Group target fine tracking method based on cooperative relation speed correction

Abstract

The invention discloses a group target fine tracking method based on cooperative relation speed correction, and the method comprises the steps: firstly obtaining a random finite set model on the premise of supposing to track a dense group target, and building an intra-group multi-target tracking model; and through a generalized label multi-Bernoulli filtering algorithm, forming and predicting a hypothetical track of a dense target in the group, a hypothetical track weight and a target probability density, and updating the hypothetical track and a target state according to the obtained group target measurement. On the basis, a group target fine tracking method for performing speed correction based on a cooperative relationship is provided, and the overall movement speed estimation of the group target is used for updating the individual hypothesis track of the dense targets in the group, so that the phenomenon of track association errors of the targets in the group is remarkably reduced. According to the method, the tracking accuracy of the dense targets in the group in the complex environment is effectively improved, and reliable technical support can be provided for fine reconnaissance and attack application of the group targets.

Description

Technical Field

[0001] This invention belongs to the field of radar signal processing technology, and in particular relates to a method for fine tracking of group targets based on cooperative relationship velocity correction. Background Technology

[0002] Group tracking can be viewed as a special type of multi-object tracking problem, which differs from traditional multi-object tracking problems. Traditional multi-object tracking problems typically assume that the motion of each object is independent. However, in group tracking problems, although the individual sub-objects in the group can exhibit independent motion to some extent, the group moves as a whole. The individual sub-objects need to move synchronously to avoid collisions, thereby ensuring that the group as a whole maintains a certain motion pattern.

[0003] Differences in tracking problems will lead to variations in the objects of focus, the utilization of prior information, and the technical approaches. In traditional multi-target tracking methods, the multi-target assumption and joint probabilistic data association algorithms can achieve the tracking of multiple individual targets while solving the multi-target association problem. However, when facing dense multi-target or group target problems, association errors can easily occur due to measurement errors, observation noise signals, and missed target detection.

[0004] In practical engineering applications, such as intercepting low-altitude formations penetrating targets and tracking targets of special value within a swarm, accurate navigation of individual targets within the swarm is crucial while tracking the entire swarm. Therefore, fine-grained swarm target tracking is necessary. Most existing fine-grained swarm target tracking methods retain the characteristics of multi-target tracking, primarily tracking swarm targets by adding control terms to the target motion model, without considering the impact of internal swarm cooperation relationships on the swarm structure. Therefore, designing an accurate and efficient fine-grained swarm target tracking method is of significant research importance. Summary of the Invention

[0005] The purpose of this invention is to address the problem of high track association error rates in the fine tracking of dense targets within a group of targets in existing technologies, and to provide a fine tracking method for group targets based on cooperative relationship velocity correction. This method can provide accurate and efficient tracking tracks for dense group targets. To achieve the above objective, this invention adopts the following technical solution.

[0006] This application provides a fine-grained tracking method for group targets based on cooperative relationship velocity correction, the method comprising:

[0007] Step 1: Obtain the random finite set model and establish a multi-target tracking model within the swarm;

[0008] Step 2: Using the generalized label multi-Bernoulli filtering algorithm, form and predict the hypothetical tracks of dense targets within the group, as well as the hypothetical track weights and target probability densities;

[0009] Step 3: Update the hypothetical track and target status based on the acquired group target measurements;

[0010] Step 4: Update the hypothetical trajectory for the group of targets using a velocity correction strategy.

[0011] Compared with the prior art, the significant advancement of this invention lies in the following: The significant advantages of this application compared with the prior art are:

[0012] (1) By introducing the cooperative relationship between individual targets within a group into the target tracking method, and considering the motion characteristics of the group targets, the accuracy of fine tracking is improved and the probability of incorrect correlation of dense target tracks within the group is reduced.

[0013] (2) The proposed method is applicable to targets moving in a straight line at a constant speed and targets moving in a turning direction at a constant speed.

