A Multi-Information Fusion Ghost Node Removal and Target Tracking Method

Abstract

This invention discloses a multi-information fusion ghost point removal and target tracking method, which utilizes the dispersion characteristics of target angle collaborative localization to remove false associated points and improve target tracking accuracy. Addressing the problem of decreased tracking accuracy caused by a large number of false associated ghost points, a two-stage ghost point removal and target tracking algorithm based on the fusion of angle measurement and target motion characteristics is proposed. This algorithm adopts a "association first, estimation later" collaborative localization strategy, constructing a view-domain grid map and energy accumulation matrix by establishing a mapping relationship between angle measurement noise and localization error. It analyzes the spatial geometric distribution characteristics between real targets and false associated ghost points within the view domain, designs a novel removal criterion based on the Hough transform, and achieves first-level coarse removal of ghost points. By studying the dispersion characteristics and motion characteristics of the target localization ambiguity area, a predictive tracking gate is constructed using motion parameter identification, achieving second-level fine removal of ghost points from a kinematic perspective.

Description

Technical Field

[0001] This invention relates to the field of multi-target tracking, and more particularly to a method for multi-information fusion ghost point elimination and target tracking. Background Technology

[0002] In the field of passive detection, aircraft swarms equipped with optical and infrared guidance devices estimate the spatial position of targets by extracting the line-of-sight angle information relative to the target and utilizing the principle of direction finding intersection. During multi-target tracking, when the measurement origin is uncertain, multi-path direction finding cross-location inevitably produces a large number of false positioning points, forming spurious tracks in the time series that are difficult to eliminate, severely impacting the accuracy of multi-target tracking. With the increase in the number of aircraft and targets, the number of false points generated by direction finding cross-location grows exponentially. In addition, complex battlefield electromagnetic environments also present measurement noise and significant clutter interference; in some mission scenarios, even decoy targets may appear. Measurement errors cause the azimuth rays of the aircraft targeting the same target to fail to intersect precisely after positioning, instead dispersing within a certain range of positioning ambiguity. These ambiguous points overlap with false correlation points, significantly increasing the difficulty of identifying false points. When clutter and decoy targets are observable, their cross-location points have the same geometric characteristics as the real targets, forming stubborn ghost points that are difficult to distinguish, placing even higher demands on target tracking accuracy.

[0003] Current research on "ghost point" removal mainly revolves around two core paradigms: data association optimization and feature discrimination enhancement. In terms of methodology, researchers have formed a multi-dimensional technical approach by integrating geometric constraint theory, pattern recognition technology, and uncertainty reasoning methods: First, based on geometric feature difference analysis, false point identification is achieved by constructing spatial distribution constraint or motion feature discrimination models; second, by enhancing measurement redundancy, association ambiguity is reduced through multi-view observation data fusion or temporal feature joint analysis; third, improved pattern recognition algorithms are used to accumulate energy and extract features from the original measurement information, establishing more robust discrimination criteria. In the process of technological evolution, position optimization strategies based on sensor maneuvering have improved the completeness of observation information, while improved transform domain analysis methods have enhanced feature separability in complex scenarios. However, existing methods still face significant contradictions between accuracy and computational efficiency, and between scenario universality and algorithm complexity, particularly manifested in insufficient robustness to dynamic environments, limited efficiency in high-dimensional data processing, and discrimination failure under multiple interference coupling.

[0004] In summary, with the increase in the number of observed targets and the increasing complexity of the battlefield electromagnetic environment, the large number of false targets under direction finding and location significantly reduces the accuracy of multi-target tracking and the computational efficiency of the system, thus restricting the development of multi-target tracking in the field of multi-vehicle swarms. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-information fusion ghost point removal and target tracking method, which, from multiple dimensions such as field of view, homology, and consistency, achieves effective removal of associated ghost points and improves the accuracy of multi-target tracking.

[0006] The technical solution to achieve the purpose of this invention is: a multi-information fusion ghost point removal and target tracking method, comprising:

[0007] Step 1: Establish the mapping relationship between angle measurement noise and positioning error, and construct the view grid map and energy accumulation matrix;

[0008] Step 2: Analyze the spatial geometric distribution characteristics between real targets and false association ghost points within the field of view, and design a removal criterion based on the idea of ​​Hough transform to achieve first-level coarse removal of ghost points;

[0009] Step 3: By using the scattering features and motion characteristics of the target localization ambiguity area, a predictive tracking gate is constructed using motion parameter identification to achieve secondary fine removal of ghost points from a kinematic level.

[0010] Further, in step 1, constructing the view raster map and energy accumulation matrix is ​​as follows:

[0011] Based on the aircraft's position, detection range, and detection angle, a fuzzy detection field of view is constructed for the aircraft. In a multi-aircraft, multi-target tracking system, the position of the aircraft members at the cluster boundary is taken as the origin O, with the forward detection direction as the y-axis, to construct a relative rectangular coordinate system. The distance between two aircraft is the positioning baseline length D, the maximum forward detection range of the aircraft is denoted as r, and the target's azimuth angle relative to the aircraft is denoted as θ. Using the aircraft's position as a reference, the boundary of the primary fuzzy locatable field of view in the two-dimensional XOY plane is determined as follows:

[0012]

[0013] Among them, S l and S r S is the x-axis view boundary. u and S b The y-axis is the boundary of the field of view. and These are the minimum values ​​of the spacecraft coordinate set, and Let be the maximum values ​​of the aircraft coordinate set; and let the coordinates of the target associated point be denoted as . N is the total number of associated points; the criterion for the associated intersection points to fall into the primary fuzzy locatable field of view is designed as follows:

[0014]

