A method for joint tracking of a target of a UAV cluster

Abstract

The application discloses a kind of unmanned aerial vehicle cluster target joint tracking method, first, according to the three interactive rules of unmanned aerial vehicle cluster separation, aggregation and adjustment, the cooperative movement of unmanned aerial vehicle cluster is effectively described using stochastic differential equation;Second, based on the definition of new distance measure of target motion characteristics, the division of cluster in field of view is realized using DBSCAN clustering algorithm based on the measure;Finally, in the framework of Bayesian filtering, DBSCAN clustering algorithm is combined with JPDA algorithm, and the cluster division and target joint state estimation in clutter environment are realized simultaneously;In simulation experiment, the behavior of cluster is simulated and state estimation is carried out, and the results show that the algorithm can effectively describe the interactive motion process of unmanned aerial vehicle cluster target, and the tracking algorithm can effectively track the joint tracking of cluster target, which proves that the method has the characteristics of good tracking and classification effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle tracking, and particularly relates to a joint tracking method for unmanned aerial vehicle cluster targets. BACKGROUND

[0002] With the increasing variety of unmanned aerial vehicles and the continuous expansion of their task fields, unmanned aerial vehicles have gradually shifted from performing intelligence, surveillance and reconnaissance tasks in safe airspace to performing mainstream combat tasks in confrontational airspace. New types of unmanned aerial vehicles are constantly emerging, and the rapid development of autonomous control technology and artificial intelligence technology has given birth to a new combat mode represented by cluster combat. The United States has launched multiple programs since 2014 for developing various types of unmanned aerial vehicles that can be used for cluster combat.

[0003] To effectively respond to the aerial threat of unmanned aerial vehicle clusters, the primary task is to continuously detect and stably track the motion of unmanned clusters. However, due to the small distance between the members of an unmanned aerial vehicle cluster and the cooperative interaction between them, as well as the merging and splitting of the cluster, traditional multi-sensor multi-target tracking algorithms may have problems such as severe overlap of tracking gates, increased difficulty of data association, and difficulty in maintaining tracks. To address these issues, it is necessary to consider the cluster as a whole, jointly track the targets within the cluster, combine clustering algorithms with tracking algorithms, and jointly participate in the filtering and tracking process to achieve better tracking results.

[0004] Generally speaking, the concept of unmanned aerial vehicle cluster combat originates from the collective behavior of low-level social animals such as fish schools, bird flocks, and bee swarms in nature. The individuals in these animal groups can autonomously determine their motion states based on local perception and simple communication rules, and can emerge coordinated overall behavior from simple local rules. The coordination mode of unmanned aerial vehicle cluster combat is similar to this, and to accurately describe the cooperative interaction within the cluster, it is necessary to model the motion of the cluster. Reynolds first summarized three internal coordination rules of the cluster: separation, adjustment, and aggregation. Vicsek attributed the coordination effect to the averaging adjustment of target speed direction and proposed the Vicsek model. In addition, the interaction model based on the force between group members is also one of the directions of cluster motion modeling. The model abstracts the interaction between individuals as "force", and individuals complete maneuvering actions under the action of the resultant force generated by surrounding members.

[0005] After establishing the cluster motion model, another problem to be solved is the clustering of targets in the field of view. However, there is no corresponding algorithm in the prior art to effectively track and classify cluster targets, and therefore there is an urgent need to design a joint tracking method for unmanned aerial vehicle cluster targets to solve the above problems in the prior art. SUMMARY

[0006] To address the aforementioned problems, this invention aims to provide a joint tracking method for UAV swarm targets. This method combines the DBSCAN clustering algorithm, which eliminates the influence of noise points, with the JPDA algorithm. During the prediction process, it classifies the swarm targets, and during the update process, it achieves accurate correlation between the targets and measurements, ultimately completing accurate tracking of the swarm targets. It features good tracking and classification performance.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A joint tracking method for a swarm of unmanned aerial vehicle (UAV) targets, comprising the steps of

[0009] Step 1. Construct a drone swarm motion model

[0010] Step 101. Construct SDE motion models for cluster members within the drone swarm;

[0011] Step 102. Based on the SDE motion model of the cluster members established in Step 101, construct the SDE motion model of the cluster joint motion state;

[0012] Step 2. Establish a joint tracking model

[0013] Step 201. Based on the SDE motion model of the swarm joint motion state, the SDE motion model of the swarm joint motion state is discretized and combined with the joint measurement equation of the target to obtain the joint tracking system equation of the UAV swarm target, and a Bayesian filtering process is introduced to estimate the position of the members in the swarm.

