A Radar Target Tracking Method Based on Spatiotemporal Clustering and Multiphysics Model

Abstract

This invention discloses a radar target tracking method based on spatiotemporal clustering and a multi-physics model, belonging to the field of radar target tracking technology. The method includes the following steps: S1, inputting radar point target data; S2, adaptive spatiotemporal clustering to remove noise; S3, establishing a multi-physics model and adaptive threshold weight matching point-by-point and track-by-track; S4, cyclic matching of predicted points; S5, integrating matching scores S1, S2, ..., Sn from the matching score queue; S6, taking the point corresponding to the optimal matching score as the output result. This invention proposes a multi-physics model matching and judgment approach, constructing multiple target motion physical models and combining the target's historical motion state with the current point's adaptability analysis. Even in complex scenarios such as changes in target motion patterns and multiple trajectory intersections, it can accurately identify point ownership, effectively solving the problems of insufficient robustness and poor adaptability of traditional methods, and significantly improving the continuity and accuracy of target tracking in complex scenarios.

Description

Technical Field

[0001] This invention relates to the field of radar target tracking technology, and more specifically to a radar target tracking method based on spatiotemporal clustering and multi-physics models. Background Technology

[0002] In recent years, the rapid development of drone technology has led to a surge in demand for observing aerial point targets that are "low, small, and slow." These targets are characterized by their small size, low altitude, slow speed, and small radar cross-section, posing stringent challenges to the accuracy and anti-interference capabilities of observation systems. Radar can only acquire discrete target point data, while practical applications such as airspace surveillance and security early warning rely on effectively processing these discrete points to achieve accurate target trajectory tracking. Only by constructing continuous and reliable trajectories can subsequent target identification and risk warning be supported.

[0003] In the development of target tracking technology, two core challenges have long hindered breakthroughs: First, radar point target data is mixed with environmental noise, equipment interference and other noise, and efficient noise removal to extract effective target points is the foundation of trajectory construction; Second, different target trajectories have significantly different characteristics and are often accompanied by complex situations such as multiple trajectories intersecting and running concurrently, which makes it extremely difficult to associate and match target points with existing trajectories, and easily leads to trajectory breakage and mismatch problems.

[0004] In denoising and clustering, the traditional DBSCAN algorithm is a classic solution, but its shortcomings become apparent when dealing with complex data of "low, small, and slow" targets: it uses a fixed neighborhood radius and a single spatial distance metric, lacking the ability to adapt to data distribution. Sparse regions are prone to losing trajectory points, while dense regions are prone to mis-clustering noise points. Furthermore, it does not consider key information such as time intervals and height differences, has insufficient multi-dimensional data association mining, poor robustness to noise, and difficulty in adapting to complex motion characteristics.

[0005] In addition, in terms of target point association and matching, traditional methods mostly adopt the nearest neighbor principle, which determines the ownership of the point track based solely on spatial distance. This has extremely strong limitations: it is highly dependent on radar performance and is prone to deviation when the accuracy of the point track is insufficient; it does not consider the target motion characteristics, and is prone to mismatch and trajectory breakage when the target motion pattern changes abruptly. Summary of the Invention

[0006] The purpose of this invention is to provide a radar target tracking method based on spatiotemporal clustering and multi-physics model in order to solve the above-mentioned technical problems.

[0007] To achieve the above objectives, the present invention specifically adopts the following technical solution:

[0008] This invention provides a radar target tracking method based on spatiotemporal clustering and a multi-physics model, comprising the following steps:

[0009] S1, Radar point target data input;

[0010] S2. Adaptive spatiotemporal clustering for noise removal;

[0011] S3. Establish a multiphysics model, and then perform adaptive threshold weight matching point by point and track by track;

[0012] S4. Cyclic matching of prediction points;

[0013] S5. Integrate the matching scores S1, S2, ..., Sn in the matching score queue;

[0014] S6. Take the point corresponding to the best matching score as the result output.

[0015] Specifically, the raw point target data input by radar contains a large number of noise points, which severely interfere with the accuracy of trajectory identification and association. Therefore, it is necessary to first filter out noise from this raw point target data. This scheme uses an adaptive spatiotemporal clustering method to filter out noise points in the raw point target data. Unlike the traditional DBSCAN algorithm, which relies solely on spatial distance for clustering and has insufficient adaptability to spatiotemporal multidimensional data, this paper proposes a method that integrates multidimensional metrics and adaptive neighborhood adjustment.