[0015] By statistically analyzing the grid distribution of associated points in the view map, an energy accumulation matrix is ​​constructed. Combined with the geometric distribution characteristics of ghost points and real targets, a first-level removal of false associated ghost points is achieved. Let I0 be the horizontal and vertical view indices in the two-dimensional XOY plane. row and I col Based on the coordinates of the suspected target falling into the primary fuzzy locatable field of view The index calculation rules can be designed as follows:

[0016]

[0017] Where g represents the prior parameters of the grid cell designed based on the optimal topology configuration;

[0018] When calculating the index according to the above rules, the point position and the grid size are not necessarily multiples of each other, so a rounding operation is required to obtain a valid and predetermined index. Considering that when the associated point position falls on the boundary of the view map, the rounding operation may cause the grid map boundary index to not fall within the view map, a padding process is required. Forced constraints are applied to index numbers that exceed the map boundary to ensure they fall into the map, preventing point loss. In the presence of observation errors, the landing points of the associated points of the actual target will form a positioning ambiguity area. Since the grid cell division is fixed and ordered, the intersection points within the positioning ambiguity area of ​​the same target may be scattered in different grids, leading to energy accumulation and dissipation problems. To avoid the single grid cell being unable to function when the target ambiguity area is located at the grid boundary... The method covers all suspected target points, which leads to incomplete statistics of target associated points under the raster index. To prevent the loss of block boundary features, multiple sets of cross-raster maps are constructed and the raster density is increased, thereby ensuring the accuracy of target positioning. The cross-design of the raster map is carried out from the x-axis and y-axis of the view, and the raster density is achieved by moving the raster cross offset. Although the index boundaries of different raster maps are different, the position coordinates of suspected target points under the locatable view map are uniform, ensuring the consistency of multiple raster indexes. By arranging the discrete points of the direction finding cross-positioning in the view plane and statistically voting on the discrete points accumulated by the raster cells, the energy accumulation matrix under the raster map is constructed.

[0019] Furthermore, in step 2, a novel elimination criterion is designed based on the idea of ​​Hough transform, specifically as follows:

[0020] Theoretically, the intersection points of corresponding targets should coincide, meaning the points will fall within the same accumulation unit. Accumulating the voting results for each accumulation unit will yield energy matrices for different accumulation units; the more overlapping points, the higher the accumulation value and the maximum energy value. In each observation cycle, under dual-station collaborative angle cross-positioning, each target can always obtain... There are several intersection points, and these intersection points coincide with each other. However, false positioning points obtained due to incorrect angle correlation will be scattered throughout the detection field of view, and their distribution characteristics are determined by the spatial layout of the aircraft and the position of the target. Regardless of the relative positions of the aircraft and the target, and the number of sensors selected, the target location containing the correct angle correlation intersection points will always have the most points. Therefore, to obtain the measurement belonging to the target from numerous intersection points, it is necessary to judge based on the overlap of the intersection points. It is known that the target's true intersection points coincide and the number of coincidences is... Based on the calculated energy accumulation matrix, two cases are considered:

[0021] (1) When the number of intersections that do not coincide or coincide with each other is less than When the intersection point is identified as a ghost point, it is removed.

[0022] (2) When the number of overlapping intersections is equal to If there are m such overlapping points, then this m point is determined to be the target point; if there are m+m * If there are such overlapping points, then there still exists m. * A false point still needs to be judged.

[0023] Furthermore, in step 3, a predictive tracking gate is constructed using motion parameter identification, and a two-stage fine-tuning scheme for ghost point removal is designed, specifically as follows:

[0024] Secondary point cluster elimination is based on identifying target motion characteristics using historical trajectories and calculating the target dispersion area by combining the target positioning ambiguity area with the current motion characteristics, thereby generating a secondary predictive tracking gate. Point clusters falling within the predictive tracking gate are valid target cross-location points in the continuous time series, while targets falling outside the predictive tracking gate are false targets due to measurement noise and will not comply with motion constraints at the track information level. By identifying the target's current motion parameters, assuming the target's current velocity is v... d The flight sampling time is t, and the target's maximum speed is v. m Then the target's maximum maneuver distance is

[0025] L t =v m t (4)

[0026] Using a point on the boundary of the ambiguous positioning area as the center, and L... t Draw a circle with radius L, which is the target maneuver dispersion circle centered at that point. The outer edges of all target maneuver dispersion circles form an extended closed boundary, approximately equivalent to an ellipse; let L be the radius L. a and L bLet be the major and minor axes of the elliptical target scattering region, respectively. The minor axis is approximately half of the major axis, and the major axis is approximately the sum of the distance from the current target position to the longitudinal endpoint of the ambiguity region and the target prediction step size. The approximate calculation formula is as follows:

[0027]

[0028] Δθ max For the direction finding error, θ1 and θ2 are the direction finding angles between the carrier aircraft and the target, and γ is the intersection angle for cross-positioning.

[0029] It can be seen that at the target's maximum speed v m Under certain conditions, the geometric characteristics of the target elliptical dispersion region and the direction finding error Δθ max The azimuth angle θ and the flight sampling interval t are related; by selecting the current cross-positioning coordinates as the center of the elliptical target dispersion area, and combining the major and minor axis parameters of the elliptical target dispersion area obtained from identification calculation, the equation of the elliptical dispersion area can be generated; let the coordinates of the target direction finding cross-positioning be (x T ,y T If a point lies within the ellipse, then the criterion for that point can be denoted as:

[0030]

[0031] Considering the target's motion, the dispersion area will have directionality, meaning the target dispersion ellipse will have a rotation angle α. For an ellipse with a rotation angle α, the general criterion for a point falling within the ellipse can be denoted as:

[0032]

[0033] Wherein, the rotation angle α is determined by the target's current position (x) T ,y T ) and the predicted position at the next time step (x′) T ,y′ T Perform difference calculations:

[0034] α=aractan((y T -y′ T ) / (x T -x′ T (8)

[0035] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0036] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the above-described method.

