the first The specific value of each volume point Estimate its mean. The standard volume point is obtained by referring to a table; Let be the state dimension. For the first The flight path is The square root of the covariance matrix at time step is calculated as follows: , In the formula, For the first The flight path is The time-time covariance matrix is ​​initialized with a value and updated at each subsequent time step. 212) According to the state transition function For the The volume points of the flight path undergo a nonlinear transformation: , In the formula, For the first The track is in The values ​​of each volume point after nonlinear transformation by the state transition function.

[0029] 213) Calculate the first mean of state prediction for each track With predicted covariance : The following calculation was performed: , In the formula, For the first The mean of the state prediction for each track.

[0030] The following calculation was performed: , In the formula, Its prediction error covariance, Let be the process noise covariance.

[0031] 22) Calculate predictive measurements and statistics; 221) The measurement volume point is obtained by nonlinear transformation of the state volume point using the measurement equation: , In the formula, For the first The track is in Measurement space sample values ​​at each volume point For the state space volume point, This is the angle measurement function.

[0032] 222) Calculate the mean of the predicted measurements. Covariance of Measurement Residuals : The following calculation was performed: , In the formula, For the first The flight path is The mean of the predicted measurements at any given time; The following calculation was performed: , In the formula, For the corresponding measurement residual covariance, Let be the process noise covariance.

[0033] 23) Candidate correlation pairs are screened based on elliptical tracking gates, and a cost matrix is ​​constructed; 231) Select the threshold value of the tracking gate according to the optimality criterion. This ensures that the echo located within the tracking gate has the highest probability of originating from the target rather than other noise, thus constructing the optimal tracking gate threshold value as follows: , In the formula, The probability of detection is obtained from sensor hardware specifications or statistical experiments; For the new echo density (target echo and redundant echo), For the observation dimension, for The measurement prediction covariance at time.

[0034] 232) Calculate the residuals Distance from Mahalanobis : The following calculation was performed: , In the formula, for The residual between the measurement data at time point and the mean of the measurement prediction. for The first moment Individual measurement, For the first The flight path is The mean of the predicted measurements at any given time; The following calculation was performed: , In the formula, for Time of the first track and the first Mahalanobis distance between measurements.

[0035] 233) Criteria for determining whether a measurement falls within the tracking gate: , In the formula, The threshold value of the selected elliptical tracking gate is dynamically adjusted according to the clutter density of the current environment. In areas with dense clutter, the threshold value is appropriately reduced to suppress false tracking, while in areas with greater target maneuverability, the threshold value is appropriately increased to prevent falling behind.

[0036] 234) Forming a candidate set: , In the formula, for The moment fell into the first Measurement set within the track tracking gate.

[0037] 235) Construct the measurement-track cost matrix: , In the formula, The cost matrix is ​​constructed. for Time of the first The measurement residual covariance matrix of the flight path. It is the natural logarithm function.

[0038] 236) The optimal measurement is selected using global nearest neighbors, and the objective function of the planning is determined: , In the formula, for The number of measurements at any given time for Number of target tracks at any given time For the first The first measurement and the first The cost of association between flight paths 0–1 decision variables, representing measurement Whether assigned to a track .

[0039] 237) Global nearest neighbor constraint: A track is associated with at most one measurement, and a measurement is associated with at most one track. The mathematical representation of this constraint is: 238) Solve the objective function to obtain the optimal solution: The optimal solution is in the form of ,in .

[0040] 0) Based on the optimal association result For each target track Determine its current time. Is there a measurement that is uniquely associated with it? If it exists Then it is considered that the measurement This is a valid update measurement for the trajectory.

[0041] This invention provides a real-time collaborative fusion method for tracks under weak computational conditions. Each platform independently performs single-frame measurement and local track association matching based on global nearest neighbor. Cross-platform association is completed only during the initialization phase through statistical matching over multiple consecutive time windows, and the stable correspondence is solidified into a track mapping lookup table. Subsequent time steps do not repeat complex association calculations. This avoids the exponential computational growth of traditional high-precision algorithms, can be executed in real-time on embedded platforms, and simplifies cross-platform track association to statistical solidification during the initialization phase and extremely low-overhead reading during the online phase. It avoids multi-platform combination optimization per frame, significantly reducing the online computational pressure during multi-sensor collaboration. This method replaces online joint probability calculation with table lookup, allowing cross-platform collaborative updates to directly read the corresponding track state without repeating complex track-track association calculations. If there is no local measurement at the current time step, the corresponding cross-platform track is determined first using the track mapping lookup table, and its state information is obtained as collaborative information for compensation updates. Only when both local measurements and collaborative information are unavailable does it degenerate into prediction maintenance based on motion models. This invention leverages cross-platform observation complementarity to compensate for intermittent measurement gaps on a single platform. Even if a platform is unable to detect a target due to obstruction or limited field of view, it can still use the stable trajectory of the target from other platforms for filtering and updating through a fixed mapping relationship, effectively suppressing the accumulation of state uncertainty. It avoids the accuracy degradation and trajectory interruption caused by relying solely on motion models. Traditional methods rely solely on model prediction when measurements are missing, leading to rapid error divergence. This invention maintains trajectory continuity and accuracy and extends trajectory lifetime through collaborative information injection. Simultaneously, it considers system robustness; collaborative information and local measurements form redundancy, ensuring that temporary failure of any platform will not cause global trajectory disruption.