[0014] Step 202. Introduce the DBSCAN clustering algorithm to cluster the cluster targets and calculate the predicted position value of any target in its respective cluster;

[0015] Step 203. Introduce the JPDA algorithm process to perform measurement association and filtering tracking on the cluster targets.

[0016] Preferably, the process of constructing the cluster member SDE motion model in step 101 includes:

[0017] Assuming there are drone clusters The member, the The dynamic model of each member is as follows:

[0018] (1)

[0019] In equation (1), and Representing the first One goal Position and velocity at any given moment represents a position control parameter, represents a velocity control parameter, represents the motion noise of an individual in the group, represents the motion noise of the entire group, , respectively represent the position center of the group and the mean velocity of all members of the group, respectively represent the position center of the group and the mean velocity of all members of the group, represents the target is subjected to the total potential field force of other targets in the group, the potential field force is the negative gradient of the potential field function , the potential field force can be defined according to the requirements of continuity, differentiability and non-negativity of the potential field function, for example:

[0020] (2)

[0021] In formula (2), is the Euclidean distance between member i and member j, R 11 , R 12 , R 21 , R 22 are potential field force control parameters.

[0022] Preferably, the construction process of the group joint motion state SDE motion model in step Step102 includes

[0023] Step1021. According to the SDE motion model of the members, the members belonging to the same group are combined, and the linear SDE representation of the group joint state in the two-dimensional space is as follows:

[0024] (3)

[0025] In formula (3)

[0026] (4)

[0027] The matrix is defined as

[0028] (5)

[0029] In formula (5)

[0030] (6)

[0031] The matrix is defined as

[0032] (7)

[0033] The joint noise from individual motion noise and group motion noise combined, the covariance matrix of which is

[0034] (8)

[0035] the coefficient matrix

[0036] the matrix is defined as

[0037] (9)

[0038] wherein in equation (9)

[0039] (10)

[0040] the matrix is defined as

[0041] (11)

[0042] Step 1022. Solve the cluster joint motion state SDE and discretize it, and the unmanned aerial vehicle cluster target joint tracking system equation can be obtained after being combined with the target joint measurement equation as:

[0043] (12)

[0044] in equation (12), , ;

[0045] is a zero-mean Gaussian noise, and the covariance

[0046] (13)

[0047] the matrix is defined as

[0048] (14)

[0049] wherein in equation (14)

[0050] (15)

[0051] the measurement noise is a Gaussian white noise, and the covariance matrix

[0052] (16).

[0053] Preferably, the process of introducing the Bayesian filtering process to estimate the position of the members in the group in Step 201 comprises

[0054] Step 2011. Estimating the position of the members in the group In a two-dimensional space, the distance between the members in the group is selected as The state of the members in the group at time t is The joint state of all the targets in the cluster in the field of view of the sensor is ;

[0055] Step 2012. Estimating the cooperative interaction within the group, the state of the group and the state of the targets are jointly estimated in the filtering process

[0056] Step 2013. After obtaining the state estimation value of the group at time t, the field of view of the cluster is first divided by a clustering algorithm, and then the state prediction value is obtained according to the interaction relationship between the target and other targets in the group, that is, the cluster state prediction formula at time t is

[0057] (17)

[0058] (18)

[0059] In formula (18), represents the number of cluster divisions in the field of view at time t, represents the transition probability obtained by the joint motion state equation

[0060] Step 2014. Updating the cluster state at time t The cluster state update formula at time t is as follows

