An unmanned aerial vehicle group target tracking method based on dynamic topology constraint

Abstract

The application provides a UAV group target tracking method based on dynamic topology constraint, and solves the technical problems that target association is unstable, identity is easily exchanged, occlusion is missed, and track is interrupted after missing detection, and it is difficult to realize continuous and stable tracking when existing UAV group target tracking. It comprises the following steps: acquiring observation data of the UAV group target, generating UAV candidate targets, and calculating corresponding motion state information; constructing a dynamic topology graph of the UAV group, extracting group structure features in combination with the motion state information, and determining the UAV group state according to the group structure features; constructing a target association cost function, adjusting the association parameters of the target association cost function according to the group state, associating the UAV candidate targets at the current moment with historical tracks, and obtaining a target association result; and updating the tracks of each UAV target in the UAV group according to the target association result, so as to realize continuous tracking of the UAV group target. The application can be widely applied in the technical field of multi-target tracking.

Description

Technical Field

[0001] This application belongs to the field of multi-target tracking technology, and more specifically, it relates to a method for tracking unmanned aerial vehicle (UAV) swarm targets based on dynamic topology constraints. Background Technology

[0002] In recent years, with the rapid development of unmanned aerial vehicle (UAV) technology, miniaturized, low-cost, and highly maneuverable UAVs have been widely used in many fields. UAV swarms, consisting of multiple UAVs, have gradually become an important focus in low-altitude airspace surveillance, security early warning, and target defense. UAV swarms generally possess characteristics such as large numbers, small targets, low flight altitude, high maneuverability, and significant group coordination, placing higher demands on traditional multi-target tracking methods. Therefore, swarm target tracking is a major research direction in contemporary multi-target tracking.

[0003] The main task of multi-target tracking is to estimate the motion states of multiple targets from continuous observation data while maintaining consistency in the identities of each target. Existing multi-target tracking methods are typically based on techniques such as Kalman filtering, particle filtering, joint probabilistic data association, multiple hypothesis tracking, and labeled random finite sets. These methods generally achieve good tracking results in typical multi-target tracking scenarios, but they still have certain limitations in UAV swarm target tracking scenarios.

[0004] First, the targets within a drone swarm are close together, and their speeds, headings, and maneuvering characteristics are similar. When multiple drone targets are densely distributed or intersect within the sensor's field of view, relying solely on the motion prediction results of a single target for data association can easily lead to problems such as target identity changes, trajectory jumps, or false associations.

[0005] Secondly, during actual flight, UAV swarms may exhibit various state changes, such as formation flight, dispersed search, convergence penetration, splitting and bypassing, and merging and withdrawing. Traditional multi-target tracking methods typically treat each target as an independent tracking object, lacking an expression of the internal spatial structure and topological relationships of the UAV swarm, and making it difficult to fully utilize the stability information of the overall motion and relative structure of the UAV swarm.

[0006] Furthermore, in complex low-altitude environments, UAV swarm targets are susceptible to ground obstruction, sensor noise, clutter interference, and false alarms. When one or more UAV targets are temporarily missing, traditional methods typically rely on single-target motion models to maintain the trajectory, failing to fully utilize group constraint information such as the swarm center motion, adjacent target relationships, and historical relative positions, resulting in insufficient trajectory continuity.

[0007] Furthermore, existing tracking methods often use fixed settings for correlation parameters, threshold parameters, or model weights, making it difficult to adapt to dynamic changes in the state of UAV swarms. For example, in stable formations, the swarm topology is relatively stable, and topology consistency constraints should be strengthened; while in dispersed states, the relative relationships between targets change significantly, and the weights of topology constraints should be appropriately reduced while enhancing the prediction of individual motions. Fixed parameter strategies struggle to balance tracking accuracy and stability across different swarm states.

[0008] Therefore, there is an urgent need to provide a new method for target tracking of drone swarms to solve the above problems. Summary of the Invention

[0009] The purpose of this application is to provide a target tracking method for UAV swarms based on dynamic topology constraints, so as to solve the technical problems in the prior art of UAV swarm target tracking, such as unstable target association, easy identity swapping, trajectory interruption after occlusion and missed detection, and difficulty in achieving continuous and stable tracking.

[0010] To achieve the above objectives, embodiments of this application provide a method for tracking unmanned aerial vehicle (UAV) swarm targets based on dynamic topology constraints, comprising the following steps: Acquire observation data of UAV swarm targets, generate UAV candidate targets and calculate corresponding motion state information; The edge weights are determined based on the candidate targets of the UAVs and a dynamic topology graph of the UAV swarm is constructed. The swarm structure features are extracted by combining motion state information, and the state of the UAV swarm is determined based on the swarm structure features. Construct a target association cost function, adjust the association parameters of the target association cost function according to the group state, associate the current UAV candidate target with the historical trajectory, and obtain the target association result; The trajectory of each drone target in the drone swarm is updated based on the target association results, enabling continuous tracking of drone swarm targets.

[0011] Preferably, the process of determining edge weights and constructing a dynamic topology graph of the UAV swarm based on UAV candidate targets includes: mapping each UAV candidate target to a node in the dynamic topology graph, mapping the relative positions and relative motion relationships between UAV candidate targets to edges between nodes; and determining whether edges can be established between nodes by combining preset distance adjacency conditions and nearest neighbor conditions. The edge weights between two nodes are calculated by combining the relative distance, relative speed difference, heading angle difference, and historical adjacency stability. A dynamic topology graph is then constructed by combining the node set, edge set, and edge weight set.