[0042] Each platform independently performs local tracking, forming a local track set. Local tracks are exchanged only during the initialization phase to construct the cost matrix; subsequently, only the status information of the corresponding tracks (not the original measurements) is transmitted, and collaborative information is only acquired on demand when local measurements are missing. This significantly reduces communication bandwidth requirements, eliminating the need to continuously transmit original measurements to a central node; only the associated track status is transmitted, resulting in a much smaller data volume than the original measurements. Decentralization enhances system resilience; there is no single central node, and each platform operates independently while supporting each other through a mapping table. The failure of any node does not affect the track maintenance and collaborative updates of the remaining platforms. It meets the resource constraints of distributed platforms (missiles, UAVs), prioritizing local processing with supplementary collaboration, conforming to practical engineering conditions limited by size, power consumption, and computing resources. Global optimal matching based on track consistency cost is performed within the initialization time window, and the matching results are solidified into a track mapping lookup table, maintaining this mapping relationship in subsequent time steps. This avoids the problem of inconsistent track numbers for the same target across different platforms due to independent tracking by each platform. After the mapping table is solidified, even if subsequent measurements are intermittent, the track reference relationship of each platform for the same target remains stable, providing a reliable association basis for collaborative integration and situational awareness, which meets the core requirement of track identity consistency in multi-target tracking.

[0043] Cross-platform matching is the most computationally and bandwidth-intensive part. On weak computing platforms (such as RK3588), if global matching is performed for every frame, the system will freeze as the number of targets increases.

[0044] Based on the above problems, in this embodiment of the invention, the track consistency cost of pairing all tracks within the local track set of any two platforms within the platform during the initialization time window is obtained, including the following steps: Pair the tracks in the local track sets of any two platforms across platforms to construct candidate track pairs; For each candidate track pair, obtain its state residuals and composite covariance at each time point within the initialization time window, and calculate the Mahalanobis distance between tracks at the corresponding time points. The Mahalanobis distance between tracks at each time point within the initialization time window is statistically processed to obtain the track consistency cost of the candidate track pair.

[0045] Traditional correlation methods only consider distance within a single frame. However, in cross-platform collaboration, the bias and noise between different sensors are significant, making single-frame correlation prone to errors.

[0046] Based on the above problems, in this embodiment of the invention, the statistical processing involves calculating an average value, and the track consistency cost is obtained by averaging the Mahalanobis distance within the time window: , In the formula, For candidate track pairs and Cost of track consistency between different routes; The length of the window; For the track With track Mahalanobis distance between them.

[0047] By performing temporal accumulation within the initialization time window, the instantaneous noise impact of single-frame measurements is effectively smoothed out, providing a solid confidence foundation for the solidification of the lookup table. Only cross-platform track pairs that consistently exhibit high consistency within this continuous time window are determined to point to the same real target.

[0048] Specifically, the present invention provides a real-time collaborative fusion method for weak computational conditions of flight tracks, including: S103: accumulating and calculating the state consistency cost between cross-platform flight track sequences within a set initialization time window; In this embodiment, the example is illustrated by two sensor platforms, each of which is simultaneously tracking multiple target tracks, but the present invention is not limited thereto.

[0049] Includes the following steps: 31) Select the initialization time window and construct a cross-platform track sequence. 311) Select a continuous initialization time window, the initialization time window containing continuous... Each time step is denoted by the following formula: , In the formula, The window length is preset based on system stability and computing resource conditions.

[0050] 312) Within this time window, respectively from the platform With the platform Extract the target tracks currently in the root state and form track sets accordingly: , In the formula, , Platforms With the platform The total number of tracks is calculated, and each track consists of filtered state estimates and corresponding covariance matrices from multiple consecutive moments within a time window. That is, for the platform... The track Its trajectory sequence can be represented as: , In the formula, for Moment Platform No. Filtered state estimation of a single track. for Moment Platform No. The corresponding covariance matrix of the flight path, platform The way the track sequence is defined is the same.