[0061] (19)

[0062] Under the clutter environment, the key to tracking multiple targets is to solve the measurement likelihood Assuming that there are K associations between the target and the measurement, then

[0063] (20)

[0064] Preferably, the process of introducing the DBSCAN clustering algorithm to cluster the cluster targets in Step 202 comprises

[0065] Step 2021. According to the motion characteristics of the cluster center and the consistent speed of the unmanned aerial vehicle cluster, the cluster target motion characteristic distance measure is defined , ​​​​​A value between 0 and 1 is used to reflect the distance between the motion characteristics of the two target clusters, reflecting the possibility of the two targets belonging to the same cluster:

[0066] (21)

[0067] In formula (21), and are the relative Euclidean distance and relative velocity of the two targets, and are the distance-related standard deviation and velocity-related standard deviation, respectively; describes the constraint of the relative distance between the two targets;

[0068] Step 2022. Obtain cluster division using DBSCAN clustering algorithm After that, joint state prediction is performed within each cluster, and the final prediction value of all targets is the set of joint prediction values of each cluster target :

[0069] (22)

[0070] Step 2022. For the first cluster, the prediction value expression is as follows

[0071] (23)

[0072] Preferably, the specific process of introducing the JPDA algorithm flow to associate the group target described in step Step203 includes

[0073] Step 2031. In the JPDA algorithm, first, according to the geometric relationship between the multi-target tracking gates, it is divided into multiple clusters, and the targets and measurements in each cluster are processed in turn. Let the number of targets be , the number of measurements be , the JPDA introduces a confirmation matrix , which is expressed as follows

[0074] (24)

[0075] where, is a binary variable, indicates that the measurement j falls within the confirmation gate of target i, and

[0076] Step 2032. For the clutter model conforming to the Poisson distribution, the final association probability matrix of the effective measurements is obtained :

[0077] (25)

[0078] In formula (25), The jth effective measurement is associated with the target i, and the calculation formula is as follows

[0079] (26)

[0080] Step 2033. In the state updating process under the multi-target tracking filtering framework, the innovation of the measurement and the predicted value is calculated for any target i in the sensor field of view , the filtering gain , the state updating value , the state covariance updating value , and the expressions are as follows

[0081] (27)

[0082] The updating value of all targets is the set of the updating value of each target , that is

[0083] (28).

[0084] The beneficial effects of the present application are that the present application discloses a joint tracking method for unmanned aerial vehicle cluster targets, and compared with the prior art, the improvement of the present application is that:

[0085] The present application proposes a joint tracking method for unmanned aerial vehicle cluster targets based on the combination of the DBSCAN clustering algorithm and the JPDA algorithm, the internal interaction model of the unmanned aerial vehicle cluster members is established by using the random differential equation according to the interaction characteristics of the cluster targets, the interactive motion process of the unmanned aerial vehicle cluster is described, and the motion equation of the Bayesian filtering tracking is obtained after discretization; meanwhile, according to the characteristics of the cluster targets, a new motion characteristic distance measure is defined, and the DBSCAN clustering algorithm is used to realize the division of the cluster; the division method is combined with the JPDA tracking algorithm, the JPDA tracking algorithm of the cluster targets is derived under the Bayesian filtering framework, and the problem that the target measurement is difficult to be effectively associated is effectively solved; in the simulation experiment, the cluster behavior of the cluster is simulated and the state estimation is carried out, and the results show that the algorithm proposed in the present application can effectively describe the interactive motion process of the unmanned aerial vehicle cluster targets, the tracking algorithm can effectively carry out the joint tracking of the cluster targets, and it is proved that the present method has the advantages of good tracking and classification effects. BRIEF DESCRIPTION OF DRAWINGS

[0086] Figure 1 It is the algorithm flowchart of the joint tracking method for unmanned aerial vehicle cluster targets of the present application.

[0087] Figure 2 Schematic diagram of potential field region of the present application.

[0088] Figure 3 Flow chart of Bayesian filtering of the present application.