[0012] Preferably, the process of extracting group structure features by combining motion state information includes: calculating the group center position, group radius, average distance between targets, variance of distance between targets, heading consistency parameter, topology edge change rate and number of topology connected components based on the dynamic topology graph and the motion state information of UAV candidate targets, so as to obtain the group structure features of the UAV swarm; The group center position and group radius are used to characterize the overall spatial distribution of the UAV swarm, the average distance between targets and the variance of the distance between targets are used to characterize the change in the distance between targets within the UAV swarm, the heading consistency parameter is used to characterize the consistency of the UAV swarm's direction of motion, and the topological edge change rate and the number of topological connected components are used to characterize the evolution state of the UAV swarm's topology.

[0013] Preferably, the process of determining the state of the UAV swarm based on the swarm structure characteristics includes: calculating the change in swarm radius, the change in variance of distance between targets, the target density, the change in target density, and the change in the number of topological connected components based on the swarm structure characteristics at the current time and the previous time; comparing the swarm structure characteristics and the corresponding changes with a preset discrimination threshold according to a preset discrimination priority; and determining the current swarm state of the UAV swarm from the preset swarm state set. The preset group state set includes stable formation state, dispersed state, aggregated state, split state and merged state; the preset judgment priority is to judge the split state or merged state first, and then judge the stable formation state, dispersed state or aggregated state.

[0014] Preferably, the formula for the target association cost function is: ; In the formula, For the current moment t No. i The historical trajectory and the first j The target association cost between candidate drone targets. The cost of motion prediction matching, For the cost of topological consistency, The cost of matching observed features, These are all association weight parameters adjusted based on the group state. This is when the observed feature matching cost is not used. .

[0015] Preferably, the process of adjusting the association parameters of the target association cost function according to the group state includes: when the group state is a stable formation state, increasing the weight parameter corresponding to the topology consistency cost relative to the baseline weight; when the group state is a dispersed state, decreasing the weight parameter corresponding to the topology consistency cost relative to the baseline weight and increasing the weight parameter corresponding to the motion prediction matching cost; when the group state is an aggregated state, relaxing the target association threshold and using the predicted position determined by the predicted position of the group center at the current time and the historical relative position to participate in the calculation of the motion prediction matching cost; when the group state is a split state, performing subgroup target association based on the topology connected components; when the group state is a merged state, reconstructing the dynamic topology graph of the merged group and performing target association based on the reconstructed dynamic topology graph.

[0016] Preferably, the process of obtaining the target association result includes: based on the group state at the current moment, using a state adjustment function to normalize the weight parameters corresponding to the motion prediction matching cost, topology consistency cost, and observation feature matching cost, so that the sum of each weight parameter is 1; using an allocation algorithm to associate the UAV candidate target at the current moment with the historical trajectory to obtain the target association result.

[0017] Preferably, the process of updating the trajectory of each UAV target in the UAV swarm includes: when the candidate UAV target in the target association result is successfully associated with the historical trajectory, the target status and trajectory are updated through a filter, and the UAV swarm target trajectory is output; When a candidate drone target fails to be associated with a historical trajectory, a compensation prediction position is generated based on the predicted position of the group center and the historical relative position of the target with respect to the group center. The trajectory is maintained and the drone swarm target trajectory is output.

[0018] Preferably, during the continuous tracking of the UAV swarm target, the changes in the connected components of the dynamic topology graph are monitored in real time; if the splitting of the UAV swarm target is detected, the UAV swarm is divided into multiple subgroups and tracked separately; if the center distance and movement direction differences of multiple subgroups meet the preset merging conditions, the subgroups are merged and the dynamic topology graph is reconstructed.

[0019] Preferably, the formula for compensating for the predicted location is: ; In the formula, For the first i A historical trajectory at the present moment t The compensation prediction location, For the current moment t Predicted location of the group center for t The location of the group center at time -1. For the firsti A historical trajectory at every moment t -1 target position, This is the overall correction matrix for the group.

[0020] The beneficial effects of this application are as follows: This application proposes a target tracking method for UAV swarms based on dynamic topology constraints. First, it acquires observation data of UAV swarm targets, generates candidate UAV targets, and calculates their corresponding motion state information, providing reliable data support for subsequent construction of the dynamic topology graph and extraction of swarm structure features. Next, it uses the candidate UAV targets as nodes in the dynamic topology graph, constructs edges based on the relative positional and motion relationships between targets, sets edge judgment conditions, determines edge weights, and constructs the UAV swarm dynamic topology graph. This explicitly expresses the structural coupling relationships within the UAV swarm, reducing mismatches and identity swaps among neighboring UAVs and improving the tracking stability of the UAV swarm. Furthermore, it combines the UAV swarm dynamic topology graph with... This paper extracts group structure features from motion state information, determines the state of the UAV swarm based on these features, and enables the discrimination of stable formation, dispersion, aggregation, splitting, and merging states of the UAV swarm. Next, a target association cost function is constructed, and its association parameters are adaptively adjusted according to the group state, allowing the tracking strategy to dynamically adjust with changes in the UAV swarm state. Then, the association matching of candidate UAV targets at the current moment with historical trajectories is completed to obtain the target association result. This allows for dynamic adaptation of the association strategy based on different motion evolution states of the UAV swarm, utilizing group topological constraints to assist in target matching and discrimination. Finally, the trajectories of each UAV target in the swarm are updated based on the target association result, achieving continuous tracking of the UAV swarm targets. This application introduces topological consistency constraints into the target association process to ensure the consistency of the identity of each UAV target. Simultaneously, it combines the group center and historical relative positions to compensate for missed targets and predict and maintain trajectories. This approach can adapt to various complex flight conditions such as stable formation, dispersion, aggregation, splitting, and merging of UAV swarms, improving the overall stability and reliability of UAV swarm tracking. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating the overall algorithm of a UAV swarm target tracking method based on dynamic topology constraints, provided in an embodiment of this application. Figure 2 A three-dimensional tracking trajectory diagram of a drone swarm obtained in this application is provided as an embodiment of this application; Figure 3 A three-dimensional tracking trajectory map of an unmanned aerial vehicle swarm obtained by a conventional method according to an embodiment of this application; Figure 4 A comparison diagram of the average position error between the conventional method provided in one embodiment of this application and that of this application; Figure 5 A dynamic topology diagram of an unmanned aerial vehicle (UAV) swarm at the initial simulation moment provided in an embodiment of this application; Figure 6 This application provides a dynamic topology diagram of the UAV swarm splitting state at an intermediate moment in a simulation, as shown in one embodiment of the present application. Figure 7 The above is a dynamic topology diagram of the drone swarm merging state at a later stage of the simulation, provided as an embodiment of this application. Detailed Implementation