[0014] (3) Based on achieving more accurate target tracking within the group, it can simultaneously and continuously track individual targets within the group as well as the overall group target. Attached Figure Description

[0015] Figure 1 A flowchart illustrating a fine-grained group target tracking method based on cooperative relationship speed correction provided in this application embodiment.

[0016] Figure 2 This is a schematic diagram of group formation and movement direction provided for an embodiment of this application.

[0017] Figure 3 The image shows the tracking of a uniformly moving target using the conventional generalized label multi-Bernoulli filtering tracking method in the embodiments of this application; where (a) is the overall track map and (b) is a partial track map.

[0018] Figure 4 The image shows the tracking diagram of a target moving at a constant speed using the group target fine tracking method based on cooperative relationship speed correction in this embodiment of the present application; where (a) is the overall track map and (b) is a partial track map.

[0019] Figure 5 The image shows the tracking diagram of a target moving at a constant speed while turning, using the traditional generalized label multi-Bernoulli filtering tracking method in this embodiment of the application; where (a) is the overall track map and (b) is a partial track map.

[0020] Figure 6 The image shows the tracking diagram of a target moving at a constant speed while turning, based on the group target fine tracking method with cooperative relationship speed correction in this embodiment of the present application; where (a) is the overall trajectory diagram and (b) is a partial trajectory diagram.

[0021] Figure 7 This is a comparison chart of OSPA distance errors for tracking a target moving at a constant speed in the embodiments of this application.

[0022] Figure 8 This is a comparison chart of OSPA distance errors for tracking a target moving at a constant speed while turning, as shown in the embodiments of this application. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] A fine-grained tracking method for group targets based on cooperative relationship velocity correction, characterized in that the method includes:

[0025] Step 1: Obtain the random finite set model and establish a multi-target tracking model within the swarm;

[0026] Step 2: Using the generalized label multi-Bernoulli filtering algorithm, form and predict the hypothetical tracks of dense targets within the group, as well as the hypothetical track weights and target probability densities;

[0027] Step 3: Update the hypothetical track and target status based on the acquired group target measurements;

[0028] Step 4: Update the hypothetical trajectory for the group of targets using a velocity correction strategy.

[0029] Furthermore, a random finite set model is obtained, and a multi-target tracking model within the swarm is established, including:

[0030] Suppose that at time k, the target state space is... There exists N k If there are several target states, then the random finite set of target states can be represented as:

[0031]

[0032] In the formula, Representing the target state space The set of all finite subsets of x, with element x i,k This represents the i-th target state at time k, and there is no corresponding relationship between the element order at time k and time k+1;

[0033] Similarly, the measurement set at time k is represented as:

[0034]

[0035] In the formula, Represents measurement space The set of all finite subsets of z, with element z i,k M represents the i-th measurement at time k. k This represents the number of measurements taken at time k.

[0036] The target state set X at time k k Including the target X at time k-1 k-1 The three parts—survival, proliferation, and the target newborn at time k—are represented as follows:

[0037]

[0038] In the formula, S k|k-1 (x k-1 Let p represent a random finite set of targets that survive at time k-1, and let p be the probability of the targets surviving. S,k (x k-1 The probability of death is 1-p. S,k (x k-1 );B k|k-1 (x k-1 Γ represents the random finite set derived from the target at time k-1. k Let k represent a random finite set of newly generated targets at time k;

[0039] Measurement set Z at time k k It consists of two parts: the measurement of target generation and the measurement of clutter generation, represented as:

[0040]

[0041] In the formula, Θ k (x k Let p be a random finite set of measurements generated by the targets, where each target has p. D,k (x k The probability of ) is detected by the sensor, and simultaneously expressed as a likelihood function g k (z k |x k Generate measurement z k Conversely, there is also 1-p D,k (x k The probability of a sensor missing a detection is 0. This represents a random finite set of measurements that indicate clutter generation.