[0089] Figure 4 Schematic diagram of DBSCAN algorithm of the present application.

[0090] Figure 5 Trajectory estimation value graph of group scene of embodiment 2 of the present application.

[0091] Figure 6 Filtering and measurement error graph of group scene of embodiment 2 of the present application.

[0092] Figure 7 DBSCAN cluster division quantity graph of embodiment 2 of the present application.

[0093] Figure 8 Trajectory estimation value graph of subgroup scene of embodiment 2 of the present application.

[0094] Figure 9 Filtering and measurement error graph of subgroup scene of embodiment 2 of the present application.

[0095] Figure 10 DBSCAN cluster division quantity graph of embodiment 2 of the present application. DETAILED DESCRIPTION

[0096] In order for those skilled in the art to better understand the technical solutions of the present application, the technical solutions of the present application will be further described below in combination with the drawings and embodiments.

[0097] Embodiment 1: referring to the unmanned aerial vehicle cluster target joint tracking method shown in the accompanying drawings, comprising the steps of Figures 1-10

[0098] Step 1. Constructing an unmanned aerial vehicle cluster motion model

[0099] Step 101. In the unmanned aerial vehicle cluster, a cluster member SDE motion model is constructed;

[0100] Step 102. According to the cluster member SDE motion model established in step Step 101, a cluster joint motion state SDE motion model is constructed;

[0101] Step 2. Establishing a joint tracking model

[0102] ​Step 201. On the basis of the cluster joint motion state SDE motion model, the unmanned aerial vehicle cluster target joint tracking system equation is obtained by discretizing the cluster joint motion state SDE motion model and combining the target joint measurement equation, and the Bayesian filtering process is introduced to estimate the position of the group members;

[0103] Step 202. The DBSCAN clustering algorithm is introduced to cluster the cluster targets, and the position prediction value of any target is calculated under the respective cluster;

[0104] Step 203. The JPDA algorithm process is introduced to associate and filter track the cluster targets.

[0105] Preferably, the construction process of the cluster member SDE motion model in step Step 101 comprises:

[0106] The interaction between the cluster members can be abstracted as three rules of separation, aggregation and adjustment, so as to produce uniform and centralized motion and avoid collision between the members; the stochastic differential equation (SDE) can well represent the interaction between the members, assuming that there are members in the cluster, the dynamics model of the member is constructed as follows:

[0107] (1)

[0108] In formula (1), and respectively represent the position and velocity of the target at time , represents the position control parameter, represents the velocity control parameter, represents the motion noise of the individual in the group, represents the motion noise of the whole group, , respectively represent the position center of the group where the target is located and the average velocity of all members, represents the total potential field force on the target by other targets in the group, the potential field force is the negative gradient of the potential field function , and the potential field force can be defined as follows according to the requirements of continuity, differentiability and non-negativity of the potential field function:

[0109] (2)

[0110] In formula (2), is the Euclidean distance between member i and member j, R 11 is the Euclidean distance between member i and member j, R 12 is the Euclidean distance between member i and member j, R 21 is the Euclidean distance between member i and member j, R 22 is the potential field force control parameter, used to control the maximum value of repulsive force and attractive force, at this time the potential field area around the member can be formed as Figure 2 .

[0111] Preferably, the construction process of the cluster joint motion state SDE motion model in step Step102 includes:

[0112] Step1021. According to the SDE motion model of the members, the members belonging to the same cluster are combined, and the linear SDE representation of the cluster joint state in the two-dimensional space is as follows:

[0113] (3)

[0114] In formula (3)

[0115] (4)

[0116] The matrix is defined as

[0117] (5)

[0118] In formula (5)

[0119] (6)

[0120] The matrix is defined as

[0121] (7)

[0122] The joint noise is combined by the individual motion noise and the group motion noise , The covariance matrix of

[0123] (8)

[0124] The coefficient matrix

[0125] The matrix is defined as

[0126] (9)

[0127] In formula (9)

[0128] (10)

[0129] matrix , defined as

[0130] (11)

[0131] Step 1022. Solve the cluster joint motion state SDE, and discretize it, and the unmanned aerial vehicle cluster target joint tracking system equation can be obtained after being associated with the target joint measurement equation, as shown in equation (12)

[0132] (12)

[0133] In equation (12), , .