[0023] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0024] This application proposes a target tracking method for UAV swarms based on dynamic topology constraints. First, candidate UAV targets are generated using sensor observation data, and the motion state information of each candidate UAV target is extracted. Second, a dynamic topology map is constructed based on the relative positional and motion relationships between the candidate UAV targets. Third, swarm structure features are extracted based on the dynamic topology map, and the current swarm state is determined according to these features. Then, the association parameters in the target association cost function are adjusted based on the swarm state. Finally, the candidate UAV targets at the current moment are associated with historical trajectories, and the trajectories of each UAV target in the swarm are updated based on the association results. Through these steps, this application can maintain good target association stability and trajectory continuity in scenarios such as dense UAV swarm flight, splitting, merging, short-term missed detection, and clutter interference.

[0025] Please see Figure 1 An embodiment of this application provides a method for tracking unmanned aerial vehicle (UAV) swarm targets based on dynamic topology constraints, comprising: S1: Acquire observation data of UAV swarm targets, generate UAV candidate targets, and calculate the corresponding motion state information.

[0026] Acquire observational data of the UAV swarm targets at the current moment. This data can be obtained from radar, electro-optical, infrared, radio detection equipment, or a multi-sensor fusion system. The observational data includes target position, target velocity, heading angle, target confidence level, and radar scattering characteristics.

[0027] Each candidate UAV target is generated based on each valid observation data, and the motion state information of each candidate UAV target is extracted and calculated.

[0028] At the present moment t A set of candidate targets for the UAV is generated based on sensor observation data, using the following formula: ; In the formula, For the current moment t The set of candidate drone targets, For the first i One candidate drone target, This represents the number of candidate drone targets at the current moment.

[0029] For the i The system identifies candidate UAV targets and obtains their motion state information. In an optional embodiment, the motion state information includes the target position, target velocity, target heading angle, and target confidence level. The target position and target heading angle can be obtained directly from observation data; the target velocity can be obtained directly from observation data or estimated from the target position at consecutive time points; the target confidence level can be obtained directly from observation data or set to a preset constant if the target confidence level is not obtained.

[0030] No. i The candidate drone targets are given by the following formula: ; In the formula, Let the target position vector be... For the target velocity vector, For the target heading angle, The target confidence level.

[0031] In a 3D UAV swarm target tracking scenario, the target position vector can be represented as: ; In the formula, They represent the first i The horizontal, vertical, and vertical positions of each candidate UAV target in three-dimensional space.

[0032] The target velocity vector can be represented as:

[0033] In the formula, These are the velocity components in the corresponding directions.

[0034] When the target velocity cannot be directly obtained, the target velocity vector can be estimated by the position difference between adjacent time steps, as shown in the following formula: ; In the formula, This represents the sampling time interval.

[0035] S2: Determine edge weights based on candidate drone targets and construct a dynamic topology graph of the drone swarm. Extract swarm structure features by combining motion state information and determine the drone swarm state based on the swarm structure features.

[0036] Specifically, this application no longer treats each target of the UAV swarm as an independent tracking object, but maps the UAV candidate targets as nodes in a dynamic topology graph, maps the relative positional and relative motion relationships between targets as edges between nodes, forms an edge between any two UAV candidate targets that satisfy the adjacency relationship, and constructs the dynamic topology graph of the UAV swarm by combining the edge weights.

[0037] At the present moment t The dynamic topology graph is represented as: ; In the formula, For the current moment t Dynamic topology graph, For a set of nodes, Let be the set of edges. Let be the set of edge weights. The set of nodes is denoted by . Corresponding to the current set of candidate drones, that is: ; In the formula, Indicates the first i The topology nodes corresponding to each candidate drone target.

[0038] edge set The adjacency condition is determined based on the adjacency conditions between candidate drone targets. In one optional embodiment, the adjacency conditions include distance adjacency conditions and nearest neighbor conditions. When the relative distance between two candidate drone targets is less than or equal to a preset adjacency distance threshold, or when two candidate drone targets are each other's nearest neighbors, an edge is established between the corresponding two nodes.