[0042] Assuming the swarm as a whole moves independently and follows a Markov process, the motion model of the swarm as a whole can be represented as:

[0043] x c,k =Fx c,k-1 +wk (5)

[0044] In the formula, x c,k The vector represents the motion state of the entire group of targets at time k; F is the state transition matrix; w k It is Gaussian white noise;

[0045] When the movements of individual goals within a group are coordinated and the overall goal of the group is taken as the main objective, the movement model of individual goals within the group can be represented as follows:

[0046] x j,k =x c,k +DIS j +w k (6)

[0047] In the formula, x j,k DIS represents the motion state vector of the j-th individual target within the group at time k; j w represents the translation vector between the overall group objective and the objective of the j-th individual within the group; k It is Gaussian white noise.

[0048] The measurement model for individual goals within a group is expressed as follows:

[0049] z j,k =Hx j,k +v k (7)

[0050] In the formula, z j,k Represents the position of the j-th individual target in the group at time k as measured by the sensor; H is the measurement matrix; v k To measure covariance noise.

[0051] Furthermore, using a generalized label-based multi-Bernoulli filtering algorithm, hypothetical trajectories of dense targets within the swarm are formed and predicted, along with the hypothetical trajectory weights and target probability densities, including:

[0052] Suppose that the survival probability of an object with state (x, l) in the next time step is p. S (x,l), with a probability of extinction of q S (x,l)=1-p S (x,l), the newborn target follows a weighted order of w. B The density is p B The label (x,l) is a Bernoulli distribution, predicting the density π+(X). + ) is represented as:

[0053]

[0054] Where X + I is the state prediction quantity. + For predicting the label set; For the state label space, Provide a space for new student target tags. The label space is for prediction, where Ξ is a discrete space; a two-dimensional array. Represents the set of track prediction labels I + When there is an associated mapping process Assuming, Historical information is mapped to the correlation between flight track and measurement; the relevant parameters in equation (8) are calculated as follows:

[0055]

[0056] In the formula, The weights representing the predicted hypothetical trajectory are the survival tag weights. and the weight of the new student label w γ The product; for a given label, This represents the predicted single-target probability density, which can be categorized as either the surviving target density or the newly generated target density. The surviving target density is determined by the survival probability. With weighted probability transition density It was obtained through single-objective prediction.

[0057] Furthermore, the hypothetical track and target state are updated based on the acquired group target measurements, including:

[0058] The updated posterior distribution using measurements is:

[0059]

[0060] The relevant parameters in equation (15) are calculated as follows:

[0061]

[0062] In the formula, Let θ represent the weights of the posterior hypothesis trajectory. -1 ({0:|Z|}) is the mapping function from the label space to the measurement index, where Θ is its space. It means only needs to be considered Mapping; The pseudo-measurement likelihood function consists of two parts: detected and undetected. The detected part g(z|·,l) is a single-target measurement likelihood function. Let be the single-target normalization constant; for a given label, the single-target posterior probability density is... By predicting density and pseudo-measure likelihood function The result was obtained through calculation.

[0063] Furthermore, a velocity correction strategy is applied to the group of targets to update the hypothetical trajectory, including:

[0064] Constraining target states belonging to the same group, the average of all target states in the group is used to estimate the overall motion state of the group, i.e.:

[0065]

[0066] In the formula, x c v represents the positional component of the overall target of the group. c For the velocity components of the group as a whole, x j Let v be the position component of the j-th target within the group. j Let n be the velocity component of the j-th target in the group, and n be the number of targets in the group.

[0067] Based on the estimation of the overall motion state of the group, the velocity of all targets within the group is corrected using the following method:

[0068]

[0069] In the formula, To correct the velocity component of the j-th target within the group, S sen To adjust parameters and control the sensitivity of correction, w is a proportionality coefficient that describes the strength of velocity correction on targets within the swarm. It is used to preserve certain individual state changes when the swarm structure changes.

[0070] The method provided in this application is further illustrated below with reference to specific embodiments. This embodiment is described in the following context.