[0134] is a zero-mean Gaussian noise, and the covariance

[0135] (13)

[0136] matrix , defined as

[0137] (14)

[0138] wherein, in equation (14),

[0139] (15)

[0140] measurement noise is a Gaussian white noise, and the covariance matrix

[0141] (16)

[0142] By the method, the unmanned aerial vehicle cluster target joint tracking system equation is obtained, that is, the cluster joint motion state SDE motion model.

[0143] Preferably, the process of introducing the Bayesian filtering process to filter the members in the cluster in step Step201 includes:

[0144] Step 2011. Since the traditional filtering tracking method represented by Kalman filtering can be unified in the Bayesian filtering framework, in this embodiment, the Bayesian filtering framework for cluster target tracking is given; for the members in the cluster , in a two-dimensional space, the moment state is selected as , then the joint state of all cluster targets in the sensor field of view ;

[0145] Step2012. For accurately describing the cluster behavior, the interaction within the cluster is estimated, and the joint estimation of the cluster state and the target state is needed in the filtering process, for example, the cluster state is modeled as , and there are three targets in the cluster, one of the cluster states is , which means that the targets 1 and 2 belong to the cluster 1, and the target 3 belongs to the cluster 2; the joint tracking Bayesian filtering process of the UAV cluster target is shown in Figure 3

[0146] Step2013. The predicted value of the state at the time t is obtained , and then the estimated value of the state at the time t is obtained , and the predicted value of the cluster target state at the time t is obtained : firstly, the clusters in the field of view are divided by the clustering algorithm, and then the predicted value of the state is obtained according to the interaction between the target and other targets in the cluster, that is, the predicted value of the state at the time t is obtained , and the predicted formula of the cluster state at the time t is as follows

[0147] (17)

[0148] (18)

[0149] In the formula (18), represents the number of cluster divisions in the field of view at the time t, represents the transition probability obtained by the joint motion state equation;

[0150] Step2014. The updated formula of the cluster state at the time t is as follows

[0151] (19) The key to tracking multiple targets in the clutter environment is to solve the measurement likelihood

[0152] , assuming that there are kinds of association results of the target and the measurement, then

[0153] (20)

[0154] that is, the updated formula of the cluster state at the time t is obtained as shown in the formula (20). Preferably, the process of clustering the cluster target by introducing the DBSCAN clustering algorithm in the step Step202 includes:

[0155]

[0156] ​​​​​Because of the merging and splitting behavior of the cluster, the number of clusters at a certain moment is difficult to determine effectively, and the traditional k-means clustering method is difficult to effectively apply, so it is crucial to design a clustering method that can only rely on the distance relationship between targets, and the DBSCAN (Density-based spatial clustering of application with noise) clustering method meets the above requirements well;

[0157] The core idea of the DBSCAN algorithm is to cluster the high-density region of the target point data into a cluster, and the number of adjacent targets within the given radius of each target in the cluster must reach a certain threshold. The clustering result divides the spatio-temporal data with similarity in the target point set into the same group according to the similarity measurement standard, so that the similarity difference within the same group is as small as possible, and the difference between groups is as large as possible;

[0158] Figure 4 For the clustering example of the DBSCAN algorithm, the dashed circle radius is the clustering radius Eps, and MinPts=2 is set. Points 2, 3, 4, and 5 in the neighborhood with Eps as the clustering radius have at least MinPts points, so they are core points. Point 1 has only one point in its neighborhood, so it is a boundary point. Point 6 has no other points in its neighborhood, and point 7 is considered a noise point. Finally, {1, 2, 3, 4, 5} belong to the same cluster.

[0159] The flowchart of the DBSCAN algorithm is as follows:

[0160]

[0161] Step 2021. According to the motion characteristics of the unmanned aerial vehicle cluster center, speed consistency, and motion characteristics distance measure , Use a value between 0 and 1 to reflect the motion characteristics distance of two target clusters, that is, the possibility of two targets belonging to the same cluster.