[0037] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] (1) To address the uncertainty in target positioning caused by inaccurate angle measurement in collaborative positioning, the mapping relationship between angle measurement noise and target positioning error was explored. A field-of-view grid map and energy accumulation matrix were constructed, and a first-level ghost point removal criterion was designed based on the ghost point distribution characteristics.

[0040] (2) Considering the distribution characteristics of the fuzzy geometric region of target positioning and the target motion characteristics, a predictive tracking gate was constructed by combining motion parameter identification, which realized the effective removal of ghost points in two stages.

[0041] (3) Adopting the collaborative positioning strategy of "association first and estimation later", a two-level ghost point elimination and target tracking algorithm based on the fusion of angle measurement and target motion characteristics is proposed. Attached Figure Description

[0042] Figure 1 This is a single-target localization and tracking scenario using multiple observation platforms in this invention.

[0043] Figure 2 This is a scenario of false correlation among multiple observation platforms and multiple targets in this invention.

[0044] Figure 3 This is the two-dimensional in-plane angle cooperative target observation model in this invention.

[0045] Figure 4 This is a schematic diagram of the ambiguous area in the dual-station angle collaborative positioning of this invention.

[0046] Figure 5 This is a schematic diagram of the dual-station angle collaborative positioning prediction tracking gate design in this invention.

[0047] Figure 6 This is a schematic diagram of the multi-maneuver target mission setting in this invention.

[0048] Figure 7 This is a schematic diagram of the target tracking number elimination result in this invention.

[0049] Figure 8 This is a schematic diagram of the entire OSPA process in this invention.

[0050] Figure 9 This is a schematic diagram of the target tracking number elimination result in this invention.

[0051] Figure 10 This is a schematic diagram of the target tracking number elimination result in this invention.

[0052] Figure 11 This is a schematic diagram of the discrete-time tracking scenario in the simulation experiment of this invention.

[0053] Figure 12 This is a schematic diagram of the discrete-time two-stage prediction gate in the simulation experiment of this invention.

[0054] Figure 13 This is a schematic diagram (cluster) of discrete-time target removal in the simulation experiment of this invention.

[0055] Figure 14 This is a schematic diagram of the target's x-axis motion velocity identification under simulation experiment in this invention.

[0056] Figure 15 This is a schematic diagram of the target's y-axis motion velocity identification under simulation experiment in this invention.

[0057] Figure 16 This is a schematic diagram of the probabilistic interaction results of the maneuver model under simulation experiments in this invention.

[0058] Figure 17 This is a comparison chart of the point removal performance under different grid sizes in the simulation experiment of this invention. Detailed Implementation

[0059] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0060] This invention proposes a two-stage ghost point removal and target tracking algorithm based on a "association-then-estimation" strategy. By analyzing the mapping relationship between angle measurement noise and target localization error, a view-domain grid map and energy accumulation matrix are constructed, and a first-stage ghost point removal criterion is designed based on ghost point distribution characteristics. Taking into account the distribution characteristics of the target localization fuzzy geometric region and target motion characteristics, a predictive tracking gate is constructed by combining motion parameter identification, achieving effective second-stage ghost point removal. The implementation process includes the following five specific steps.

[0061] Step 1: Multi-station angle collaborative positioning model

[0062] As can be seen from the principle of infrared passive detection, the target position can be calculated using the position information and target angle information provided by different observation base stations. For a multi-vehicle, multi-target tracking system, assuming the target's position coordinates in this scenario are X... T =[x T ,y T ,z T ] T The coordinates of the i-th aircraft member are S. i =[x i ,y i ,z i The observation is the azimuth angle β between the i-th aircraft and the target. iand pitch angle The specific expressions are as follows:

[0063]

[0064] β i =arctan((y T -y i ) / (x T -x i (10)

[0065] Considering a dual-station angle-coordinated target localization scenario, a three-dimensional Cartesian coordinate system is established using the xoy plane as the horizontal plane, such as... Figure 1 As shown. When the coordinates of the aircraft are known, formulas (9)-(10) can be rearranged to obtain:

[0066]

[0067] It is worth mentioning that the distribution relationship between the target and the aircraft observation platform does not affect the azimuth angle and relative positioning relationship, but the range of azimuth angle value β must be guaranteed. i ∈[0,2π]. Considering that the number of aircraft in the cluster system is n and satisfies n≥2, then formula (11) can be written as:

[0068] A i X T =b i (12)

[0069] Among them, A i and b i The specific format is as follows:

[0070]

[0071] At this point, we can obtain the least-squares expression for solving the target position matrix:

[0072] X T =[x T ,y T ,z T ] T =(A T A) -1 A T b (15)

[0073] In the formula, A and b are A i and b i A matrix formed by sorting and combining rows by index from smallest to largest.

[0074] Step 2: Multi-angle, multi-target, correlation, and ghost point analysis

[0075] When multiple targets exist, each aircraft acquires multiple angle information, and it is unknown whether the measurement information was generated by a target or by which target. The cross-localization achieved by associating different direction-finding information can generate a large number of false points. To illustrate the relationship between the aircraft's tracking line-of-sight and its known parameters, and to visually demonstrate the generation and distribution characteristics of these false points, let the aircraft's position on the two-dimensional plane be S. i (x i ,y i ), i = 1, 2, ..., n, where n is the total number of aircraft in the swarm system, and the target position is T. k (x k ,y k Let k = 1, 2, ..., m, where m is the total number of targets being tracked. Taking three aircraft tracking three targets as an example... Figure 2 A schematic diagram of spurious associations in a typical task scenario within a two-dimensional plane is provided. In the diagram, black solid dots represent real targets, and green solid dots represent spurious ghost points. Ghost point problems are unavoidable in multi-target tracking. Analyzing the main factors contributing to ghost point occurrence and employing different ghost point removal algorithms based on different application scenarios can reduce their impact on multi-target tracking accuracy.