[0039] The distance adjacency condition is: when the first... i The candidate drone targets and the first j When the relative distance between candidate drone targets meets the preset adjacency distance threshold, an edge is established between the corresponding two nodes: ; In the formula, For a moment t The relative distance between two drone candidate targets This is a preset adjacency distance threshold.

[0040] The nearest neighbor condition is: In another alternative embodiment, the first... i The candidate drone targets and the first j When two candidate drone targets are each other's nearest neighbors, an edge is also established between the corresponding two nodes. By combining distance adjacency and nearest neighbor adjacency, excessive fragmentation of the topology graph can be avoided when the group is sparse or locally dispersed, while also avoiding the generation of too many invalid edges when the targets are dense.

[0041] Furthermore, the edge weights in the dynamic topology graph are determined. The edge weights in the dynamic topology graph are determined based on the relative distance, relative speed difference, heading angle difference, and historical adjacency stability between corresponding UAV candidate targets. By incorporating spatial distance, motion consistency, heading consistency, and historical adjacency relationships into the edge weight calculation, the dynamic topology graph not only reflects the spatial structure of the UAV swarm at the current moment but also reflects the historical stable relationships between targets within the swarm, thereby enhancing the ability to maintain identities in dense target scenarios.

[0042] In an optional embodiment, the edge weight calculation formula is: ; in: ; ; In the formula, For the current moment t No. i The candidate drone targets and the first j Boundary weights between candidate drone targets The relative distance between the two. The difference in relative speed between the two. The difference in heading angle between the two. For the historical adjacency stability of the two, For normalization parameters, The weight coefficients are non-negative and satisfy the following conditions: ; In one alternative embodiment, historical adjacency stability This can be determined based on whether two targets have maintained an adjacency relationship over a consecutive number of time points in the past. For example, in a length of... L Within the historical time window, if the first i The first goal and the first j One goal is among them M If an adjacency relationship exists at any given time, then the historical adjacency stability can be expressed as: ; In the formula, LThe length of the historical time window. M This represents the number of moments within which two targets maintain an adjacency relationship. By introducing historical adjacency stability, short-term noise or brief proximity between different UAV swarms can prevent immediate changes to the swarm topology, thus improving the stability of the UAV swarm structure representation.

[0043] In another alternative embodiment, if the current time is the initial stage of tracking and historical adjacency information is insufficient, then it can be set to... The initial values ​​can be preset, or the edge weights can be determined solely based on the relative distance, the difference in relative speed, and the difference in heading angle.

[0044] Furthermore, after constructing the dynamic topology graph, swarm structure features are extracted based on the dynamic topology graph and the motion state information of the UAV candidate targets. These swarm structure features include the swarm center position, swarm radius, average distance between targets, variance of distance between targets, heading consistency parameter, topological edge change rate, and number of topologically connected components. Specifically, the swarm center position and swarm radius characterize the overall spatial distribution of the UAV swarm; the average distance between targets and the variance of distance between targets characterize the variation in distance between targets within the UAV swarm; the heading consistency parameter characterizes the degree of consistency in the motion direction of the UAV swarm; and the topological edge change rate and the number of topologically connected components characterize the evolutionary state of the UAV swarm's topology.

[0045] In an optional embodiment, the group center location is calculated using the following formula: ; In the formula, For the current moment t The center of the drone swarm, For the first i The center weight of each drone candidate target is calculated. For the first i The position vectors of the candidate drone targets. For the current moment t The number of candidate drone targets.

[0046] The group radius is calculated using the following formula: ; In the formula, For the current moment t The center of the drone swarm, For the first i The center weight of each drone candidate target is calculated. For the first i The position vectors of the candidate drone targets. For the current moment t The group radius.

[0047] When the observation data includes target confidence scores, the weight for central calculation is taken as the target confidence score of the corresponding UAV candidate target: .

[0048] When the observed data does not contain the target confidence level, the center calculation weight is taken as a preset constant: .

[0049] By weighting the data, the impact of low-confidence observations on the estimation of the swarm center can be reduced, making the location of the swarm center more reflective of the true movement trend of the UAV swarm.

[0050] In an optional embodiment, the average distance between targets is calculated using the following formula: ; In the formula, For the current moment t The average distance between targets For the first i The candidate drone targets and the first j The relative distance between the candidate drone targets For the current moment t The number of candidate drone targets.

[0051] The variance of the distance between targets is calculated using the following formula: ; In the formula, For a moment t The average distance between targets For a moment t The variance of the distance between targets For the first i The candidate drone targets and the first j The relative distance between the candidate drone targets For a moment t The number of candidate drone targets.

[0052] In an optional embodiment, the heading consistency parameter is calculated using the following formula: ; In the formula, For the current moment t The heading consistency parameters, For the first i The velocity vectors of the candidate UAV targets To prevent constants with a denominator of zero, when the motion directions of all targets in a drone swarm tend to be consistent, Larger; when the movement directions of targets within the group are dispersed, Smaller.

[0053] In an optional embodiment, the rate of change of topological edges is calculated using the following formula: ; In the formula, For the current moment t The rate of change of the topological edges, To make the current moment t The set of labeled edges obtained by mapping nodes in a dynamic topology graph according to their historical trajectory labels. For the previous moment t The set of labeled edges with a value of -1. This is a symmetric difference operation for sets.

[0054] The number of topologically connected components is the number of non-connected subgraphs in the current dynamic topology graph, denoted as . When a drone swarm splits from a single entity into multiple subswarms, the number of topologically connected components typically increases; when multiple subswarms regroup and merge, the number of topologically connected components decreases.