[0162] (21)

[0163] In formula (21), and are the relative Euclidean distance and relative speed of two targets, and are the distance-related standard deviation and speed-related standard deviation; The relative distance between two targets is described, which is related to the expansion boundary of the cluster. When the relative distance and relative speed of two targets are smaller, the distance between them is closer and the motion similarity is better, and the possibility of belonging to the same cluster is greater;

[0164] Step2022. Obtain cluster partition by DBSCAN clustering algorithm After that, state joint prediction can be performed in each cluster, and the final prediction value of all targets is the set of joint prediction values of each cluster target

[0165] (22)

[0166] Step2022. For the first cluster, the prediction value expression is as follows

[0167] (23)

[0168] That is, the joint prediction value of any cluster target can be obtained by equation (22) and equation (23).

[0169] Preferably, the specific process of introducing the JPDA algorithm flow to associate the group targets in step Step203 comprises:

[0170] In order to solve the measurement likelihood in the multi-target case, the association between the measurement and the track needs to be performed, and the joint probability data association method has been widely concerned since its inception because of its good multi-target correlation performance.

[0171] Step2031. In the JPDA algorithm, first, according to the geometric relationship between the multi-target tracking gates, it is divided into multiple clusters, and the targets and measurements in each cluster are processed in turn, assuming that the number of targets is , and the number of measurements is In order to represent the complex relationship between the effective echo and the track gate, the JPDA introduces a confirmation matrix , which is represented as follows

[0172] (24)

[0173] Wherein, any matrix element is a binary variable, , indicating that the measurement j falls within the confirmation gate of target i, , indicating that it does not fall within; for the case where the measurement falls within the intersection region of the tracking gate, it means that the measurement may come from multiple targets; the purpose of JPDA is to calculate the probability of each measurement and its possible various source targets associated with each other, so after establishing the confirmation matrix, the confirmation matrix needs to be split into all interconnection matrices representing interconnection events ; the splitting process must be based on two assumptions: each measurement has a unique source; for a given target, at most one measurement takes it as a source.

[0174] ​​​Step2032. For the clutter model conforming to Poisson distribution, the correlation probability matrix of the effective measurement is finally obtained :

[0175] (25)

[0176] In formula (25), Pij represents the correlation probability of the jth effective measurement and the target i, and the calculation formula is as follows

[0177] (26)

[0178] Since the state updating process does not involve the interaction of the cluster and the calculation related to , the measurement data can be used after the correlation probability matrix is obtained.

[0179] Step2033. The state updating process is performed under the multi-target tracking filtering framework, and the innovation of the measurement and the predicted value is calculated for any target i in the field of view of the sensor , the filtering gain , the state updating value , the state covariance updating value , and the expressions are as follows

[0180] (27)

[0181] Finally, the updating value of all targets is the set of the updating value of each target , that is

[0182] (28)

[0183] Embodiment 2: Step3. To verify the effectiveness of the unmanned aerial vehicle cluster target joint tracking method described in embodiment 1, the following experiment is designed for verification.

[0184] Step301. Simulation scenario

[0185] To verify the rationality of the unmanned aerial vehicle cluster motion model constructed in embodiment 1 and the effectiveness of the tracking algorithm, two simulation scenarios of cluster motion and sub-cluster motion are set for verification; in the simulation, there are a total of 9 targets, which are divided into 3 groups;

[0186] The motion model and measurement model parameter settings are shown in Table 1:

[0187] Table 1: Tracking system equation parameter settings

[0188]

[0189] ​The DBSCAN algorithm parameter settings are shown in Table 2.

[0190] Table 2: DBSCAN algorithm parameter settings

[0191]

[0192] The JPDA algorithm parameter settings are shown in Table 3.