[0016] In one implementation, step S2 involves adaptive spatiotemporal clustering to remove noise from the original data. The specific steps are as follows:

[0017] S21. Calculate the spatiotemporal weighted distance between any two points:

[0018] Considering the fusion of spatial, altitude, and temporal multi-dimensional information of radar point targets, a metric is defined. and Spatiotemporal weighted distance between two points The calculation formula is as follows:

[0019] ;

[0020] In the formula, , Spatial location; and For height; and For time; These are spatial, vertical, and temporal weights, respectively.

[0021] S22. Calculate the global density:

[0022] For the entire radar point target dataset Define global density as Global Denseness This reflects that the distance between any two points in the dataset is less than The probability mean, This is the global reference radius;

[0023] S23. Calculate the local density:

[0024] For a certain point Its local density is defined as:

[0025] ;

[0026] Local density is accumulated and The inverse mean of the distances to neighboring points is used to quantify the density of a local region; a larger value indicates a denser local area. In the formula, for Count of neighboring points; To avoid constants with a denominator of zero;

[0027] S24. Calculate the adaptive neighborhood radius using the following formula:

[0028] ;

[0029] In the formula, To prevent the smoothing factor from being too heavy; To adjust the amplitude control parameters, by setting... The size (usually between 0.5 and 2) adjusts the adaptive adjustment strength of the neighborhood radius;

[0030] S25. Clustering decision based on adaptive neighborhood:

[0031] Clustering decision rules: two points and Satisfy space-time weighted distance At that time, belong The neighborhood; during the clustering process, a multi-category hierarchical judgment logic is adopted, covering core clusters, associated clusters and discrete points under different density distributions. At the same time, a cluster merging mechanism is applied, and the clustering results are finally processed. If the number of points in a cluster is less than 5 after clustering, it is judged as a noise point set and removed. This threshold can be adjusted according to the actual scenario.

[0032] In one implementation, in step S3, a multi-physics model is established, and the specific definitions of each physical model are as follows:

[0033] S31. Establish a time interval calculation model:

[0034] Assumption The scan time is , The scan time is ,but and Time difference:

[0035] ;

[0036] like Less than the preset time interval threshold If the condition is met, continue calculating the next physical quantity constraint; otherwise, end the current round of matching.

[0037] S32. Establish a spatial distance calculation model:

[0038] Assumption Location information is , Location information is ,So , The spatial distance calculation model is as follows:

[0039] ;

[0040] like Less than the preset distance threshold If the condition is met, continue calculating the next physical quantity constraint; otherwise, end the current round of matching.

[0041] S33. Establish a height difference calculation model:

[0042] Assumption The altitude is , The height of is ,but and Height difference:

[0043] ;

[0044] like Less than the preset height difference threshold If so, continue calculating the next physical quantity constraint.

[0045] S34. Establish a model for calculating the angle between motion vectors:

[0046] If the length of a trajectory is greater than or equal to 3, then the last three points of the trajectory are taken as follows:

[0047] , , ;

[0048] These three points can form two vectors. and ;

[0049] and and Angle between two vectors This is the vector angle at the latest match of the trajectory.

[0050] Earth's curvature correction radius: ;

[0051] Angle to radians: , ;

[0052] Formula for spatial rectangular coordinates: ;

[0053] Constructing motion vectors:

[0054] ;

[0055] ;

[0056] Calculate the magnitude of the motion vector:

[0057] ;

[0058] ;

[0059] Calculate the vector dot product:

[0060] ( ) ;

[0061] Calculate the included angle :

[0062] .

[0063] Specifically, based on the physical characteristics of the motion of various targets observed by radar, such as large aircraft, drones, and airborne objects, four major physical models are constructed. Each model calculates the corresponding physical quantity values ​​through formulas, and then combines them with preset thresholds to form a layer-by-layer constraint matching and judgment mechanism. These models quantify the continuity in the time dimension, the proximity in the spatial dimension, the stability in the vertical dimension, and the consistency of the motion direction, respectively. The calculated four key values—time difference, spatial distance, height difference, and motion vector angle—serve as the core judgment basis for radar point target trajectory point matching.