[0076] Generally, in a multi-vehicle swarm cooperative multi-target tracking system, we assume that the number of aircraft n > 2, the number of moving targets is m, the detection probability of the target by the electro-optical platform on each aircraft is 1, and all angle observations originate from the target. Within each observation period, the aircraft are grouped in pairs, and the maximum number of intersection points is... Among them are There are 10 real intersection points (intersection points corresponding to the target), and the number of intersection points corresponding to each target is 10. There are [number] intersections, and without considering measurement errors, these intersections coincide. Based on the relationship between the number of aircraft and targets, the distribution of intersections at the ghost point is as follows:

[0077] (1) If n > m, the maximum number of intersections at the ghost point is:

[0078] (2) n = m, the maximum number of intersections at the ghost point is Furthermore, only one such point can exist.

[0079] (3) If n < m, the maximum number of intersections at the ghost point is: Furthermore, there may be multiple points in this situation.

[0080] Step 3: Grid Map Construction in Clutter Environment

[0081] Based on parameters such as aircraft position, detection distance, and detection angle, a fuzzy detection field of view is constructed for the aircraft. This allows for the initial elimination of false correlation intersections falling outside the detection field of view, reducing computational complexity and minimizing the impact of ghost points on target tracking accuracy. In a multi-aircraft, multi-target tracking system, the location of the aircraft members at the cluster boundary is taken as the origin O, with the forward detection direction as the y-axis, constructing a relative Cartesian coordinate system. The distance between two aircraft is the positioning baseline length D, and the maximum forward detection distance of the aircraft is denoted as r. Using the aircraft position as a reference, the boundary of the initial fuzzy locatable field of view can be determined as follows:

[0082]

[0083] Among them, S l and S r S is the x-axis view boundary. u and S b The y-axis is the boundary of the field of view. and These are the minimum values ​​of the spacecraft coordinate set, and These are the maximum values ​​of the aircraft coordinate set. Let the coordinates of the target associated point be denoted as . N is the total number of associated points. The criterion for determining whether an associated intersection point falls within the primary fuzzy locatable field of view can be designed as follows:

[0084]

[0085] The accuracy of target positioning will be limited by the influence of angle measurement errors. The three-dimensional spatial target observation model can be derived using a two-dimensional approximation method. Therefore, Figure 3 A two-dimensional plane angle-coordinated target observation model is presented to analyze the coupling relationship between angle error and positioning accuracy. The angle measurement information of the two aircraft is as follows:

[0086]

[0087] In the formula: (θ1, θ2) are the true azimuth angles; (v1, v2) represents the azimuth angle measurement; (v1, v2) represents the noise measurement.

[0088] Let (x1, y1) and (x2, y2) be the actual positions of the two spacecraft, respectively. The target position information contains errors, γ is the azimuth angle between the two aircraft relative to the target, and γ = θ2 - θ1. For example... Figure 3 As shown, their directional and positional relationships can be described as follows:

[0089]

[0090] At the same time, the target coordinates (x, y) can be solved based on the relative positions of the three. T ,y T )

[0091]

[0092] Expanding the above equation at (θ1, θ2) using Taylor and neglecting higher-order terms (second order and above), we can approximate the expression for the position error.

[0093]

[0094] Calculate the distance (r1, r2) from the aircraft to the target based on geometric relationships. Construct a right triangle with r2 as the hypotenuse. The geometric relationships are as follows:

[0095]

[0096] Therefore, we can solve it.

[0097]

[0098] Calculate the partial differentials of each term, and equation (21) can be rewritten as follows:

[0099]

[0100] The target estimation position error is

[0101]

[0102] If we disregard the positional error of the aircraft and assume that the maximum direction-finding error of the two aircraft (or a single aircraft at different times) is the same and is Δθ, then... max Therefore, the target's true location should be within the quadrilateral ABCD where the spacecraft's detection sector intersects, such as... Figure 4 As shown. Since the direction-finding error is within ±Δθ... max The target's true location could be any point within the quadrilateral ABCD region, meaning the target's actual position could be anywhere within that region. Since the target's true location within quadrilateral ABCD cannot be determined, the quadrilateral ABCD region is called the positioning ambiguity region. According to the sine theorem, the ratio of the sine of each side to the sine of its opposite angle is equal. When the baseline distance D is constant, the lengths of sides AU1, BU1, and CU1 can be obtained using the triangle cosine theorem as follows:

[0103]

[0104] Clearly, quadrilateral ABCD is an axially symmetric figure, and its area is equal to twice the area of ​​triangle ABC. The area of ​​triangle ABC can be calculated by considering AB as the base, and the altitude corresponding to the base can be approximated by a segment of a circle.

[0105]

[0106] If we take Δθ max Since it is a small quantity, its influence in trigonometric function calculations can be ignored to simplify the calculation; that is, the area of ​​quadrilateral ABCD can be approximated as...

[0107]

[0108] It can be seen that S ABCD The magnitude of the direction finding error Δθ, besides the baseline distance D, is also related to the distance from the baseline. max In addition to the above, it is also related to the direction finding θ1 and θ2 of the carrier aircraft relative to the target. Generally speaking, Δθ max It depends on the performance of the carrier aircraft's infrared direction finding. In Δθ max Under certain conditions, the smaller the baseline distance D, the greater the S ABCD The smaller the value, the better. If the baseline distance D is constant, then S... ABCD The size of θ1 and θ2 depends mainly on θ1 and θ2.