[0051] 32) Calculate the cross-platform track state residuals and covariance; It should be noted that before calculating the cross-platform track state residuals, the track states of the cooperating platforms must be uniformly transformed to the coordinate system of one of the platforms. For the platform... flight path With the platform flight path At any given moment ,platform The local state is estimated as The corresponding state estimation covariance matrix is ;platform The trajectory status is converted to the platform's relative pose between platforms. The state estimate after the local coordinate system is denoted as The transformed state estimation covariance matrix is ​​denoted as... Construct cross-platform track state residuals:

[0052] , In the formula, for Moment Platform track With the platform track The residuals. Assuming that the trajectory estimation errors of different platforms are independent, the composite covariance matrix corresponding to the state residuals is obtained:

[0053] , In the formula, State residual The corresponding composite covariance matrix, for Moment Platform No. The covariance matrix of the flight path, for Moment Platform No. The covariance matrix of the track after transformation.

[0054] 33) Calculate the track-level Mahalanobis distance time-by-time; In the time window Calculate the distance between cross-platform track pairs time-by-time, at any given time. ,track With track The Mahalanobis distance between tracks is defined as: , In the formula, For the track With track Mahalanobis distance between them.

[0055] 34) Calculate the track consistency cost within the initialization time window; For time windows The Mahalanobis distances calculated time-by-time are statistically processed to form the overall consistency cost at the track-to-track level. The average Mahalanobis distance within the time window is used as the track consistency cost. , In the formula, For the track With track The cost of track consistency between platforms can be reduced by statistically aggregating the track state residuals at multiple time points within a time window, which can effectively suppress the impact of single-frame measurement noise on cross-platform track matching results.

[0056] 35) Construct a cross-platform track-track cost matrix; For the platform With the platform Construct a cross-platform track-track cost matrix from the set of tracks: , In the formula, For the platform With the platform The set of tracks is used to construct a cross-platform track-track cost matrix, and the matrix elements are... For the platform The Tracks and Platforms The The consistency cost of a track within the initialization time window.

[0057] Optionally, the correspondence can be solidified into a track mapping lookup table, including the following steps: Introducing binary assignment variables The assigned variables must satisfy the feasibility condition of one-to-one pairing; Using binary assignment Cost of consistent flight path A global cost function for cross-platform trajectory matching is constructed by performing linear weighting. ; global cost function To optimize the objective, a global nearest neighbor algorithm is used to optimize the cross-platform track-track cost matrix. Solving for the objective function under the given constraints, we obtain the result. The global optimal matching decision matrix that yields the minimum value ; Through continuous Within an initialization time window, extract the globally optimal matching decision matrix. Track index pairs with a value of 1 The frequency of occurrence of each track index pair is counted, and index pairs that meet the stability threshold requirements are solidified into cross-platform track number mapping relationships and stored in the track mapping lookup table.

[0058] Optional, binary assignment variable Platform for quantitative description With the platform The track matching relationship between them, where: , To ensure that the track matching results satisfy the one-to-one constraint, a binary assignment variable is used. The following feasibility conditions must be met: , , In the formula, and They are respectively traversing the platform With the platform The index number of the track set; Candidate track pairs; For the platform Total number of tracks; For the platform Total number of flight tracks.

[0059] Optional, global cost function This can be expressed by the following formula: , In the formula, Cost matrix Located in the middle line, number The matrix elements of the column represent the physical meaning of the track consistency cost.

[0060] Optional, globally optimal matching decision matrix We obtain it from the following formula: , In the formula, Represents a set of binary assignment variables To optimize the object, while satisfying the one-to-one matching constraint, a search is performed to find the global cost function. The optimal matching combination that achieves the minimum value; The optimal matching decision matrix is ​​obtained by solving for the elements of a binary variable.

[0061] Optionally, index pairs that meet the stability threshold requirements can be fixed into cross-platform track number mapping relationships and stored in a track mapping lookup table, including the following steps: For any cross-platform track index pair Define its stability index for: , In the formula, To initialize the total number of time windows; where, the indicator variable Defined as: , When the stability index meets Then the trajectory is considered With track The mapping relationship between the numbers is stable within a time window, and this mapping relationship is solidified into a cross-platform track number mapping relationship, with a stability threshold. The setting is based on the system's fault tolerance capability for erroneous associations; Output a fixed cross-platform track number mapping relationship, which serves as the basis for cross-platform collaboration during system operation, and enter the online collaboration phase.

[0062] Specifically, the present invention provides a real-time collaborative fusion method for weak computational conditions of flight tracks, including: S104: completing global matching of cross-platform flight tracks based on multi-period statistical stability, and then constructing and solidifying a flight track mapping lookup table, including the following steps: 41) Introduce track matching assignment variables; Introducing binary assignment variables Platform for quantitative description With the platform The track matching relationship between them, where: .