[0055] Furthermore, the current group state of the UAV swarm is determined based on the aforementioned group structure characteristics. The group state is determined from a preset group state set, which includes stable formation state, dispersed state, aggregated state, split state, and merged state.

[0056] Before determining the group state, calculate the changes in group radius, the changes in variance of distance between targets, the changes in target density, and the changes in the number of topologically connected components.

[0057] The change in population radius is: ; In the formula, For the current moment t The change in the population radius, For the current moment t The group radius, For the previous moment t -1 is the group radius.

[0058] The change in variance of the distance between targets is: ; In the formula, This represents the change in the variance of the distance between targets. For the current moment t The variance of the distance between targets For the previous moment t -1 is the variance of the distance between targets.

[0059] The target density is: ; In the formula, For the current moment t Target density, For the current moment t The number of candidate drone targets, For the current moment t The group radius, It is a very small positive number.

[0060] The change in target density is: ; In the formula, The target density change For the current moment t Target density, For the previous moment t -1 target density.

[0061] The change in the number of topologically connected components is: ; In the formula, This represents the change in the number of topologically connected components. For the current moment t The number of topologically connected components, For the previous moment t -1 is the number of topologically connected components.

[0062] Furthermore, when the number of topologically connected components in the dynamic topology graph increases, the UAV swarm is determined to be in a split state; when the number of topologically connected components in the dynamic topology graph decreases, the UAV swarm is determined to be in a merging state; when the number of topologically connected components remains unchanged, the group radius changes little, the target distance variance is small, and the heading consistency parameter is high, the UAV swarm is determined to be in a stable formation state; when the number of topologically connected components remains unchanged, the group radius increases, and the target distance variance increases, the UAV swarm is determined to be in a dispersed state; when the number of topologically connected components remains unchanged, the group radius decreases, and the target density increases, the UAV swarm is determined to be in a convergent state.

[0063] In an optional embodiment, to avoid state conflicts when multiple group discrimination conditions are met simultaneously, the group state is determined according to a preset discrimination priority. The preset discrimination priority is: first determine the split state or the merge state, then determine the stable formation state, the dispersed state, or the aggregated state.

[0064] when At that time, it was determined that the drone swarm was in a split state; when At that time, it was determined that the drone swarm was in a merged state; when And satisfy At that time, the drone swarm was determined to be in a stable formation state; when And satisfy At that time, it was determined that the drone swarm was in a dispersed state; when And satisfy At that time, it was determined that the drone swarm was in a state of aggregation; in, All of these are preset threshold values. These threshold values ​​can be preset based on specific application scenarios, sensor measurement accuracy, drone swarm size, and flight speed range, or they can be automatically adjusted based on historical tracking data.

[0065] Through the above-described group state discrimination process, this application can identify typical states such as formation maintenance, diffusion, contraction, splitting and merging based on the dynamic changes in the topology of the UAV swarm, providing a basis for the adaptive adjustment of subsequent target association parameters.

[0066] S3: Construct a target association cost function, adjust the association parameters of the target association cost function according to the group state, associate the current UAV candidate target with the historical trajectory, and obtain the target association result.

[0067] After obtaining the set of candidate drone targets at the current moment, it is necessary to associate the current candidate drone targets with the historical trajectories formed at the previous moment or in historical moments. To improve the association stability in dense drone swarm scenarios, this application constructs a target association cost function that includes motion prediction matching cost and topology consistency cost. When the observation data contains observation features that can be used to distinguish targets, the target association cost function can also include observation feature matching cost.

[0068] In an optional embodiment, the target association cost function is expressed as: ; In the formula, For the current moment t No. i The historical trajectory and the first j The target association cost between candidate drone targets. The cost of motion prediction matching, For the cost of topological consistency, The cost of matching observed features, All are association weight parameters adjusted according to the group state. When the observed feature matching cost is not used, let .

[0069] In one alternative embodiment, the motion prediction matching cost can be determined based on the deviation between the predicted position of the historical trajectory and the observed position of the current UAV candidate target. Specifically, the motion prediction matching cost... It can be represented as: ; In the formula, For the first j The candidate drone targets at time t The observation location For the first i A historical trajectory at the present moment t The predicted location, This represents the corresponding prediction error covariance matrix. The aforementioned cost can be understood as the motion matching cost based on Mahalanobis distance; the closer the current observation position is to the historical trajectory prediction position, the smaller the motion prediction matching cost.

[0070] In one optional embodiment, the topology consistency cost is determined based on the difference between the adjacency relationships of targets corresponding to historical trajectories in the historical topology graph and the adjacency relationships of current UAV candidate targets in the current dynamic topology graph. Topology consistency cost It can be represented as: ; In the formula, For the first i Each historical trajectory corresponds to the set of adjacent trajectories of the target at the previous moment. For the previous moment i A historical trajectory and its adjacent trajectories r The relative topological relationship vector between them For the current moment j Candidate drone targets and adjacent trajectories r The relative topological relationship vector between predicted or associated locations. For adjacent trajectories r Reliability weight.

[0071] In one alternative embodiment, the relative topological relationship may include relative distance, relative orientation, and relative velocity difference.

[0072] ; In the formula, Indicates the first i The first goal and the first r The relative distance between the targets Indicates relative location. This indicates the difference in relative speed.