[0109] In practical aircraft swarm applications, the baselines between aircraft can be acquired and maintained for formation adjustments, and can be considered as prior information. However, the observation angles for non-cooperative targets change dynamically and randomly with the target's movement, constituting unknown observation information and containing observation errors. Therefore, the area and distribution of the positioning ambiguity zone are more influenced by the observation angle. Considering a fixed baseline distance, let...

[0110] W = sinθ1sinθ2 / sin 3 (θ1+θ2) (32)

[0111] To make S ABCD Minimum, should satisfy The solution can be obtained by arranging the parts.

[0112]

[0113] In other words, when the baseline distance is constant, the area of ​​the corresponding positioning ambiguity region is only at its minimum value when the base angles satisfy θ1 = θ2 = π / 6, at which point the intersection angle γ = 120°. The area of ​​the positioning ambiguity region can then be rewritten as:

[0114]

[0115] The basis for removing ghost points is that overlapping points fall into the same accumulation unit. Therefore, it is first necessary to construct a grid map to segment the localization field of view, generating several grid accumulation units. The size of the grid accumulation unit is related to the dispersion area of ​​the current target localization ambiguity area. Considering that the size of the target localization ambiguity area will change under dynamic localization topology, which is not conducive to the continuity of the target in the time series, it is possible to consider designing the prior parameters of the grid unit based on the optimal topology configuration and correction parameters.

[0116] Step 4: Ghost Node Removal Method at the Next Level of the Energy Accumulation Matrix

[0117] By statistically analyzing the grid distribution of related points in the view map, an energy accumulation matrix is ​​constructed. Combined with the geometric distribution characteristics of ghost points and real targets, primary elimination of falsely associated ghost points can be achieved. Specifically, the coordinates of suspected targets falling into the primary fuzzy locatable view are: The index calculation rules are designed as follows:

[0118]

[0119] When calculating the index according to the above rules, the point position and the raster size are not always multiples of each other, so a rounding operation is required to obtain a valid, predetermined index. Considering that when the associated point position falls on the view map boundary, the rounding operation may cause the raster map boundary index to not necessarily fall within the view, a certain amount of padding is needed. This involves applying a forced constraint to index numbers exceeding the map boundary to ensure they fall within the map, preventing point loss. Specifically, taking row indexes as an example:

[0120]

[0121] Among them, I max This is the theoretical maximum index boundary when constructing a raster map; exceeding this index value indicates falling outside the view map.

[0122] When observation errors exist, the landing points of the associated points of the true target will form a certain range of positioning ambiguity. Since the grid cell division is fixed and ordered, the intersection points within the positioning ambiguity area of ​​the same target may be scattered across different grids, leading to energy accumulation and dissipation. To avoid the problem that a single grid cell cannot cover all suspected target points when the target ambiguity area is located at the grid boundary, thus causing incomplete statistics of target associated points under the grid index, multiple sets of cross-grid maps can be constructed and the grid density increased to prevent the loss of block boundary features, thereby ensuring target positioning accuracy. The cross-grid map design can be carried out from both the x-axis and y-axis dimensions of the view, achieving grid density by shifting the grids by a certain cross-offset. Although the index boundaries of different grid maps differ, the position coordinates of suspected target points under the locatable view map are uniform, ensuring the consistency of multiple sets of grid indexes.

[0123] Discrete points for direction finding cross-location are arranged in the field of view plane, and the discrete points accumulated by the grid cells are statistically voted on. Theoretically, the cross-points corresponding to the target should coincide, meaning the points will fall into the same accumulation cell. The voting results for each accumulation cell are accumulated to obtain the energy matrix of different accumulation cells. Clearly, the more overlapping points, the higher the accumulated value and the maximum energy value. In each observation cycle, under dual-station cooperative angle cross-location, each target can always obtain... There are several intersection points, and these intersection points overlap with each other. However, false positioning points obtained due to incorrect angle correlation will be scattered throughout the detection field of view, and their distribution characteristics are determined by the spatial layout of the aircraft and the position of the target. Regardless of the relative positions of the aircraft and the target, and the number of sensors selected, the target location containing the correct angle-correlated intersection points will always have the most points. Therefore, to obtain measurements belonging to the target from numerous intersection points, one can determine this based on the overlap of the intersection points. It is known that the target's true intersection points overlap, and the number of overlaps is... Based on the accumulated results, we can make judgments in two cases:

[0124] (1) When the number of intersections that do not coincide or coincide with each other is less than When the intersection point is identified as a ghost point, it can be removed.

[0125] (2) When the number of overlapping intersections is equal to If there are m such overlapping points, then this m point is determined to be the target point; if there are m+m * If there are such overlapping points, then there still exists m. * A false point still needs to be judged.

[0126] Let the energy matrix of the accumulation unit be p, and the minimum energy value for the suspected target cluster to be established be p. Where c (c < 1) is the scaling empirical factor. If p < p T The cluster does not hold if p ≥ p T The cluster was determined to be a suspected target point cluster.