[0071] Tracking tests were conducted on dense groups of targets moving at constant speed in straight lines and turning at constant speeds. The traditional generalized label multi-Bernoulli filtering tracking method and the fine-grained group target tracking method based on cooperative relationship velocity correction proposed in this application were used to track and compare the group targets in this case. The OSPA distance was used as the performance evaluation index for both methods. This index considers three aspects: the number of targets, the location of targets, and the association of target identities. The smaller the value, the better the tracking performance.

[0072] In a two-dimensional coordinate system, the sensor monitors a spatial range of 2000m × 2000m, where λ = 2.5 × 10⁻⁶. -6 / m 2 P represents clutter intensity. S =0.99 is the target survival probability, P D =0.95 represents the detection probability. There is one cluster within the region, containing 25 targets. The initial coordinates of the cluster center target are... The distance between all targets within the group is 5 meters, and the total movement time is 100 seconds. The group formation and direction of movement are as follows: Figure 2 As shown.

[0073] For a target moving at a constant velocity in a straight line, the velocity is... The track data obtained for tracking targets within the group are as follows: Figure 3 , 4 As shown, the OSPA distance is as follows Figure 7 As shown.

[0074] For a target moving at a constant speed while turning, the speed is... The turning began at 20 seconds and ended at 60 seconds, with a turning angle of 3°. The resulting tracks for tracking targets within the group are as follows: Figure 5 , 6 As shown, the OSPA distance is as follows Figure 8 As shown.

[0075] Comparative analysis of flight path charts shows that the proposed fine-grained tracking method for swarm targets based on cooperative relationship velocity correction significantly reduces flight path intersections, making the fine-grained correlation paths of individual targets within a dense swarm more stable. Furthermore, the constraint of the swarm as a whole on individual targets greatly reduces the possibility of individual targets exceeding the swarm's target range due to correlation errors. Comparative analysis of OSPA distances demonstrates that the proposed fine-grained tracking method for swarm targets based on cooperative relationship velocity correction provides more accurate estimates of the swarm's target status and quantity.

[0076] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fine-grained tracking method for group targets based on cooperative relationship speed correction, characterized in that, The method includes: Step 1: Obtain the random finite set model and establish a multi-target tracking model within the swarm; Step 2: Using the generalized label multi-Bernoulli filtering algorithm, form and predict the hypothetical tracks of dense targets within the group, as well as the hypothetical track weights and target probability densities; Step 3: Update the hypothetical track and target status based on the acquired group target measurements; Step 4: Update the hypothetical trajectory for the group of targets using a velocity correction strategy.

2. The fine-grained group target tracking method based on cooperative relationship speed correction according to claim 1, characterized in that, Obtain a random finite set model and establish a multi-target tracking model within the swarm, including: Suppose that at time k, the target state space is... There exists N k If there are several target states, then the random finite set of target states can be represented as: In the formula, Representing the target state space The set of all finite subsets of x, with element x i,k This represents the i-th target state at time k, and there is no corresponding relationship between the element order at time k and time k+1; Similarly, the measurement set at time k is represented as: In the formula, Represents measurement space The set of all finite subsets of z, with element z i,k M represents the i-th measurement at time k. k This represents the number of measurements taken at time k. The target state set X at time k k Including the target X at time k-1 k-1 The three parts—survival, proliferation, and the target newborn at time k—are represented as follows: In the formula, S k|k-1 (x k-1 Let p represent a random finite set of targets that survive at time k-1, and let p be the probability of the targets surviving. S,k (x k-1 The probability of death is 1-p. S,k (x k-1 );B k|k-1 (x k-1 Γ represents the random finite set derived from the target at time k-1. k Let k represent a random finite set of newly generated targets at time k; Measurement set Z at time k k It consists of two parts: the measurement of target generation and the measurement of clutter generation, expressed as: In the formula, Θ k (x k Let p be a random finite set of measurements generated by the targets, where each target has p. D,k (x k The probability of ) is detected by the sensor, and simultaneously expressed as a likelihood function g k (z k |x k Generate measurement z k , and vice versa 1-p D,k (x k The probability of a sensor missing a detection is 0. This represents a random finite set of measurements that indicate clutter generation. Assuming the swarm as a whole moves independently and follows a Markov process, the motion model of the swarm as a whole can be represented as: x c,k =Fx c,k-1 +w k (5) In the formula, x c,k The vector represents the motion state of the entire group of targets at time k; F is the state transition matrix; w k It is Gaussian white noise; When the movements of individual goals within a group are coordinated and the overall goal of the group is taken as the main objective, the movement model of individual goals within the group can be represented as follows: x j,k =x c,k +DIS j +w k (6) In the formula, x j,k DIS represents the motion state vector of the j-th individual target within the group at time k; j w represents the translation vector between the overall group objective and the objective of the j-th individual within the group; k It is Gaussian white noise. The measurement model for individual goals within a group is expressed as follows: z j,k =Hx j,k +v k (7) In the formula, z j,k Represents the position of the j-th individual target in the group at time k as measured by the sensor; H is the measurement matrix; v k To measure covariance noise.