[0193] Table 3: JPDA algorithm parameter settings

[0194]

[0195] The sensor acquisition measurement period is , the total duration of the UAV cluster group motion , the total duration of the group motion , the initial covariance of the target ; at the initial moment, the launch center and launch speed of the target group motion and group motion are randomly located in a certain area of the plane, and the specific settings are as follows:

[0196] Table 4: Group motion initial condition settings

[0197]

[0198] Table 5: Group motion initial condition settings

[0199]

[0200] Tables 4 and 5 show that in the group scene, 3 clusters are launched according to the areas and speeds shown in , , cluster 1 and cluster 2 merge, and cluster 3 continues to move; , cluster 3 and , the merged cluster continues to merge, at which time the 3 clusters in the scene are completely merged; in the group scene, 1 cluster is launched according to the positions and speeds shown in , , cluster 3 suddenly slows down and separates from the original cluster; , cluster 2 suddenly speeds up and separates from the original cluster, at which time 3 groups appear in the scene.

[0201] Step 302. Result analysis

[0202] Step 3021. Grouping

[0203] The group scene simulation results are shown in Figure 5 , Figure 6 , Figure 7 , the motion trajectory estimation value of the cluster is shown in Figure 5 , which reflects the trajectory point of the target, and the figure is drawn according to interval plot, Figure 6 the filtering error and measurement error of the target are shown, Figure 7 reflects the result of target cluster division using the DBSCAN algorithm; as can be seen from the figure, after the target is launched, it first goes through a cluster self-organizing stage, cluster 1 and cluster 2 merge at , and after merging, it goes through a self-organizing stage, the speed, size and direction of the two clusters gradually become consistent, cluster 3 merges at t = 20 s; the DBSCAN algorithm accurately identifies the two merging times, but since the algorithm divides the clusters according to the distance and speed relationship between the targets, there is a phenomenon of partial advance or delay in identifying the actual interaction time, which will affect the cluster state prediction value, and at this time the filtering error will increase;

[0204] Step 3022. Cluster division

[0205] The cluster division simulation result is shown in Figure 8 , Figure 9 , Figure 10 The figure has similar meanings to the merging figure. As can be seen from the figure, after the target is launched, it first goes through a cluster self-organizing stage, the cluster divides at to produce cluster 3, to produce cluster 1 and cluster 2; the DBSCAN algorithm accurately identifies the two division times, and the stable identification times are and , both have a certain lag, the cause is similar to that when merging, at this time the filtering error will increase, but since the robustness of the tracking algorithm itself and the influence of the interaction between the targets due to the small distance between the targets is not great, the algorithm proposed in the present application can still achieve stable tracking.