[0064] In one implementation, the specific steps of point-by-point adaptive threshold weight matching in step S3 are as follows:

[0065] S35. Calculate the time matching score C1;

[0066] S36. Calculate the spatial distance matching score C2;

[0067] S37. Calculate the high-match score C3;

[0068] S38. Calculate the motion vector angle matching score C4;

[0069] S39. Accumulate the scores to obtain the match score S for this round.

[0070] In one implementation, the specific method of cyclic matching of predicted points in step S4 is as follows:

[0071] S41. Obtain the predicted points: P1, P2, ..., Pn through the extrapolation model;

[0072] S42. The predicted points return to step S41 to enter the physical constraint model to calculate the matching score.

[0073] In one implementation, step S4, the adaptive threshold weight matching and cyclic matching are as follows: During the matching process, whenever the radar scans a new point, it needs to match the new point with all existing trajectories in the current situation one by one; if the match is successful, the new point is added to the end of the matched trajectory; if the match fails, the new point is initialized as the starting point of a new trajectory; during the matching process, each trajectory in the current situation needs to be matched with the new point, and the new point can be matched with multiple trajectories at the same time. This requires calculating a matching score when the new point is matched with each trajectory, and selecting the trajectory corresponding to the best score to match the new point.

[0074] For a trajectory, each new point is matched with the last point of the trajectory. During the matching process, the weights of the time interval, spatial distance, height difference, and vector angle are calculated.

[0075] Let the weight of the time interval be... The time matching score is:

[0076] ;

[0077] Let the spatial distance weight be... 2. The spatial distance matching score is:

[0078] ;

[0079] Let the height difference weight be... The high-match score is:

[0080] ;

[0081] Let the weight of the included angle of the vectors be... The matching score for the angle between the motion vectors is:

[0082] ;

[0083] Therefore, the total matching score is:

[0084] .

[0085] In one implementation, in step S4, during radar scanning, environmental factors often lead to situations such as point loss or dropping. These situations directly cause trajectory association interruptions. To solve this problem, it is necessary to extrapolate trajectory points based on uniformly accelerated linear motion. Specifically, when a new point completes a round of matching calculation with a trajectory, it is necessary to extrapolate and predict points for that trajectory, and then use the extrapolated points as the end point of the trajectory to continue matching with the new point. The specific extrapolation method is as follows:

[0086] , , .

[0087] The two time intervals are calculated as follows:

[0088] , ;

[0089] The two velocity vectors are calculated as follows:

[0090] , ;

[0091] The acceleration vector is calculated as follows:

[0092] ,

[0093] Calculate the extrapolation point, assuming the extrapolation point is located at... ;

[0094] ;

[0095] Therefore, the total number of matches between a new point and a trajectory is... Round 1, the first round matches the new point with the end point of the trajectory, the second round matches the new point with the first predicted point of the trajectory, ..., the ... The first round is for adding points and trajectories. The matching score is obtained by matching each predicted point. , The number of scores is determined by the number of extrapolated prediction points.

[0096] In one implementation, in step S6, the point corresponding to the optimal matching score is taken as the result output, specifically as follows:

[0097] Based on the previous section, for a new point, after experiencing... After round-matching, a queue of matching scores can be obtained. , , The matching score queues correspond to the trajectory end point, the first prediction point, the second prediction point, ..., the ... The matching score between the predicted point and the new point is calculated. According to the calculation principle of the matching score, the lower the score, the higher the matching degree between the new point and the corresponding reference point. Therefore, the reference point corresponding to the smallest score in the queue is selected as the final matching point. If the matching score queue is empty, it is determined that the new point fails to match the current trajectory.

[0098] The beneficial effects of this invention are as follows:

[0099] 1. This invention is reasonably designed. The raw point target data input by radar contains a large number of noise points, which severely interfere with the accuracy of trajectory identification and association. Therefore, it is necessary to first filter out noise from this raw point target data. This scheme uses an adaptive spatiotemporal clustering method to filter out noise points in the raw point target data. Unlike the traditional DBSCAN algorithm, which relies solely on spatial distance for clustering and has insufficient adaptability to spatiotemporal multidimensional data, this paper proposes a method that integrates multidimensional metrics and adaptive neighborhood adjustment.