[0127] Step 5: Method for Secondary Ghost Target Removal under Predictive Tracking Gate

[0128] The core idea of ​​secondary point cluster elimination is to identify the target motion characteristics based on historical trajectories and calculate the target dispersion area by combining the target positioning ambiguity area with the current motion characteristics, thereby generating a secondary predictive tracking gate. Point clusters falling within the predictive tracking gate are the effective target cross-location points in the continuous time series, while targets falling outside the predictive tracking gate are false targets due to measurement noise, which will not comply with motion constraints at the track information level. Therefore, under reasonable design conditions, the problem of false targets can be further eliminated through predictive tracking gates based on motion characteristics. Specifically, the size and shape of the target dispersion area depend not only on the geometric characteristics of the positioning ambiguity area but also on the target's heading and speed. When the target's speed and heading are unknown, the target may start from any point within the positioning ambiguity area and maneuver in any direction within a certain speed range. By identifying the target's current motion parameters, assuming the target's current speed is v... d The flight sampling time is t, and the target's maximum speed is v. m Then the target's maximum maneuver distance is

[0129] L t =v m t (37)

[0130] Using a point on the boundary of the ambiguous positioning area as the center, and L... t Draw a circle with radius , which is the target maneuver dispersion circle centered at that point. The outer edges of all target maneuver dispersion circles form an extended closed boundary, approximately equivalent to an ellipse, such as... Figure 5 As shown in the figure. U1 and U2 are the positions of the carrier aircraft. Taking the position of U1 as the origin O, and placing U2 on the positive x-axis, the distance D between U1 and U2 is the positioning baseline length, Δθ. max The direction finding error is represented by θ1 and θ2, which are the direction finding angles between the carrier aircraft and the target. γ is the intersection angle for cross-positioning. Quadrilateral ABCD is the positioning ambiguity area. T is the target dispersion center, which is also the cross-positioning point at the current moment.

[0131] Remember L a and L b Let be the major and minor axes of the elliptical target scattering region, respectively. The minor axis is approximately half of the major axis, and the major axis is approximately the sum of the distance from the current target position to the longitudinal endpoint of the ambiguity region and the target prediction step size. The approximate calculation formula is as follows:

[0132]

[0133] It can be seen that at the target's maximum speed v m Under certain conditions, the geometric characteristics of the target elliptical dispersion region and the direction finding error Δθ maxThe azimuth angle θ and the flight sampling interval t are related. By selecting the current cross-location coordinates as the center of the elliptical target dispersion area, and combining this with the major and minor axis parameters of the elliptical target dispersion area obtained from identification calculations, the equation for the elliptical dispersion area can be generated. Let the coordinates of the target direction finding cross-location be (x...). T ,y T If a point lies within the ellipse, then the criterion for that point can be denoted as:

[0134]

[0135] Considering the target's motion, the dispersion area will have a certain directionality, meaning the target dispersion ellipse will have a certain rotation angle α. For an ellipse with a rotation angle α, the general criterion for a point to fall within the ellipse can be denoted as:

[0136]

[0137] The rotation angle α can be obtained by differential calculation of the target's current position and the predicted position information of the next time step.

[0138] α=aractan((y T -y′ T ) / (x T -x′ T (41)

[0139] Example

[0140] To verify the effectiveness of the proposed algorithm in ghost point elimination during maneuvering target tracking and the accuracy of tracking model parameter identification, a simulation case was designed, assuming four friendly aircraft hovering and tracking six enemy moving targets. The task settings are as follows: Figure 6 As shown. The simulation parameters are as follows: the hovering positions of the aircraft are (0m, 500m, 3000m), (900m, 600m, 3000m), (2000m, 850m, 3000m), (3000m, 500m, 3000m), and the movement patterns and parameters of the enemy targets are shown in Table 1.

[0141] Table 1 Initial Parameters and Task Settings for Multi-Objective Motion

[0142]

[0143] The forward detection range of the aircraft equipped with the reconnaissance payload is 5000m, the detection azimuth angle is 80°, and the detection elevation angle is 25°. It is also assumed that the sampling period of the infrared seeker is T = 0.1s, and the direction-finding error is δ. θ=0.1°. In the algorithm design, the energy accumulation unit size is approximately 25m, 30m, 30m. Clutter in each cycle is randomly distributed uniformly within the aircraft's line-of-sight space. Let the given unit clutter number be λ = 5, and the clutter parameter per unit area be γ = λ × 10. -6 That is, every 1×10 -6 m 2 λ clutter is generated within the field of view. The total number of clutter within the tracking field of view in each cycle can be determined based on the field of view size, while assuming a clutter detection probability of 0.5. Considering that the CV model can be regarded as a special form of the CT model that maintains a zero conversion rate, the CA model and CT model are used as maneuvering target tracking models in this section. For the turning tracking model, the turning rate parameters are selected as [-8° / s, 2° / s, 5° / s] to generate three turning tracking models under different turning rates. The initial probability of the model set is μ = [0.5, 0.5], and the model transformation probability matrix is ​​selected as...

[0144] To verify the effectiveness of the algorithm, we analyzed its performance in terms of both the ability to eliminate false associations and ghost points, and the accuracy of target tracking. OSPA distance is a commonly used evaluation metric in multi-target tracking, used to measure the difference between the estimated target and the actual situation; essentially, it measures the degree of difference between different sets. The distance between any two vectors is defined as follows:

[0145] d c (x,y)=min(c,d(x,y)) (42)

[0146] In the formula, parameter c represents the cutoff point, which is a constant greater than 0, reflecting the sensitivity of OSPA distance to potential error. Let there be a set of true target positions X = {x1, x2, ..., x...} m The target estimated location set Y = {y1, y2, ... y} n The p-order OSPA distance between the two is defined as follows:

[0147]

[0148] In the formula A n Let c be the set of all permutations of {1,2,…,n}, where π(i) represents the i-th element of the π-th combination; p represents the order reflecting the sensitivity of the OSPA distance to outliers. The sensitivity of target estimation error is adjusted by adjusting the values ​​of c and p. In this patent, c = 100m and p = 1 are selected.