[0073] In another alternative embodiment, when the observation data includes radar scattering features, radio signal features, or other features that distinguish targets, the observation feature matching cost can be determined based on the difference between the observation features of the current UAV candidate target and the target features maintained by the historical trajectory. For example, if the target feature vector is... Then the cost of observing feature matching It can be represented as: ; In the formula, For the first j The observed feature vectors of each UAV candidate target. For the first i Historical characteristics of targets corresponding to historical trajectories are estimated.

[0074] Furthermore, this application adjusts the association weight parameters in the target association cost function according to the current group state of the UAV swarm, so that the tracking strategy can adapt to different motion states of the UAV swarm.

[0075] In an optional embodiment, adjusting the association weight parameters of the target association cost function according to the group state includes: When the group state is a stable formation state, the weight parameter corresponding to the increase in topology consistency cost relative to the baseline weight is... In a stable formation, the relative positions and adjacency relationships between targets are relatively stable. Therefore, topological consistency has high reference value for maintaining target identity.

[0076] When the group state is dispersed, the weight parameter corresponding to the reduction of the topology consistency cost relative to the baseline weight is... And increase the weight parameters corresponding to the motion prediction matching cost. In a dispersed state, the relative topological structure between targets varies considerably. Overemphasizing topological consistency may lead to incorrect constraints, thus requiring enhanced individual motion prediction capabilities.

[0077] When the group is in an aggregated state, the target association threshold is relaxed, and the predicted position determined by the current predicted position of the group center and its historical relative position is used in the calculation of motion prediction matching cost. In the aggregated state, the distance between targets decreases, observations are more likely to interfere with each other, and the distinguishability of individual motion predictions decreases. Therefore, using the predicted position jointly determined by the group center's motion trend and its historical relative structure in motion matching can assist in target association by maintaining individual identity discrimination and utilizing the overall motion consistency of the group.

[0078] When the group is in a split state, the subgroup target association is performed based on the topological connected components. That is, the dynamic topological graph is first divided into multiple connected components, and then the target association is performed within each connected component to reduce the probability of erroneous association between different subgroups.

[0079] When the group is in a merged state, a dynamic topology graph is reconstructed for the merged group, and target associations are performed based on the reconstructed dynamic topology graph. By reconstructing the topology graph, adjacency relationships can be updated when multiple subgroups are brought closer together again, avoiding association errors caused by continuing to use the old subgroup topology.

[0080] In an alternative embodiment, the association weight function can be determined by normalization using a state adjustment function: ; ; ; In the formula, For the current moment t The group state The state adjustment function is the cost of motion prediction matching. The state adjustment function represents the cost of topological consistency. The state adjustment function is the cost of matching observed features.

[0081] Through the above normalization process, the weights can be made to satisfy: ; This avoids unbounded growth of the weight parameters in the association cost function and improves the stability of the association parameter adjustment process.

[0082] After adjusting the association weight parameters in the target association cost function, the Hungarian algorithm, nearest neighbor algorithm, or other allocation algorithms can be used to determine the association relationship between the current UAV candidate target and the historical trajectory. The target association result can be expressed as: ; In the formula, For the current moment t The target association results For the first i A historical trajectory, For the first j One candidate drone target.

[0083] S4: Update the trajectory of each UAV target in the UAV swarm based on the target association results to achieve continuous tracking of UAV swarm targets.

[0084] After obtaining the target association results, the corresponding historical trajectories are updated based on the observation information of the current UAV candidate targets. For historical trajectories successfully associated with the target association results, the target state can be updated according to filter update rules. For example, particle filtering, Kalman filtering, and other state estimation methods can be used for trajectory updating.

[0085] In an alternative embodiment, if the first i The historical trajectory and the first j If multiple drone candidate targets are successfully associated, the updated target location can be represented as: ; In the formula, For the first i A historical trajectory at every moment t The predicted location, For the corresponding observation location, To update the gain for the state, For the first j The candidate drone targets at time t The predicted location.

[0086] For historical trajectories that have not been successfully associated with UAV candidate targets, this application does not immediately delete the trajectory, but generates a compensation prediction position based on the predicted position of the group center and the historical relative position of the target with respect to the group center in order to maintain the trajectory.

[0087] In an optional embodiment, the compensated predicted location is calculated using the following formula: ; In the formula, For the first i A historical trajectory at the present moment t The compensation prediction location, For the current moment t Predicted location of the group center for t The location of the group center at time -1. For the first i A historical trajectory at every moment t -1 target position, This is the overall correction matrix for the group.

[0088] Through the above-mentioned methods of compensation for missed detections, even if a drone target is obscured or not detected by sensors for a short period of time, its trajectory can be maintained by utilizing the overall trend of the drone and its historical relative structure, thereby reducing the probability of trajectory interruption.

[0089] Furthermore, during target tracking by a UAV swarm, the dynamic topology graph may change from one connected component to multiple connected components, or from multiple connected components to fewer connected components. To accommodate this, this application performs sub-swarm management in both split and merge states.

[0090] When the dynamic topology graph changes from one topologically connected component to multiple topologically connected components, it is determined that the UAV swarm has split. At this point, the UAV swarm is divided into multiple subgroups, and dynamic topology graph construction, swarm structure feature extraction, target association, and trajectory update are performed on each subgroup. By performing target association within the subgroups, erroneous associations caused by spatial proximity or overlapping movements between different subgroups can be avoided.