[0100] 2. The adaptive spatiotemporal clustering method proposed in this invention addresses these problems in a targeted manner. It abandons the fixed neighborhood radius and dynamically adjusts the neighborhood radius based on the local point density. The neighborhood is expanded in sparse areas and shrunken in dense areas. At the same time, the time interval and height difference are incorporated into the clustering system and weighted. Noise points are filtered out by local density thresholds, which greatly improves the clustering accuracy and anti-interference ability.

[0101] 3. This invention proposes a multi-physics model matching and judgment approach, constructs multiple target motion physics models, and combines the target's historical motion state with the current track adaptability analysis. Even in complex scenarios such as changes in target motion mode and intersection of multiple trajectories, it can accurately identify track ownership, effectively solving the problems of insufficient robustness and poor adaptability of traditional methods, and significantly improving the continuity and accuracy of target tracking in complex scenarios. Attached Figure Description

[0102] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0103] Figure 1This is a flowchart of a radar target tracking method based on spatiotemporal clustering and multi-physics model according to the present invention.

[0104] Figure 2 This is a schematic diagram of denoising the original data.

[0105] Figure 3 This is a schematic diagram of the adaptive spatiotemporal clustering results.

[0106] Figure 4 It is a dot pattern image after noise removal.

[0107] Figure 5 This is a schematic diagram of the extrapolation points of the trajectory.

[0108] Figure 6 This is a schematic diagram of the target association results. Detailed Implementation

[0109] To make the technical problems, technical solutions, and technical effects of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0110] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0111] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0112] In the description of the embodiments of the present invention, it should be noted that the terms "inner", "outer", "upper", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the present invention.

[0113] Example 1

[0114] like Figure 1As shown, this embodiment provides a radar target tracking method based on spatiotemporal clustering and multi-physics models, including the following steps:

[0115] S1, Radar point target data input;

[0116] S2, Adaptive spatiotemporal clustering for noise removal, such as Figures 2 to 4 As shown, the specific steps are as follows:

[0117] S21. Calculate the spatiotemporal weighted distance between any two points:

[0118] Considering the fusion of spatial, altitude, and temporal multi-dimensional information of radar point targets, a metric is defined. and Spatiotemporal weighted distance between two points The calculation formula is as follows:

[0119] ;

[0120] In the formula, , Spatial location; and For height; and For time; These are spatial, vertical, and temporal weights, respectively.

[0121] S22. Calculate the global density:

[0122] For the entire radar point target dataset Define global density as Global Denseness This reflects that the distance between any two points in the dataset is less than The probability mean, This is the global reference radius;

[0123] S23. Calculate the local density:

[0124] For a certain point Its local density is defined as:

[0125] ;

[0126] Local density is accumulated and The inverse mean of the distances to neighboring points is used to quantify the density of a local region; a larger value indicates a denser local area. In the formula, for Count of neighboring points; To avoid constants with a denominator of zero;

[0127] S24. Calculate the adaptive neighborhood radius using the following formula:

[0128] ;

[0129] In the formula, To prevent the smoothing factor from being too heavy; To adjust the amplitude control parameters, by setting... The size (usually between 0.5 and 2) adjusts the adaptive adjustment strength of the neighborhood radius;

[0130] S25. Clustering decision based on adaptive neighborhood:

[0131] Clustering decision rules: two points and Satisfy space-time weighted distance At that time, belong The neighborhood; during the clustering process, a multi-category hierarchical judgment logic is adopted, covering core clusters, associated clusters and discrete points under different density distributions. At the same time, a cluster merging mechanism is applied, and the clustering results are finally processed. If the number of points in a cluster is less than 5 after clustering, it is judged as a noise point set and removed. This threshold can be adjusted according to the actual scenario.

[0132] S3. Establish a multiphysics model. The specific definitions of each physics model are as follows:

[0133] S31. Establish a time interval calculation model:

[0134] Assumption The scan time is , The scan time is ,but and Time difference:

[0135] ;

[0136] like Less than the preset time interval threshold If the condition is met, continue calculating the next physical quantity constraint; otherwise, end the current round of matching.