[0063] To ensure that the track matching results satisfy the one-to-one constraint, a binary assignment variable is used. The following feasibility conditions must be met: , , In the formula, and They are respectively traversing the platform With the platform The index number of the track set; Candidate track pairs; For the platform Total number of tracks; For the platform Total number of flight tracks.

[0064] 42) Construct a global optimization objective function for cross-platform trajectory matching; Using binary assignment The cross-platform trajectory association cost matrix elements in Perform linear weighting to construct a global cost function : , In the formula, The cost matrix Located in the middle line, number The matrix elements of the column have the physical meaning of the aforementioned track consistency cost.

[0065] 43) The global nearest neighbor algorithm is used to solve for optimal track matching; The global nearest neighbor algorithm is used to solve the problem, and the globally optimal matching result that satisfies the constraints is obtained: , In the formula, Represents a set of binary assignment variables To optimize the object, while satisfying the one-to-one matching constraint, a search is performed to find the global cost function. The optimal matching combination that achieves the minimum value; The optimal matching decision matrix is ​​obtained by solving for the elements of a binary variable.

[0066] 44) Generate cross-platform track number mapping relationships; Establish a platform With the platform The mapping relationship between track numbers is defined, and the track mapping set is defined as follows: , In the formula, This is a set of track number mappings, where each element represents a platform. track index With the platform track index They were determined to originate from the same target.

[0067] 45) Determine and solidify the stability of the track number mapping relationship. In multi-platform collaboration, a single match (i.e., a single time window) may be affected by noise or outliers, leading to erroneous associations. Directly fixing the mapping relationship poses a significant risk.

[0068] To address the aforementioned issues, in this embodiment of the invention, cross-platform track matching is repeatedly performed within multiple adjacent time windows to statistically analyze the consistency of track number mapping relationships across different time windows. Let's assume that in the most recent consecutive... The track matching results obtained within each time window are as follows: For any cross-platform track pair Its stability index is defined as:

[0069] , In the formula, For stability indicators; To initialize the total number of time windows; indicator variable Defined as: , When the stability index meets Then the track index is considered to be With track index The mapping relationship between the numbers is stable within a time window, and this mapping relationship is solidified into a fixed track number correspondence between platforms, with a stability threshold. The setting is based on the system's fault tolerance capability for erroneous associations.

[0070] By smoothing and statistically analyzing the matching results over M consecutive time windows, the recurrence rate of the mapping relationship is calculated. This transforms an instantaneous matching problem into a statistical confidence problem. The mapping is only solidified when its stability exceeds a threshold. This ensures the accuracy of lookup-based collaboration, avoids the transmission of erroneous information between platforms, and is a key threshold for achieving low computational complexity and high reliability.

[0071] 46) Output track number mapping and enter the online collaboration phase; Once the track number mapping relationship is fixed, it will be used as the basis for cross-platform collaboration during system operation, and the cross-platform track matching operation will not be repeated at every time step.

[0072] Distributed fusion places extremely high demands on clock synchronization and data alignment. In environments with weak computing and communication, frequent fusion can lead to computational redundancy and track jitter.

[0073] To address the aforementioned issues, this embodiment of the invention utilizes cooperative information to compensate and update the local track, including the following steps: If no valid local measurement exists at the current moment, the state estimate of the corresponding cooperative platform matching target at the current moment is obtained according to the track mapping lookup table. and the corresponding covariance matrix And project the collaborative target state onto this platform. In the local coordinate system, the transformed state estimate is obtained. and covariance matrix Construct this platform Collaborative pseudo-measurement and its measurement error covariance : , , Introducing a collaborative measurement mapping matrix This is used to map the target state to the collaborative measurement space; Constructing the State – Collaborative Measurement Cross-Covariance Matrix and covariance of measurement residuals : , , In the formula, To predict the covariance matrix; The collaborative filter gain is derived. : .

[0074] Perform cross-platform collaborative track state and its state covariance matrix update: , .

[0075] In the formula, for The target track is updated in real time. The state; for Mean of time-state prediction; for Track updated in real time The state covariance matrix.

[0076] This method constructs a gradient backoff logic of local priority -> table lookup collaboration -> model prediction. This reduces interference by not using collaboration when local data is available; when the target is lost locally (e.g., due to occlusion), table lookup is used immediately for instant compensation. The lightweight architecture eliminates the need for complex weighted fusion calculations, directly introducing collaborative trajectory state, making it ideal for edge deployment.