[0206] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above examples, the above examples and descriptions in the specification are only to illustrate the principles of the present application, and various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1.A method for joint tracking of a UAV swarm target, the method comprising: comprising the steps of Step 1. Constructing the UAV swarm motion model Step 101. Constructing the swarm member SDE motion model in the UAV swarm; Step 102. Constructing the swarm joint motion state SDE motion model according to the swarm member SDE motion model established in Step 101; Step 2. Establishing the joint tracking model Step 201. Discretizing the swarm joint motion state SDE motion model on the basis of the swarm joint motion state SDE motion model, and obtaining the UAV swarm target joint tracking system equation by combining the discretized swarm joint motion state SDE motion model with the target joint measurement equation, and introducing the Bayesian filtering process to estimate the position of the swarm member; Step 202. Introducing the DBSCAN clustering algorithm to cluster the swarm targets, and calculating the position prediction value of any target under the respective swarm; Step 203. Introducing the JPDA algorithm process to associate and filter track the swarm targets; The construction process of the swarm member SDE motion model in Step 101 comprises Suppose that there are members in the UAV swarm, and the dynamic model of the th member is (1) In equation (1), and Representing the first One goal Position and velocity at any given moment Indicates position control parameters, Indicates speed control parameters, This represents the motion noise of individuals in the group. Represents the motion noise of the entire group. , Representing the target The location center of the group and the average speed of all members, Indicate target Subject to the total potential force of other targets within the group, potential force Let be the potential field function The negative gradient, according to the requirements of continuity, differentiability, and non-negativity of the potential field function, can be used to define the potential force as follows: (2) In formula (2), is the Euclidean distance between member i and member j, R 11 , R 12 , R 21 , R 22 is the potential field force control parameter; The construction process of the swarm joint motion state SDE motion model in Step 102 comprises Step 1021. According to the SDE motion model of the members, the SDE of the members belonging to the same swarm is combined to obtain the linear SDE representation of the swarm joint state in the two-dimensional space as follows: (3) In formula (3) (4) Matrix is defined as (5) In formula (5) (6) Matrix is defined as (7) Joint noise from individual motion noise and group motion noise combined, the covariance matrix is (8) coefficient matrix ; matrix , defined as (9) In formula (9) (10) matrix , defined as (11) Step 1022. Solving the swarm joint motion state SDE, and discretizing and combining the discretized swarm joint motion state SDE with the target joint measurement equation to obtain the UAV swarm target joint tracking system equation as follows: (12) In formula (12), , ; is zero-mean Gaussian noise with covariance (13) Matrix is defined as (14) In formula (14), (15) Measurement noise is a Gaussian white noise with covariance matrix (16)。 2.The method of claim 1, wherein: The process of estimating the position of the swarm member by introducing the Bayesian filtering process in Step 201 comprises Step 2011. The cluster member In two-dimensional space, select its momentary state is The joint state of all cluster targets within the sensor field of view ; Step 2012. Estimating the swarm internal cooperative interaction, and jointly estimating the swarm state and the target state in the filtering process; Step 2013. Prediction is obtained Time state estimate value After that, first of all, the field of view is divided into clusters by clustering algorithm, and then the state prediction value is obtained according to the interaction relationship between the target and other targets in the cluster to which the target belongs, that is Time cluster state prediction formula (17) (18) In formula (18), denotes the number of cluster divisions in the field of view at the moment, denotes the transition probability obtained by the joint motion state equation; Step 2014. Update The cluster state update formula at time t is as follows (19) The key to track multiple targets in cluttered environment is to solve the measurement likelihood Suppose the association results between targets and measurements have kinds, then (20)。 3.The method of claim 1, wherein: The process of clustering the swarm targets by introducing the DBSCAN clustering algorithm in Step 202 comprises Step 2021. According to the characteristics of the centralized and consistent speed movement of the UAV cluster center, the cluster target movement characteristic distance measure is defined , The value between 0 and 1 is used to reflect the movement characteristic distance of two target clusters, and the possibility that two targets belong to the same cluster: (21) In formula (21), and are the relative Euclidean distance and relative velocity of the two targets, respectively, and are the distance-related standard deviation and velocity-related standard deviation, respectively; The restriction of the relative distance of the two targets is described. Step 2022. Obtain cluster partition by using DBSCAN clustering algorithm After that, state joint prediction is performed within each cluster, and finally the prediction value of all targets is the set of joint prediction values of each cluster target ​ (22) Step 2022. For the first cluster, the predicted value expression is as follows (23)。 4.The method of claim 1, wherein: The specific process of associating the swarm targets by introducing the JPDA algorithm process in Step 203 comprises Step 2031. In the JPDA algorithm, firstly, according to the geometric relationship between the multi-target tracking gates, it is divided into multiple clusters, and the targets and measurements in each cluster are processed in turn. Let the number of targets be , the number of measurements be , the JPDA introduces a confirmation matrix , which is expressed as follows (24) wherein, is a binary variable, denotes that the measurement j falls within the acceptance gate of target i, denotes that it does not fall within; Step 2032. For the clutter model that follows a Poisson distribution, the correlation probability matrix of the effective measurements is finally obtained : (25) In formula (25), represents the association probability of the jth effective measurement and the target i, and the calculation formula is as follows (26) Step 2033. Perform a state update process under a multi-target tracking filter framework to compute the innovation of the measurement and the predicted value for any target i in the sensor field of view , filter gain , state update value , state covariance update value , as follows (27) the final all-targets the update value for each target the set of update values, i.e. (28)。

Images