[0137] S32. Establish a spatial distance calculation model:

[0138] Assumption Location information is , Location information is ,So , The spatial distance calculation model is as follows:

[0139] ;

[0140] like Less than the preset distance threshold If the condition is met, continue calculating the next physical quantity constraint; otherwise, end the current round of matching.

[0141] S33. Establish a height difference calculation model:

[0142] Assumption The altitude is , The height of is ,but and Height difference:

[0143] ;

[0144] like Less than the preset time interval threshold If so, continue calculating the next physical quantity constraint.

[0145] S34. Establish a model for calculating the angle between motion vectors:

[0146] If the length of a trajectory is greater than or equal to 3, then the last three points of the trajectory are taken as follows:

[0147] , , ;

[0148] These three points can form two vectors. and ;

[0149] and and Angle between two vectors This is the vector angle at the latest match of the trajectory.

[0150] Earth's curvature correction radius: ;

[0151] Angle to radians: , ;

[0152] Formula for spatial rectangular coordinates: ;

[0153] Constructing motion vectors:

[0154] ;

[0155] ;

[0156] Calculate the magnitude of the motion vector:

[0157] ;

[0158] ;

[0159] Calculate the vector dot product:

[0160] ( ) ;

[0161] Calculate the included angle :

[0162] .

[0163] Specifically, based on the physical characteristics of the motion of various targets observed by radar, such as large aircraft, drones, and airborne objects, four major physical models are constructed. Each model calculates the corresponding physical quantity values ​​through formulas, and then combines them with preset thresholds to form a layer-by-layer constraint matching and judgment mechanism. These models quantify the continuity in the time dimension, the proximity in the spatial dimension, the stability in the vertical dimension, and the consistency of the motion direction, respectively. The calculated four key values—time difference, spatial distance, height difference, and motion vector angle—serve as the core judgment basis for radar point target trajectory point matching.

[0164] The specific steps for point-by-point and track-by-track adaptive threshold weight matching are as follows:

[0165] S35. Calculate the time matching score C1;

[0166] S36. Calculate the spatial distance matching score C2;

[0167] S37. Calculate the high-match score C3;

[0168] S38. Calculate the motion vector angle matching score C4;

[0169] S39. Accumulate the scores to obtain the match score S for this round.

[0170] The information is as follows: During the matching process, whenever the radar scans a new point, it needs to match the new point with all existing trajectories in the current situation one by one. If the match is successful, the new point is added to the end of the matched trajectory; if the match fails, the new point is initialized as the starting point of a new trajectory. During the matching process, each trajectory in the current situation needs to be matched with the new point, and the new point can be matched with multiple trajectories at the same time. Therefore, a matching score needs to be calculated when the new point is matched with each trajectory, and the trajectory corresponding to the best score is selected to match the new point.

[0171] For a trajectory, each new point is matched with the last point of the trajectory. During the matching process, the weights of the time interval, spatial distance, height difference, and vector angle are calculated.

[0172] Let the weight of the time interval be... The time matching score is:

[0173] ;

[0174] Let the spatial distance weight be... 2. The spatial distance matching score is:

[0175] ;

[0176] Let the height difference weight be... The high-match score is:

[0177] ;

[0178] Let the weight of the included angle of the vectors be... The matching score for the angle between the motion vectors is:

[0179] ;

[0180] Therefore, the total matching score is:

[0181] .

[0182] S4. Loop matching of predicted points, the specific method is as follows:

[0183] S41. Obtain the predicted points: P1, P2, ..., Pn through the extrapolation model;

[0184] S42. The predicted points return to step S41 to enter the physical constraint model to calculate the matching score.

[0185] S5. Integrate the matching scores S1, S2, ..., Sn in the matching score queue; the detailed solution is as follows: During radar scanning, environmental factors often lead to dropped or lost points; this directly causes the trajectory association to be interrupted. To solve this problem, it is necessary to extrapolate the trajectory points based on uniformly accelerated linear motion; the specific operation is as follows: when a new point completes a round of matching calculation with a trajectory, it is necessary to extrapolate the predicted points of the trajectory, and then use the extrapolated points as the end points of the trajectory to continue matching with the new point. The specific extrapolation method is as follows:

[0186] , , .