[0077] Specifically, the present invention provides a real-time collaborative fusion method for weak computational conditions of flight tracks, including S105: online determination of the validity of local measurements for each flight track and the availability of collaborative information based on fixed mapping; 51) The validity of local measurements is determined jointly by the elliptic gate and the global nearest neighbor assignment: , And this measurement-track pair is selected in the optimal assignment matrix: , 52) If there exists a measurement that satisfies the above conditions Then determine the track. At any moment If a valid local measurement exists, it is considered that the local measurement is missing.

[0078] 53) If cooperative information matching the target is found based on the fixed track mapping table, then the track is determined. At any moment The conditions for collaborative compensation are met; otherwise, the collaborative information is deemed unavailable.

[0079] S106: When local measurements exist, prioritize using local data to perform regular filtering updates, including the following steps: 61) When there is a valid measurement that satisfies the elliptical tracking gate constraint and is selected by the platform association at the current time step, the corresponding target track adopts the conventional state update strategy.

[0080] 62) Calculate the state – measurement cross-covariance and filter gain: For tracks with valid correlation measurements Based on the principle of capacitive Kalman filtering, using the state volume points and measurement volume points obtained in step 22), the cross-covariance matrix between the state and the measurement is calculated: , And calculate the corresponding filter gain matrix: , In the formula, For the first The flight path is State at any given time—measurement cross-covariance The mean of the state prediction. For the first The flight path is Filter gain at time t, For the first The measurement residual covariance matrix corresponding to each flight track.

[0081] 63) Perform track status update The target track status is updated using filter gain and correlated measurements: , Simultaneously update the state covariance matrix of the trajectory: , In the formula, for The target track is updated in real time. state, for Mean of state prediction at time point for The filtering gain matrix of the target trajectory at any given time. for Track updated in real time The state covariance matrix.

[0082] S107: When local measurements are missing, compensation and updates are performed directly based on the established mapping relationship by introducing the cooperative track status, including the following steps: 71) When there is no valid measurement that satisfies the elliptical tracking gate constraint and is not selected by the platform association at the current time step, a cross-platform collaborative update strategy is adopted according to the inter-platform track number mapping relationship that has been fixed in step S104.

[0083] 72) Based on the paired track pairs, for this platform No. The target trajectory, assuming it is on the pairing platform The corresponding track number on is The state estimate of the trajectory at the current moment is obtained from the collaborative platform. and covariance matrix And project the collaborative target state onto this platform. In the local coordinate system, the transformed state estimate is obtained. and covariance matrix Construct this platform Target track Collaborative pseudo-measurements and their measurement error covariance: , , In the formula, For this platform Target track Collaborative pseudo-measurement; Let it be the measurement error covariance.

[0084] 73) Introducing a collaborative measurement mapping matrix This is used to map the target state to the cooperative measurement space. Since the cooperative pseudo-measurement in this embodiment has been unified to the local state space of this platform, i.e., a full-state linear mapping, therefore... Take as the identity matrix .

[0085] 74) Based on the aforementioned collaborative measurement mapping matrix Performing linear collaborative computation of state-collaborative measurement cross-covariance and filter gain includes the following steps: 741) Based on the predicted covariance obtained in step 21) The calculated state-coordinated measurement cross-covariance matrix is ​​as follows: , In the formula, The state-coordinated measurement cross-covariance matrix.

[0086] 742) The covariance matrix of the collaborative measurement residuals is calculated as follows: , In the formula, The state-coordinated measurement cross-covariance matrix.

[0087] 743) The collaborative filter gain is calculated as follows: , In the formula, This represents the collaborative filtering gain.

[0088] 75) Perform cross-platform collaborative state updates; The target trajectory status of this platform is updated using the calculated cooperative filtering gain and cooperative pseudo-measurement. , Simultaneously update the state covariance matrix of the target trajectory: , Through the aforementioned cross-platform collaborative updates, the system can directly introduce cross-platform collaborative information for robust compensation with extremely low computing power overhead.

[0089] Optionally, after outputting the track status updated at the current time step, the output may also include: During the subsequent online operation phase, each platform directly retrieves the track number mapping pairs between heterogeneous platforms in real time by searching the track mapping lookup table, and skips the cross-platform global matching and association calculation. Each platform, within each time step, independently performs a single-frame association determination between local measurements and local tracks: if a valid local measurement is determined, filtering updates are performed using that local measurement; if no valid local measurement is determined, track updates based on cooperative information or prediction maintenance based on the target kinematics model are performed according to the cooperative platform information corresponding to the track mapping lookup table index.

[0090] Specifically, the present invention provides a real-time collaborative fusion method for tracks under weak computational conditions, including S108: when both local measurement and collaborative information are unavailable, track continuity is maintained solely based on the motion model, including the following steps: 81) When step 51) determines that there is no valid local measurement that satisfies the elliptical tracking gate and is selected by the platform association at the current time step, and the cooperative state estimate of the corresponding target cannot be obtained from the pairing platform based on the fixed track number mapping relationship, the track is determined to be in a measurement unavailable state.