[0149] Figure 7 It shows the target's entire tracking trajectory on the x and y axes. Figure 8 This presents the OSPA results for multi-target tracking. In terms of tracking accuracy, as shown... Figure 8As shown, the OSPA distance following algorithm decreases progressively with each optimization step. The OSPA distance after removing the first-level energy accumulation matrix is ​​7.0975 meters, and after processing with the second-level predictive tracking gate, it is reduced to 1.5626 meters, resulting in an overall improvement in tracking accuracy of 77.98%. Figure 9 This is a schematic diagram illustrating the removal of associated points under each level of the algorithm. Figure 10 The number of targets tracked at each algorithm level is shown. From the perspective of tracking effectiveness, each algorithm level achieved the elimination of false associated targets to varying degrees. The total number of observable targets containing clutter in the field of view was approximately 80, which was reduced to approximately 40 after the first-level energy matrix processing, with an overall elimination rate of 50.384%. After further refinement by the second-level prediction gate, effective target tracking was achieved. Experimental results show that in a complex environment with false associated points and clutter, all six targets were effectively tracked.

[0150] To further demonstrate and verify the effectiveness of the algorithm and the logic of the multi-level target elimination algorithm, we take discrete time t=100 as an example. Figures 11-13 The multi-target elimination and tracking performance of the algorithm proposed in this chapter is presented. Figure 11 This diagram illustrates a discrete-time target tracking scenario. In the diagram, pentagrams represent friendly aircraft, blue snowflakes represent all points associated with the observation angle (sources include both the target and clutter), red hollow circles represent suspected target points after primary rejection according to the energy accumulation criterion on the grid map, and green solid circles represent target points after secondary rejection under the predictive tracking gate. As shown in the diagram, the point rejection capability increases progressively. Table 2 records the statistical results of point and target measurement and rejection.

[0151] Table 2. Performance Statistics of Multi-Station Cooperative Multi-Maneuver Target Elimination and Tracking

[0152]

[0153] Even after the first-level ghost point removal, there may still be suspected target clusters that conform to the point cluster generation rules. In the second-level tracking prediction gate, a prediction tracking gate can be constructed based on the target's motion characteristics, such as... Figures 12-13 As shown in the yellow ellipse. Figure 12 This is a schematic diagram illustrating the removal of suspected target clusters under the secondary predictive tracking gate. Figure 13The image shows a magnified view of the target elimination process under the second-level predictive tracking gate for six targets. The diamond-shaped markers indicate the center of the predictive tracking gate at the previous moment. The center of the elliptical predictive tracking gate at the current moment (the yellow hollow circle in the image) is estimated based on the target's motion. The red dots represent the actual targets that have fallen into the second-level predictive tracking gate after the first-level elimination. This allows for the elimination of suspected target clusters outside the tracking gate. Specific tracking performance statistics are shown in Table 2. The table shows that the percentage of first-level target elimination under the energy accumulation matrix at the current moment reaches 59.14%, and the second-level target elimination capability under the predictive tracking gate reaches 91.82%. The first-level OSPA distance is 6.6096m, and the second-level OSPA distance is 1.4142m, representing an accuracy improvement of approximately 78.6%.

[0154] Figure 14 and Figure 15 The figures show the x and y axis velocities of six enemy targets. The figures demonstrate that the target motion parameters have been effectively identified. Based on prior constraints regarding target motion characteristics, decoy targets exhibiting persistent ghost points and stable trajectories can be eliminated. Considering the characteristics of target maneuvering, the algorithm design incorporates an interactive multi-model tracking algorithm. Figure 16 A schematic diagram illustrating the interaction probability switching results of each target tracking model is provided. As shown in the diagram, the center tracking model accurately switches when the motion mode of each target changes, ensuring the accuracy of multi-target tracking and providing accurate support for the construction of the second-level tracking prediction gate in the proposed algorithm. Furthermore, to explore the impact of different grid sizes on the proposed algorithm design, Figure 17 The performance comparison results of the algorithm under different parameter variations are presented. Analysis of the graph shows that when the grid size is too small, the energy accumulation of points in the accumulation cells is insufficient, easily leading to the loss of effective associated points, resulting in missed target detections and increased traversal pressure. When the grid size is too large, the accumulation cells may contain clutter or false associated points, leading to an excessively high false detection rate of target points, ultimately affecting target tracking accuracy and point removal efficiency. Therefore, parameter fitting should be combined with positioning error in the algorithm design, which is beneficial to the robustness and accuracy of the algorithm.