[0187] The two time intervals are calculated as follows:

[0188] , ;

[0189] The two velocity vectors are calculated as follows:

[0190] , ;

[0191] The acceleration vector is calculated as follows:

[0192] , ;

[0193] Calculate the extrapolation point, assuming the extrapolation point is located at... ;

[0194] ;

[0195] Therefore, the total number of matches between a new point and a trajectory is... Round 1, the first round matches the new point with the end point of the trajectory, the second round matches the new point with the first predicted point of the trajectory, ..., the ... The first round is for adding points and trajectories. The matching score is obtained by matching each predicted point. , The number of scores is determined by the number of extrapolated prediction points, such as Figure 5 As shown.

[0196] S6. Take the point corresponding to the best matching score as the result output, as follows:

[0197] Based on the previous section, for a new point, after experiencing... After round-matching, a queue of matching scores can be obtained. , , The matching score queues correspond to the trajectory end point, the first prediction point, the second prediction point, ..., the ... The matching score between the predicted point and the new point is calculated. According to the calculation principle of the matching score, the lower the score, the higher the matching degree between the new point and the corresponding reference point. Therefore, the reference point corresponding to the smallest score in the queue is selected as the final matching point. If the matching score queue is empty, it is determined that the new point fails to match the current trajectory. Figure 6 This is a schematic diagram of the results associated with the target.

[0198] Specifically, the raw point target data input by radar contains a large number of noise points, which severely interfere with the accuracy of trajectory identification and association. Therefore, it is necessary to first filter out noise from this raw point target data. This scheme uses an adaptive spatiotemporal clustering method to filter out noise points in the raw point target data. Unlike the traditional DBSCAN algorithm, which relies solely on spatial distance for clustering and has insufficient adaptability to spatiotemporal multidimensional data, this paper proposes a method that integrates multidimensional metrics and adaptive neighborhood adjustment.

Claims

1. A radar target tracking method based on spatiotemporal clustering and multi-physics model, characterized in that, Includes the following steps: S1, Radar point target data input; S2. Adaptive spatiotemporal clustering for noise removal, the specific steps are as follows: S21. Calculate the spatiotemporal weighted distance between any two points: Considering the fusion of spatial, altitude, and temporal multi-dimensional information of radar point targets, a metric is defined. and Spatiotemporal weighted distance between two points The calculation formula is as follows: ； In the formula, , Spatial location; and For height; and For time; These are spatial, vertical, and temporal weights, respectively. S22. Calculate the global density: For the entire radar point target dataset Define global density as Global Denseness This reflects that the distance between any two points in the dataset is less than The probability mean, This is the global reference radius; S23. Calculate the local density: For a certain point Its local density is defined as: ； Local density is accumulated and The inverse mean of the distances to neighboring points is used to quantify the density of a local region; a larger value indicates a denser local area. In the formula, for Count of neighboring points; To avoid constants with a denominator of zero; S24. Calculate the adaptive neighborhood radius using the following formula: ； In the formula, To prevent the smoothing factor from being too heavy; To adjust the amplitude control parameters, by setting... The size of the neighborhood radius is used to adaptively adjust the intensity. S25. Clustering decision based on adaptive neighborhood: Clustering decision rules: two points and Satisfy space-time weighted distance At that time, belong The neighborhood; during the clustering process, a multi-category hierarchical judgment logic is adopted, covering core clusters, associated clusters and discrete points under different density distributions. At the same time, a cluster merging mechanism is applied, and the clustering results are finally processed. If the number of points in a cluster is less than 5 after clustering, it is judged as a noise point set and removed. This threshold can be adjusted according to the actual scenario. S3. Establish a multi-physics model and perform adaptive threshold weight matching point-by-point and track-by-track: S31. Establish a time interval calculation model: Assumption The scan time is , The scan time is ,but and Time difference: ； like Less than the preset time interval threshold If the condition is met, continue calculating the next physical quantity constraint; otherwise, end the current round of matching. S32. Establish a spatial distance calculation model: Assumption Location information is , Location information is ,So , The spatial distance calculation model is as follows: ； like Less than the preset distance threshold If the condition is met, continue calculating the next physical quantity constraint; otherwise, end the current round of matching. S33. Establish a height difference calculation model: Assumption The altitude is , The height of is ,but and Height difference: ； like Less than the preset height difference threshold If so, continue calculating the next physical quantity constraint; S34. Establish a model for calculating the angle between motion vectors: If the length of a trajectory is greater than or equal to 3, then the last three points of the trajectory are taken as follows: ， ， ； These three points can form two vectors. and ; and and Angle between two vectors This is the vector angle at the time of the latest match of this trajectory; Earth's curvature correction radius: ; Angle to radians: , ; Formula for spatial rectangular coordinates: ; Constructing motion vectors: ； ； Calculate the magnitude of the motion vector: ； ； Calculate the vector dot product: ( ) ； Calculate the included angle : ； S35. Calculate the time matching score C1; S36. Calculate the spatial distance matching score C2; S37. Calculate the high-match score C3; S38. Calculate the motion vector angle matching score C4; S39. Accumulate the matching score S for this round; S4. Cyclic matching of prediction points; S5. Integrate the matching scores S1, S2, ..., Sn in the matching score queue; S6. Take the point corresponding to the best matching score as the result output; 2. The radar target tracking method based on spatiotemporal clustering and multi-physics model according to claim 1, characterized in that, In step S4, the specific method for cyclic matching of predicted points is as follows: S41. Obtain the predicted points: P1, P2, ..., Pn through the extrapolation model; S42. The predicted points return to step S41 to enter the physical constraint model to calculate the matching score.