[0091] 82) In this case, relying solely on the target motion model to predict and maintain the track state, the mean of the predicted state and the predicted covariance matrix calculated in step 213) are directly used as the output at the current time: , In the formula, The current target trajectory status. Let be the state covariance matrix at the current moment.

[0092] S109: Output the current track status and advance the time, including the following steps: 91) Output the track status at the current time step in a unified manner. For any of the above scenarios executed at the current time step, the track status obtained at the current time step will be uniformly represented as... as the target trajectory exist The final estimate of the time.

[0093] 92) Use the obtained track state and covariance as the basis for the next time step. The prior inputs for filtering prediction and data association: , This enables continuous track tracking across time steps. The system then moves to the next time step and, during the online operation phase, only executes steps S102 and S105-S108 in a loop. It directly calls the track number mapping relationship that has been fixed in step S104 for processing and does not repeat the cross-platform global matching in the initialization phase (i.e., skipping S103-S104) until the tracking task ends.

[0094] This invention provides a real-time collaborative fusion method for tracks under weak computational conditions, applicable to multi-sensor platform collaborative target tracking scenarios with limited computational resources. The method constructs a dual-mode architecture of "initial mapping fixation" and "online lookup table-based collaboration." In the initialization phase, each platform independently completes single-frame measurement association and accumulates computational state consistency within a set time window, performing cross-platform global matching to fix the track correspondence between platforms into a lookup table. After entering the online operation phase, the system no longer repeatedly performs cross-platform global matching, but instead cyclically executes local association and hierarchical update steps: when local measurements are valid, local filtering updates are performed first; only when local measurements are determined to be missing, the track state of the paired platform is directly introduced as a collaborative pseudo-measurement for compensation based on the fixed lookup table; when both local and cross-platform measurements are unavailable, prediction maintenance is performed. This invention employs a processing architecture combining "one-time mapping and solidification during the initialization phase" and "condition-triggered collaboration during the online phase." Throughout the process, it relies solely on finite time-window statistics and one-time global assignment calculations. During the online operation phase, a condition-triggered mechanism introduces collaborative information only when necessary, thereby significantly reducing real-time computational load and data association complexity without increasing communication burden. This method effectively solves the problems of maintaining track continuity and estimation stability under conditions of intermittent or missing observations, and is particularly suitable for multi-platform collaborative sensing and target tracking applications with limited computational resources.

[0095] To verify the effectiveness of the proposed real-time collaborative fusion method for track data under weak computational conditions, a simulation scenario involving dual-sensor collaborative observation of three maneuvering targets was set up. The scenario included clutter interference and simulated a period of complete loss of target measurements by the sensors to test the algorithm's cross-platform collaborative capability. Specifically, to verify the algorithm's "weak computational adaptability" in practical engineering applications, this embodiment constrained the algorithm's operating environment to a typical embedded edge computing node—the Rockchip RK3588. Given the computational resource limitations of this platform, this experiment rigorously quantitatively evaluated the algorithm's real-time performance. The evaluation criterion was set as follows: under a 1:10 computational resource margin (i.e., the simulated computational power is only one-tenth of that of a conventional high-performance processing unit), the algorithm's single-step time must still be significantly lower than the 10ms sensor data update cycle to prove that it can meet the hard real-time operation requirements on resource-constrained edge devices such as the RK3588. Finally, the effectiveness of this method was comprehensively evaluated by analyzing the dual-sensor track numbering results, target track estimation accuracy, and computational time statistics.

[0096] Figure 2 The target's motion was displayed, and the initial states of the two sensors were set to... , The initial states of the three targets are respectively set as follows: , and The first three dimensions of the state represent position information, and the last three dimensions represent velocity information. All three targets are making a turning motion with a turning angular velocity of... Two sensors track the target using proportional guidance.

[0097] Figures 3 to 6 The measurement timing distributions of the two sensors at azimuth and elevation angles are shown respectively. Black dots represent target measurements, and gray dots represent random clutter. Specifically, in... Figure 5 (Azimuth angle of sensor 1) and Figure 7 In the (sensor 1 pitch angle) data, it can be clearly observed that the measurement data of one of the targets is missing in the interval from 9.0s to 11.0s. Meanwhile, within the same time period... Figure 6 and Figure 8 The display sensor 2 continues to receive valid measurements of all targets, which provides the necessary cross-platform collaborative compensation data source for sensor 1.