Claims

1. A method for multi-information fusion ghost point removal and target tracking, characterized in that, include: Step 1: Establish the mapping relationship between angle measurement noise and positioning error, and construct the view grid map and energy accumulation matrix; Step 2: Analyze the spatial geometric distribution characteristics between real targets and false associations (ghost points) within the field of view. Based on the Hough transform, design a removal criterion to achieve first-level coarse removal of ghost points; specifically: Theoretically, the intersection points of corresponding targets should coincide, meaning the points will fall within the same accumulation unit. Accumulating the voting results for each accumulation unit will yield energy matrices for different accumulation units; the more overlapping points, the higher the accumulation value and the maximum energy value. In each observation cycle, under dual-station collaborative angle cross-positioning, each target can always obtain... There are several intersection points, and these intersection points coincide with each other. However, false positioning points obtained due to incorrect angle correlation will be scattered throughout the detection field of view, and their distribution characteristics are determined by the spatial layout of the aircraft and the position of the target. Regardless of the relative positions of the aircraft and the target, and the number of sensors selected, the target location containing the correct angle correlation intersection points will always have the most points. Therefore, to obtain the measurement belonging to the target from numerous intersection points, it is necessary to judge based on the overlap of the intersection points. It is known that the target's true intersection points coincide and the number of coincidences is... Based on the calculated energy accumulation matrix, two cases are considered: (1) When the number of intersections that do not coincide or coincide with each other is less than When the intersection point is identified as a ghost point, it is removed. (2) When the number of overlapping intersections is equal to And exist If there are such overlapping points, then this The point is determined as the target point; if it exists. If there are such overlapping points, then there are still [missing information]. A false point still needs to be judged; Step 3: By using the scattering features and motion characteristics of the target localization ambiguity area, a predictive tracking gate is constructed using motion parameter identification to achieve secondary fine removal of ghost points from a kinematic level; Specifically: Secondary point cluster elimination is based on identifying target motion characteristics using historical trajectories and calculating the target dispersion area by combining the target positioning ambiguity area with the current motion characteristics, thereby generating a secondary predictive tracking gate. Point clusters falling within the predictive tracking gate are valid target cross-location points in the continuous time series, while targets falling outside the predictive tracking gate are false targets due to measurement noise and will not comply with motion constraints at the track information level. By identifying the target's current motion parameters, assuming the target's current velocity is... Flight sampling time is The target maximum speed is Then the target's maximum maneuver distance is ； Using a point on the boundary of the ambiguous positioning area as the center, and with Draw a circle with radius , which is the target maneuver dispersion circle centered at that point. The outer edges of all target maneuver dispersion circles form an extended closed boundary, approximately equivalent to an ellipse; denoted as... and Let be the major and minor axes of the elliptical target scattering region, respectively. The minor axis is approximately half of the major axis, and the major axis is approximately the sum of the distance from the current target position to the longitudinal endpoint of the ambiguity region and the target prediction step size. The approximate calculation formula is as follows: ； For direction finding error, and The direction finding angle between the aircraft and the target. The intersection angle is used for cross-location. This is the length of the positioning baseline; It can be seen that at the target's maximum speed Under certain conditions, the geometric characteristics of the target elliptical dispersion region and the direction finding error azimuth angle and flight sampling interval Relevant; selecting the current cross-positioning coordinates as the center of the elliptical target dispersion area, and combining the major and minor axis parameters of the elliptical target dispersion area obtained from identification calculations, the equation of the elliptical dispersion area can be generated; let the coordinates of the target direction finding cross-positioning be... The criterion that a point lies within the ellipse can be denoted as: ； Considering the target's motion, the dispersion area will have directionality, meaning the target dispersion ellipse will have a rotation angle. For those with rotation angle The general criterion for a point to fall within an ellipse can be denoted as: ； Among them, rotation angle From the target's current position Predicted position at the next moment Perform difference calculations: 。 2. The multi-information fusion ghost point removal and target tracking method according to claim 1, characterized in that, Step 1, construct the view raster map and energy accumulation matrix, specifically as follows: Based on the aircraft's position, detection distance, and detection angle, a fuzzy detection field of view is constructed for the aircraft. In a multi-aircraft, multi-target tracking system, the positions of the aircraft members at the cluster boundary are used as the origin of the coordinate system. The forward detection direction is used as Axis, construct a relative rectangular coordinate system; the distance between the two aircraft is the positioning baseline length. The maximum forward detection range of the aircraft is denoted as The target's azimuth angle relative to the aircraft is denoted as... Based on the aircraft's position, determine the two-dimensional... The planar primary fuzzy localizable field of view boundary is: ； in, and for Axial field of view boundary, and for Axial field of view boundary, and These are the minimum values ​​of the spacecraft coordinate set, and Let be the maximum values ​​of the aircraft coordinate set; and let the coordinates of the target associated point be denoted as . , Given the total number of associated points; the criterion for determining whether an associated intersection point falls within the primary fuzzy locatable field of view is designed as follows: ； By statistically analyzing the grid distribution of related points in the view map, an energy accumulation matrix is ​​constructed. Combined with the geometric distribution characteristics of ghost points and real targets, a first-level removal of falsely associated ghost points is achieved. (The last sentence appears to be incomplete and requires further context.) The horizontal and vertical view indices in the plane are respectively and Based on the coordinates of the suspected target falling into the primary fuzzy locatable field of view The index calculation rules can be designed as follows: ； in, These are the prior parameters for the grid cells designed based on the optimal topology configuration; When calculating the index according to the above rules, the point position and the grid size are not necessarily multiples of each other, so a rounding operation is required to obtain a valid and predetermined index. Considering that when the associated point position falls on the boundary of the view map, the rounding operation may cause the grid map boundary index to not fall within the view, padding is required. Forced constraints are applied to index numbers exceeding the map boundary to ensure they fall into the map, preventing point loss. In the presence of observation errors, the landing points of the associated points of the real target will form a positioning ambiguity area. Since the grid cell division is fixed and ordered, the intersection points within the positioning ambiguity area of ​​the same target may be scattered in different grids, leading to energy accumulation and dissipation problems. To avoid the problem that a single grid cell cannot cover all suspected target points when the target ambiguity area is located at the grid boundary, resulting in incomplete statistics of target associated points under the grid index, multiple sets of cross-grid maps are constructed and the grid density is increased to prevent the loss of block boundary features, thereby ensuring target positioning accuracy. The cross-grid map design starts from the view... shaft and The grid is expanded along two dimensions, and grid density is achieved by moving the grid and offsetting it. Although the index boundaries of different grid maps are different, the position coordinates of the suspected target points under the locatable view map are uniform, ensuring the consistency of multiple grid indexes. By arranging the discrete points of the direction finding cross-location in the view plane and statistically voting on the discrete points accumulated by the grid cells, the energy accumulation matrix under the grid map is constructed.

3. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of any of the methods described in claims 1-2.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any of claims 1-2.

5. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any of claims 1-2.

Images