3. The radar target tracking method based on spatiotemporal clustering and multi-physics model according to claim 2, characterized in that, In step S4, the adaptive threshold weight matching and cyclic matching are as follows: During the matching process, whenever the radar scans a new point, it needs to match the new point with all existing trajectories in the current situation one by one; if the match is successful, the new point is added to the end of the matching trajectory. If a match fails, the newly added point is initialized as the starting point of a new trajectory. During the match process, each trajectory in the current situation needs to be matched with the newly added point. The newly added point can be matched with multiple trajectories at the same time. Therefore, a matching score needs to be calculated when the newly added point is matched with each trajectory, and the trajectory corresponding to the best score is selected to match the newly added point. For a trajectory, each new point is matched with the last point of the trajectory. During the matching process, the weights of the time interval, spatial distance, height difference, and vector angle are calculated. Let the weight of the time interval be... The time matching score is: ； Let the spatial distance weight be...

2. The spatial distance matching score is: ； Let the height difference weight be... The high-match score is: ； Let the weight of the included angle of the vectors be... The matching score for the angle between the motion vectors is: ； Therefore, the total matching score is: 。 4. The radar target tracking method based on spatiotemporal clustering and multi-physics model according to claim 3, characterized in that, In step S4, during radar scanning, it is necessary to extrapolate trajectory points based on uniformly accelerated linear motion. Specifically, when a new point completes one round of matching calculation with a trajectory, it is necessary to extrapolate the predicted points for that trajectory, and then use the extrapolated points as the end points of that trajectory to continue matching with the new point. The specific extrapolation method is as follows: ， ， ; The two time intervals are calculated as follows: ， ； The two velocity vectors are calculated as follows: ， ； The acceleration vector is calculated as follows: ， Calculate the extrapolation point, assuming the extrapolation point is located at... ; ； Therefore, the total number of matches between a new point and a trajectory is... Round 1, the first round matches the new point with the end point of the trajectory, the second round matches the new point with the first predicted point of the trajectory, ..., the ... The first round is for adding points and trajectories. The matching score is obtained by matching each predicted point. , The number of scores is determined by the number of extrapolated prediction points.

5. The radar target tracking method based on spatiotemporal clustering and multi-physics model according to claim 4, characterized in that, In step S6, the point corresponding to the optimal matching score is taken as the output result, as follows: For a new point, after experiencing After round-matching, a queue of matching scores can be obtained. , , The matching score queues correspond to the trajectory end point, the first prediction point, the second prediction point, ..., the ... The matching score between the predicted point and the new point is calculated. According to the calculation principle of the matching score, the lower the score, the higher the matching degree between the new point and the corresponding reference point. Therefore, the reference point corresponding to the smallest score in the queue is selected as the final matching point. If the matching score queue is empty, it is determined that the new point fails to match the current trajectory.

Images