[0098] Figure 7 The results of cross-platform track association are shown. In the figure, the horizontal axis represents the track number of sensor 2, the vertical axis represents the track number of sensor 1, and the black dots indicate the established correspondence between the two platforms. The results show that the global matching method based on track consistency cost proposed in this invention can accurately map tracks a, b, and c of sensor 1 to tracks e, f, and g of sensor 1, respectively, establishing a correct and stable cooperative correspondence.

[0099] Figure 8 The figure shows the trajectory estimation results of sensor 1. In the figure, boxes, circles, and triangles represent the actual trajectories of target 1, target 2, and target 3, respectively, while solid and dashed lines represent the estimated trajectories output by the algorithm. As can be seen from the figure, although sensor 1 experienced a measurement interruption midway, thanks to the cross-platform collaborative mechanism of this invention, sensor 1 successfully maintained continuous tracking of the target (intermediate trajectory) by utilizing the collaborative information provided by sensor 2, without any trajectory interruption or divergence. The estimated trajectory highly matches the true value.

[0100] Figure 9 The trajectory estimation results for sensor 2 are displayed. Sensor 2, acting as a collaborator, did not malfunction during the simulation and stably tracked all targets. (Comparison) Figure 8 and Figure 9 It is evident that the track geometry of the two platforms for the same target is consistent, verifying the synchronicity of multi-platform state estimation in this method.

[0101] Figure 10The results of a comparative analysis of the algorithm's real-time performance and computational complexity on a high-performance computing platform are presented. The figure uses a 10ms sensor data update cycle as the system's hard real-time benchmark, comparing the single-step computation time of the traditional centralized fusion method and the distributed collaborative method of this invention. Experimental data shows that the average single-step computation time of the method of this invention is only about 0.513ms, with a real-time occupancy rate as low as 8.145% within the 10ms time window. Compared with the computationally heavy centralized fusion architecture, this invention improves computational efficiency by approximately 4 times through decentralized distributed processing and a lightweight global association strategy.

[0102] Figure 11 The adaptive verification results of the computing power conversion model based on embedded edge nodes (RK3588) are presented. The gray bars on the left of the figure represent the hard real-time constraint that the system must adhere to (10ms sensor update cycle), while the black bars on the right represent the single-step execution time of the algorithm of this invention under the stringent 1:10 computing power conversion condition (approximately 5.13ms). Comparative analysis shows that even on a weak computing platform with significantly reduced computing power, the execution time of this invention is still far below the system's time safety threshold, not only meeting the hard real-time requirements but also establishing a computing power safety margin of approximately 48.7%.

[0103] The above experiments verify that the present invention can handle multi-target tracking under complex tracking conditions such as clutter interference, intermittent interruption of measurement data, and limited computing resources, and has significant theoretical and practical value.

[0104] The above inventions are merely a few specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A real-time collaborative fusion method for flight tracks under weak computational conditions, characterized in that, Includes the following steps: Establish the target kinematic model and the measurement models of each platform. Each platform independently performs the association matching and state update of single-frame measurement and local track based on global nearest neighbor to form a local track set. Within a preset initialization time window, obtain the local track sets formed by each platform independently, obtain the track consistency cost of pairing all tracks between any two platforms and within their respective local track sets within the initialization time window, and construct a cross-platform track-track cost matrix based on the track consistency cost. The cross-platform track-track cost matrix is ​​globally optimally matched to determine the correspondence between tracks on different platforms, and the correspondence is solidified into a track mapping lookup table. For each local track at the current time step, determine whether there is a valid local measurement; If it exists, the local track is filtered and updated using the local measurement; if it does not exist, the cross-platform track corresponding to the local track is determined according to the track mapping lookup table, and after obtaining its status information as cooperative information, the local track is compensated and updated using the cooperative information; if neither the local measurement nor the cooperative information is available, the local track is predicted and maintained based on the target motion model, and the track status updated at the current time step is output.

2. The real-time collaborative fusion method for flight paths under weak computational conditions as described in claim 1, characterized in that, The process of obtaining the track consistency cost between any two platforms within the platform and for pairwise pairing of all tracks within their respective local track sets within the initialization time window includes the following steps: Pair the tracks in the local track sets of any two platforms across platforms to construct candidate track pairs; For each candidate track pair, obtain its state residuals and composite covariance at each time point within the initialization time window, and calculate the Mahalanobis distance between the tracks at the corresponding time points. The Mahalanobis distances between the tracks at each time point within the initialization time window are statistically processed to obtain the track consistency cost of the candidate track pair.

3. The real-time collaborative fusion method for flight paths under weak computational conditions as described in claim 2, characterized in that, The statistical processing involves calculating an average value, and the track consistency cost is obtained by averaging the Mahalanobis distances within the time window. ， In the formula, For candidate track pairs and Cost of consistency in flight paths; The length of the window; For the track With track Mahalanobis distance between them.

