A clutter-free density requirement target tracking method

Abstract

The application provides a target tracking method without clutter density demand, comprising: obtaining predicted targets and predicted potential new-born targets under different models at a current moment based on a target and a potential new-born target at a previous moment; calculating the association probability and Mahalanobis distance between the measurement data and the predicted targets under different models; selecting the Mahalanobis distance corresponding to the model with the maximum association probability to establish a cost matrix and determine an updated target; calculating the association probability and Mahalanobis distance between the unused measurement data and the predicted potential new-born targets under different models; selecting the Mahalanobis distance corresponding to the model with the maximum association probability to establish a cost matrix and determine a new-born target; screening the target with a probability greater than a given threshold from the updated target as a surviving target, and merging the surviving target with the new-born target to form a target at the current moment; and obtaining a potential new-born target at the current moment by using the measurement data unused at the previous two moments. According to the application, the target tracking does not need the clutter density, and the calculation complexity is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multi-target tracking, and particularly relates to a target tracking method without clutter density requirement. BACKGROUND

[0002] In the presence of data association uncertainty, missed detection and clutter environment, the adaptive delta-generalized labeled multi-Bernoulli (AGLMB) filter can effectively track multiple maneuvering targets. However, the AGLMB filter still needs prior parameters such as clutter density and survival probability, and has high computational complexity and large amount of calculation. How to establish a multiple maneuvering target tracking filter without prior parameters such as clutter density and survival probability and with fast calculation speed is a problem that needs to be explored and solved at present. SUMMARY

[0003] To solve the problem of target tracking requiring prior parameters such as clutter density and survival probability in the related art and reduce the computational complexity, the present application provides a target tracking method without clutter density requirement.

[0004] The method comprises the following steps:

[0005] Step A: based on each target and each potential new target at the previous moment, predicted targets and predicted potential new targets under different models at the current moment are obtained respectively;

[0006] Step B: the association probability, Mahalanobis distance between the measurement data and the predicted targets under different models, and the updated mean vector and the updated covariance matrix of the predicted targets are calculated;

[0007] Step C: the Mahalanobis distance corresponding to the model with the maximum association probability in step B is selected to establish a cost matrix of two-dimensional assignment, and the optimal solution is obtained by using the Murty algorithm;

[0008] Step D: based on the optimal solution obtained in step C, the updated target at the current moment is determined;

[0009] Step E: the association probability, Mahalanobis distance between the unused measurement data and the predicted potential new targets under different models, and the updated mean vector and the covariance matrix of the predicted potential new targets are calculated;

[0010] Step F: the Mahalanobis distance corresponding to the model with the maximum association probability in step E is selected to establish a cost matrix of two-dimensional assignment, and the optimal solution is obtained by using the Murty algorithm;

[0011] Step G: based on the optimal solution obtained in step F, the new target at the current moment is determined;

[0012] Step H: filtering out the targets with existence probability greater than a given threshold from the updated targets as surviving targets, and merging the surviving targets and the new-born targets to form the targets at the current time, and taking the targets at the current time as the input of the filter at the next time;

[0013] Step I: obtaining the potential new-born targets at the current time by using the least square method with the measurement data unused at the previous two times, and taking the potential new-born targets at the current time as the input of the filter at the next time.

[0014] In the related art, when target tracking is performed, the calculation complexity is high, the calculation amount is large, and prior parameters such as clutter density and survival probability are needed. Through the above method, the Mahalanobis distance between the predicted target and the measurement data, and the Mahalanobis distance between the potential new-born target and the unused measurement data are constructed into cost matrices respectively, avoiding the prior parameters such as clutter density and survival probability, and reducing the calculation complexity and calculation amount in the target tracking process. At the same time, the targets at the current time are determined by the measurement data, the predicted target and the potential new-born target, and the potential new-born target at the current time is determined by the measurement data unused at the previous two times, which fully uses the measurement data in target tracking, and makes the target tracking more accurate.

[0015] In an optional implementation, the target at the previous time and the potential new-born target are respectively represented as:

[0016]

[0017]

[0018] wherein k-1 represents the previous time, k represents the current time, and represent the existence probability, the track label, the model label of the target at the k-1 time and the number of targets at the k-1 time respectively; ε i,k-1 、 and represent the measurement index associated with the potential new-born target i at the k-1 time, the existence probability, the track label, the model label of the potential new-born target and the number of potential new-born targets at the k-1 time respectively; and represent the mean vector and the covariance matrix of the target at the k-1 time, represents the mean vector of the potential new-born target i at the k-2 time, and represent the mean vector and the covariance matrix of the potential new-born target at the k-1 time;

[0019] The predicted target at the current time under different models calculated is represented as:

[0020]

[0021] where t i,k = 1,2, L, M r denotes the model label, M r denotes the total number of models used, denotes the predicted existence probability of the predicted target in model t i,k and denote the predicted trajectory label and predicted model label, respectively, and denote the predicted mean vector and covariance matrix of the predicted target in model t i,k ; and have:

[0022]

[0023]

[0024] where p(t i,k | t i,k-1 ) is the Markov one-step transition probability, Φ k-1 (t i,k ) is the state transition matrix corresponding to model t i,k , and Q k-1 (t i,k ) is the covariance in model t i,k .