3. The fine-grained group target tracking method based on cooperative relationship speed correction according to claim 1, characterized in that, Using a generalized label-based multi-Bernoulli filtering algorithm, hypothetical trajectories of dense targets within a swarm are formed and predicted, along with the hypothetical trajectory weights and target probability density, including: Suppose that the survival probability of an object with state (x, l) in the next time step is p. S (x,l), with a probability of extinction of q S (x,l)=1-p S (x,l), the newborn target follows a weighted order of w. B The density is p B The label of (x,l) is a Bernoulli distribution, predicting the density π. + (X + ) is represented as: Where X + I is the state prediction quantity. + For predicting the label set; For the state label space, Provide a space for new student target tags. The label space is for prediction, where Ξ is a discrete space; a two-dimensional array. Represents the set of track prediction labels I + When there is an associated mapping process Assuming, Historical information is mapped to the correlation between flight track and measurement; the relevant parameters in equation (8) are calculated as follows: In the formula, The weights representing the predicted hypothetical trajectory are the survival tag weights. and the weight of the new student label w γ The product; for a given label, This represents the predicted single-target probability density, which can be categorized as either the surviving target density or the newly generated target density. The surviving target density is determined by the survival probability. With weighted probability transition density It was obtained through single-objective prediction.

4. The fine-grained group target tracking method based on cooperative relationship speed correction according to claim 1, characterized in that, The hypothetical track and target state are updated based on the acquired group target measurements, including: The updated posterior distribution using measurements is: The relevant parameters in equation (15) are calculated as follows: In the formula, Let θ represent the weights of the posterior hypothesis trajectory. -1 ({0:|Z|}) is the mapping function from the label space to the measurement index, where Θ is its space. It means only needs to be considered Mapping; The pseudo-measurement likelihood function consists of two parts: detected and undetected. The detected part g(z|·,l) is a single-target measurement likelihood function. Let be the single-target normalization constant; for a given label, the single-target posterior probability density is... By predicting density and pseudo-measure likelihood function The result was obtained through calculation.

5. The fine-grained group target tracking method based on cooperative relationship speed correction according to claim 1, characterized in that, For a group of targets, a velocity correction strategy is used to update the hypothetical trajectory, including: Constraining target states belonging to the same group, the average of all target states in the group is used to estimate the overall motion state of the group, i.e.: In the formula, x c v represents the positional component of the overall target of the group. c For the velocity components of the group as a whole, x j Let v be the position component of the j-th target within the group. j Let n be the velocity component of the j-th target in the group, and n be the number of targets in the group. Based on the estimation of the overall motion state of the group, the velocity of all targets within the group is corrected using the following method: In the formula, To correct the velocity component of the j-th target within the group, S sen To adjust parameters and control the sensitivity of correction, w is a proportionality coefficient that describes the strength of velocity correction on targets within the swarm. It is used to preserve certain individual state changes when the swarm structure changes.

Images