4. The real-time collaborative fusion method for flight paths under weak computational conditions as described in claim 1, characterized in that, The correspondence is solidified into a track mapping lookup table, including the following steps: Introducing binary assignment variables The assigned variables satisfy the feasibility condition of one-to-one pairing; Using the aforementioned binary assignment variable The reported track consistency costs are linearly weighted to construct a global cost function for cross-platform track matching. ; With the global cost function To optimize the objective, a global nearest neighbor algorithm is used to solve the cross-platform track-track cost matrix. Under the constraints, the objective function is obtained. The global optimal matching decision matrix that yields the minimum value ; Through continuous Extract the globally optimal matching decision matrix within each initialization time window. track index pairs with a median value of 1 The frequency of occurrence of each track index pair is counted, and index pairs that meet the stability threshold requirements are solidified into cross-platform track number mapping relationships and stored in the track mapping lookup table.

5. The real-time collaborative fusion method for flight paths under weak computational conditions as described in claim 4, characterized in that, The binary assignment variable Platform for quantitative description With the platform The track matching relationship between them, where: ， To ensure that the track matching results satisfy the one-to-one constraint, the binary assignment variable The following feasibility conditions must be met: ， ， In the formula, and They are respectively traversing the platform With the platform The index number of the track set; Candidate track pairs; For the platform Total number of tracks; For the platform Total number of flight tracks.

6. The real-time collaborative fusion method for flight paths under weak computational conditions as described in claim 5, characterized in that, The global cost function This can be expressed by the following formula: ， In the formula, The cost matrix Located in the middle line, number The matrix elements of the column have a physical meaning as the track consistency cost.

7. The real-time collaborative fusion method for weak computational conditions of flight tracks as described in claim 4, characterized in that, The global optimal matching decision matrix We obtain it from the following formula: ， In the formula, Represents a set of binary assignment variables To optimize the object, while satisfying the one-to-one matching constraint, a search is performed to find the global cost function. The optimal matching combination that achieves the minimum value; The optimal matching decision matrix is ​​obtained by solving for the elements of a binary variable.

8. The real-time collaborative fusion method for weak computational conditions of flight tracks as described in claim 4, characterized in that, The process of solidifying the index pairs that meet the stability threshold requirements into a cross-platform track number mapping relationship and storing them in a track mapping lookup table includes the following steps: For any cross-platform track index pair Define its stability index for: ， In the formula, To initialize the total number of time windows; where, the indicator variable Defined as: ， When the stability index meets Then the track index is considered to be With track index The numbering mapping relationship between them is stable within the time window, and this mapping relationship is solidified into a cross-platform track numbering mapping relationship, wherein the stability threshold is... The setting is based on the system's fault tolerance capability for erroneous associations; Output a fixed cross-platform track number mapping relationship, which serves as the basis for cross-platform collaboration during system operation, and enter the online collaboration phase.

9. The real-time collaborative fusion method for flight tracks under weak computational conditions as described in claim 1, characterized in that, The method of using collaborative information to compensate and update the local track includes the following steps: If no valid local measurement exists at the current moment, the state estimate of the corresponding cooperative platform matching target at the current moment is obtained according to the track mapping lookup table. and the corresponding covariance matrix And project the collaborative target state onto this platform. In the local coordinate system, the transformed state estimate is obtained. and covariance matrix Construct this platform Collaborative pseudo-measurement and its measurement error covariance : ， ， Introducing a collaborative measurement mapping matrix This is used to map the target state to the collaborative measurement space; Constructing the State – Collaborative Measurement Cross-Covariance Matrix and covariance of measurement residuals : ， ， In the formula, To predict the covariance matrix; The collaborative filter gain is derived. : ； Perform cross-platform collaborative track state and its state covariance matrix update: ， ， In the formula, for The target track is updated in real time. The state; for The mean of the predicted state at any given time; for Track updated in real time The state covariance matrix.

10. The real-time collaborative fusion method for weak computational conditions of flight tracks as described in claim 1, characterized in that, After outputting the track status updated at the current time step, it also includes: During the subsequent online operation phase, each platform can directly retrieve the track number mapping pairs between heterogeneous platforms in real time by searching the track mapping lookup table, and skip the cross-platform global matching and association calculation. Each platform only needs to independently perform a single-frame association determination between local measurements and local tracks at each time step: if a valid local measurement is determined, then the local measurement is used to perform filtering updates; if a valid local measurement is determined, then track updates based on cooperative information or prediction maintenance based on the target kinematics model are performed according to the cooperative platform information corresponding to the track mapping lookup table index.

Images