[0091] When the center distance between multiple topologically connected components is less than a preset distance threshold, and the difference in their motion directions is less than a preset direction threshold, it is determined that multiple subgroups meet the merging adjustment requirement. At this point, the multiple topologically connected components are merged into the same UAV swarm, and the merged dynamic topology map is reconstructed.

[0092] In one alternative embodiment, the two subgroups and The center distance can be expressed as: ; The difference in the direction of motion between the two subgroups can be expressed as: ; In the formula, and Subgroups and At any moment t The central location, and These are the average velocity vectors of the two subgroups, respectively.

[0093] Two subgroups are considered to satisfy the merging condition if the following equation is met: ; In the formula, To preset the merging distance threshold, This is the preset merging direction threshold.

[0094] Through the aforementioned subgroup management mechanism, this application can maintain the continuity of trajectory management when the UAV swarm splits, merges, and reorganizes, and reduce the probability of target identity exchange between different subgroups.

[0095] Detailed Implementation: Simulation Example To verify the effectiveness of the method in this application, this embodiment constructs a three-dimensional UAV swarm target tracking simulation scenario. The number of UAV targets is set to 15, the simulation time step is 160, and the sampling interval is 1 second. The UAV targets move in three-dimensional space, with the X and Y coordinates mainly distributed in the range of 1000m to 2000m, and the altitude Z coordinate mainly distributed around 300m. To simulate actual detection conditions, random measurement noise is added to the target observation data, and a certain proportion of missed detections and a small number of clutter points are set. Specifically, the standard deviation of the observation noise is 7m, the missed detection rate is set to 0.06, and several random clutter points are added in each time step, thereby forming a UAV swarm observation sequence containing noise, missed detections, and clutter interference.

[0096] This embodiment sets up a drone swarm target undergoing a group structure evolution process. In the initial stage, 15 drones form two adjacent but non-overlapping swarm targets, with drones within the same swarm having similar directions of motion, speeds, and relative formation structures. In the intermediate stage, one of the swarm targets splits into a new sub-swarm target, with each sub-swarm maintaining its own similar motion patterns. In the later stage, one of the split sub-swarms gradually approaches and merges with the other original swarm target to form a new swarm target, with the merged drone ensemble again exhibiting similar group motion patterns.

[0097] During the tracking process, the method in this application constructs a dynamic topology graph based on the motion state of the UAV target and uses intra-group adjacency relationships and relative structural stability to constrain target association and trajectory updates. In contrast, the traditional nearest neighbor method is used, which mainly associates targets based on the spatial distance between the predicted and observed positions and does not introduce group topology constraints.

[0098] The simulation results include a 3D trajectory diagram, a comparison diagram of average position error, and a topology diagram at typical moments. Figure 2 The three-dimensional tracking trajectory of the UAV swarm obtained in this application is provided by Figure 2 It is evident that this application maintains good trajectory continuity and intra-group motion consistency during target splitting, close-range parallelism, and merging.

[0099] Figure 3 The three-dimensional tracking trajectory of the UAV swarm obtained by the traditional method is more susceptible to the influence of neighboring targets, missed detections and clutter when multiple group targets are close together or undergo structural changes, resulting in some trajectory deviations.

[0100] Figure 4 The figure shows a comparison curve of the average position error between the present application and the traditional method. Simulation results show that, over 160 time steps, the average position error of the present application is approximately 12.47 m, while the average position error of the traditional method is approximately 91.08 m, indicating that the present application can effectively reduce the tracking error of group targets.

[0101] Figures 5 to 7 The dynamic topology at different time steps is shown respectively. Figure 5 Given the initial group objective state, G1 and G2 in the diagram represent the two groups at the current time. Figure 6 Corresponding to the target splitting state of the group, the newly added G3 in the figure represents the new group that splits out at the current moment. Figure 7 Corresponding to the state after subgroup merging, G4 in the diagram represents the new group formed by merging G2 and G3. In the diagram, circular nodes represent UAV targets, star nodes are the topological centers of the UAV swarm, and connecting lines represent target pairs that satisfy the adjacency condition. Different subgroups are distinguished by topological connectivity. It is evident that this application can dynamically update the topological relationships as the swarm target structure changes, thereby adapting to the evolution of the UAV swarm.

[0102] Through the above embodiments, the present application can achieve the following technical effects: This application constructs a dynamic topology graph, which incorporates the relative positional relationships, relative motion relationships, and historical adjacency stability of targets in a UAV swarm into the tracking process, thereby achieving an explicit expression of the internal structural relationships of the UAV swarm.

[0103] This application can identify typical group states such as stable formation, dispersion, aggregation, splitting and merging by extracting group structure features such as group center position, group radius, average distance between targets, variance of distance between targets, heading consistency parameter, topological edge change rate and number of topological connected components.

[0104] This application adjusts the association weight parameters in the target association cost function according to the group state, enabling the tracking method to select a more suitable association strategy under different group states. For example, it enhances topology consistency constraints in stable formation states, reduces topology consistency constraints and enhances motion prediction in dispersed states, performs intra-subgroup association in split states, and reconstructs the dynamic topology graph in merged states.

[0105] In cases of short-term target miss or occlusion, this application utilizes the predicted position of the group center and the historical relative position of the target with respect to the group center to generate a compensation predicted position, thereby improving trajectory continuity and reducing the probability of trajectory interruption.

[0106] Through the above steps, this application can solve the problems of unstable target association, easy identity exchange, easy trajectory interruption under occlusion and missed detection, and difficulty of fixed tracking parameters adapting to changes in group state in existing UAV swarm target tracking, and has good engineering application value.

[0107] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0108] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for tracking unmanned aerial vehicle (UAV) swarm targets based on dynamic topology constraints, characterized in that, Includes the following steps: Acquire observation data of UAV swarm targets, generate UAV candidate targets and calculate corresponding motion state information; Based on the candidate drone targets, edge weights are determined and a dynamic topology graph of the drone swarm is constructed. The swarm structure features are extracted by combining the motion state information, and the drone swarm state is determined based on the swarm structure features. Construct a target association cost function, adjust the association parameters of the target association cost function according to the group state, associate the current UAV candidate target with the historical trajectory, and obtain the target association result; The trajectory of each drone target in the drone swarm is updated based on the target association results to achieve continuous tracking of the drone swarm targets.

2. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 1, characterized in that, The process of determining edge weights and constructing a dynamic topology graph of the UAV swarm based on the UAV candidate targets includes: mapping each UAV candidate target to a node in the dynamic topology graph, mapping the relative positions and relative motion relationships between the UAV candidate targets to edges between nodes; and determining whether edges can be established between nodes by combining preset distance adjacency conditions and nearest neighbor conditions. The edge weights between the two nodes are calculated by combining the relative distance, relative speed difference, heading angle difference, and historical adjacency stability. The dynamic topology graph is then constructed by combining the node set, edge set, and edge weight set.

3. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 1, characterized in that, The process of extracting group structure features by combining the motion state information includes: calculating the group center position, group radius, average distance between targets, variance of distance between targets, heading consistency parameter, topology edge change rate and number of topology connected components based on the dynamic topology graph and the motion state information of the UAV candidate targets, so as to obtain the group structure features of the UAV swarm; The group center position and group radius are used to characterize the overall spatial distribution of the UAV swarm. The average distance between targets and the variance of the distance between targets are used to characterize the change in the distance between targets within the UAV swarm. The heading consistency parameter is used to characterize the consistency of the UAV swarm's movement direction. The topology edge change rate and the number of topology connected components are used to characterize the evolution state of the UAV swarm's topology.

4. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 1 or 3, characterized in that, The process of determining the state of the UAV swarm based on the swarm structure characteristics includes: calculating the change in swarm radius, the change in variance of distance between targets, the target density, the change in target density, and the change in the number of topological connected components based on the swarm structure characteristics at the current time and the previous time; comparing the swarm structure characteristics and the corresponding changes with a preset discrimination threshold according to a preset discrimination priority; and determining the current swarm state of the UAV swarm from the preset swarm state set. The preset group state set includes stable formation state, dispersed state, aggregated state, split state and merged state; the preset discrimination priority first judges the split state or merged state, and then judges the stable formation state, dispersed state or aggregated state.

5. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 4, characterized in that, The formula for the target association cost function is: ； In the formula, For the current moment t No. i The historical trajectory and the first j The target association cost between candidate drone targets. The cost of motion prediction matching, For the cost of topological consistency, The cost of matching observed features, These are all association weight parameters adjusted based on the group state. This is when the observed feature matching cost is not used. .

6. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 5, characterized in that, The process of adjusting the association parameters of the target association cost function according to the group state includes: when the group state is a stable formation state, increasing the weight parameter corresponding to the topology consistency cost relative to the baseline weight; when the group state is a dispersed state, decreasing the weight parameter corresponding to the topology consistency cost relative to the baseline weight and increasing the weight parameter corresponding to the motion prediction matching cost; when the group state is an aggregated state, relaxing the target association threshold and using the predicted position determined by the predicted position of the group center at the current time and the historical relative position to participate in the calculation of the motion prediction matching cost; when the group state is a split state, performing subgroup target association based on the topology connected components; when the group state is a merged state, reconstructing the dynamic topology graph for the merged group and performing target association based on the reconstructed dynamic topology graph.

7. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 5, characterized in that, The process of obtaining the target association result includes: based on the group state at the current moment, using a state adjustment function to normalize the weight parameters corresponding to the motion prediction matching cost, topology consistency cost, and observation feature matching cost, so that the sum of each weight parameter is 1; using an allocation algorithm to associate the UAV candidate target at the current moment with the historical trajectory to obtain the target association result.

8. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 1, characterized in that, The process of updating the trajectory of each drone target in the drone swarm includes: when a drone candidate target in the target association result is successfully associated with the historical trajectory, the target state and trajectory are updated through a filter, and the drone swarm target trajectory is output; When the candidate drone target fails to be associated with the historical trajectory, a compensation prediction position is generated based on the predicted position of the group center and the historical relative position of the target with respect to the group center. The trajectory is maintained and the drone swarm target trajectory is output.

9. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 1, characterized in that, During the continuous tracking of the UAV swarm target, the changes in the connected components of the dynamic topology graph are monitored in real time; if the UAV swarm target split is detected, the UAV swarm is divided into multiple subgroups and tracked separately; if the center distance and movement direction differences of multiple subgroups meet the preset merging conditions, the subgroups are merged and the dynamic topology graph is reconstructed.

10. The UAV swarm target tracking method based on dynamic topology constraints as described in claim 8, characterized in that, The formula for the compensation prediction position is: ； In the formula, For the first i A historical trajectory at the present moment t The compensation prediction location, For the current moment t Predicted location of the group center for t The location of the group center at time -1. For the first i A historical trajectory at every moment t -1 target position, This is the overall correction matrix for the group.

Images