[0025] The predicted potential new-born target under different models at the current time is calculated as:

[0026]

[0027] where denotes the predicted existence probability of the predicted potential new-born target in model t i,k , and denote the predicted trajectory label and predicted model label, respectively, and denote the predicted mean vector and covariance of the predicted potential new-born target in model t i,k ; and have:

[0028]

[0029]

[0030] The predicted measurement vector of the predicted target under model t i,k and its error covariance matrix are:​

[0031]

[0032] The predicted measurement vector of a potential new target at model t i,k and its error covariance matrix are:

[0033]

[0034] where H k is the observation matrix.

[0035] In an alternative embodiment, the association probability between the measurement data and the predicted target at different models is represented as:

[0036]

[0037] where, z k represents the measurement data at time k, M j,k represents the number of measurement data, z D represents the jth measurement data; p k is the detection probability, R i,k is the observation noise covariance;

[0038] The Mahalanobis distance between the measurement data z j,k and the predicted target at model t j,k is represented as:

[0039]

[0040] and, obeys a chi-square distribution with n degrees of freedom, where n is the dimension of the observation data z α ; given a confidence level a, a reception threshold q j,k is obtained, if it is determined that the measurement data z i,k falls into the reception gate of the predicted target i;

[0041] The updated mean vector and covariance matrix of the predicted target i at model t

[0042]

[0043]

[0044] where,

[0045] In an alternative embodiment, the predicted target i is associated with the measurement data z j,k ​The model with the maximum association probability is expressed as:

[0046]

[0047] The Mahalanobis distance corresponding to the model with the maximum association probability is:

[0048]

[0049] The cost matrix of two-dimensional allocation is established using the Mahalanobis distance as the cost, and is expressed as:

[0050] C = [C1 C2]

[0051] Wherein:

[0052]

[0053] The optimal solution of the cost matrix is obtained using the optimal Murty algorithm, and is:

[0054] In an alternative embodiment, if 1≤θ i ≤M k , it indicates that the measurement data z j,k is associated with the prediction target i, and j = θ i is taken, the mean vector and the covariance matrix corresponding to the model with the maximum association probability are selected as the updated mean vector and the updated covariance matrix of the prediction target i at the current time, and are respectively expressed as:

[0055]

[0056] The updated existence probability, the updated model label and the updated trajectory label of the prediction target i are respectively expressed as:

[0057]

[0058] A binary variable g j,k (j = 1, 2, L, M k ) with an initial value of 0 is set to 1, i.e. g j,k = 1, for marking that the measurement data z j,k has been used;

[0059] If θ i >M k , it is considered that the prediction target i is missed, and the updated mean vector and the updated covariance matrix of the prediction target i are respectively taken as:

[0060]

[0061] The updated existence probability, the updated model label and the updated trajectory label of the prediction target i are respectively expressed as:

[0062]

[0063] The update target of the current time is obtained by the above steps, denoted as:

[0064]

[0065] In an optional embodiment, the unused measurement data at time k is sorted to obtain the unused measurement data denoted as:

[0066]

[0067] wherein, is the total number of unused measurement data, is the gth unused measurement data;

[0068] The Mahalanobis distance between the predicted potential new target i and the unused measurement data under different models is calculated, denoted as:

[0069]

[0070] The association probability of the measurement data and the predicted potential new target under different models is:

[0071]

[0072] The update mean vector and the covariance matrix of the predicted potential new target i under different models are respectively denoted as:

[0073]

[0074]

[0075] wherein,

[0076] In an optional embodiment, the model with the maximum association probability of the predicted potential new target i and the measurement data is denoted as:

[0077]

[0078] The Mahalanobis distance corresponding to the model with the maximum association probability is:

[0079]

[0080] The cost matrix of two-dimensional allocation is established with the Mahalanobis distance as the cost, denoted as:

[0081] C = [C1 C2] ​

[0082] wherein:

[0083]

[0084] Solve the cost matrix C with Murty algorithm to obtain the optimal solution wherein, and

[0085] In an optional embodiment, if Take g = θ i , the measurement data is assigned to the predicted potential new target i, a new target h is detected, and the mean vector and covariance of the new target h are represented as:

[0086]

[0087] The existence probability, trajectory label and model label of the new target h are represented as:

[0088] r h,k = 1, The mean vectors of the new target h at the k-2 time and the k-1 time are represented as:

[0089]

[0090] Remove the used measurement data and from the measurement data and , that is:

[0091]

[0092] Through the above steps, the unused measurement data and at the k-1 time and the k time are obtained, and the new target at the k time is:

[0093]

[0094] wherein, The number of new targets, the mean vectors and the trajectory label set of the new targets at the k-2 time, the k-1 time and the k time are represented as:

[0095]

[0096] In an optional embodiment, given a pruning threshold τ r ∈ [0.001, 0.009], the existence probability The target screening out as a surviving target is the target at time k, denoted as:

[0097]

[0098] wherein, is the number of surviving targets, and the mean vector and trajectory label set of the surviving targets at time k are:

[0099]

[0100] The targets at time k are obtained by merging the surviving targets and the new-born targets, and are denoted as:

[0101]

[0102] wherein, is the number of targets at time k;

[0103] The merging set and the set obtain the mean vector and trajectory label set of the targets at time k The set X k is the output of the filter at time k; the set and are used to supplement the outputs of the filter at time k-2 and time k-1, and are denoted as:

[0104]

[0105] In an optional embodiment, the least square method is used to obtain the potential new-born target at the current time by using the measurement data not used at the previous two times, including:

[0106] The measurement data and are selected from the measurement data not used at time k-1 and time k, respectively. and It is judged whether the speed threshold criterion is met:

[0107]

[0108] wherein, v min and v max respectively represent the minimum speed and the maximum speed, and T is the scanning period; the potential new-born target is formed according to the measurement data not used at the continuous two times which meet the speed threshold criterion;

[0109] If the measurement data and meet the speed threshold criterion, a potential new-born target i is detected; the mean vector of the potential new-born target i at time k-1 and time k can be respectively denoted as:

[0110]

[0111] in:

[0112]

[0113] The error covariance of the potential newborn target i at time k is:

[0114]

[0115] Potential newborn target i is represented as:

[0116]

[0117] Where g represents the measurement data index, ε i,k =g represents the measurement data It is associated with potential newborn target i;

[0118] By repeating the above steps, the potential set of newborn targets can be obtained as follows:

[0119]

[0120] in, This represents the potential target number of new students. Attached Figure Description

[0121] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0122] Figure 1 This is a flowchart of a target tracking method for eliminating clutter density requirements according to an exemplary embodiment;

[0123] Figure 2 This is a schematic diagram of the motion trajectories of all targets in one example;

[0124] Figure 3 In one example, the average OSPA (2) Distance comparison diagram;

[0125] Figure 4 This is a schematic diagram of the structure of a target tracking device that meets clutter density requirements according to an exemplary embodiment. Detailed Implementation

[0126] The technical solutions of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts are within the scope of the present application.

[0127] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as there is no conflict.

[0128] Embodiment 1

[0129] To solve the problem that prior parameters such as clutter density and survival probability are required for target tracking in the related art and reduce the calculation complexity, an embodiment of the present application proposes a target tracking method without the requirement of clutter density. As shown in Figure 1 The method comprises the following steps:

[0130] Step A: Based on each target and each potential new target at the previous moment, predicted targets and predicted potential new targets at the current moment under different models are obtained respectively.

[0131] In an example, the target and the potential new target at the previous moment are represented as:

[0132]

[0133]

[0134] Wherein, k-1 represents the previous moment, k represents the current moment, and represent the existence probability, the track label, the model label of the target at k-1 moment and the number of targets at k-1 moment respectively; ε i,k-1 、 and represent the measurement index associated with the potential new target i at k-1 moment, the existence probability, the track label, the model label of the potential new target and the number of potential new targets at k-1 moment respectively; and represent the mean vector and the covariance matrix of the target at k-1 moment, represent the mean vector of the potential new target i at k-2 moment, and represent the mean vector and the covariance matrix of the potential new target at k-1 moment;

[0135] The predicted targets at the current moment under different models calculated are represented as:

[0136]

[0137] where t i,k = 1, 2, L, M r denotes the model label, M r denotes the total number of models used, denotes the predicted existence probability of the target in model t i,k , and denote the predicted trajectory label and predicted model label, respectively, and denote the predicted mean vector and covariance matrix of the target in model t i,k , respectively; and have:

[0138]

[0139]

[0140] where p(t i,k | t i,k-1 ) is the Markov one-step transition probability, Φ k-1 (t i,k ) is the state transition matrix corresponding to model t i,k , Q k-1 (t i,k ) is the covariance in model t i,k ;

[0141] The predicted latent newborn target representation under different models at the current time is calculated as:

[0142]

[0143] where denotes the predicted existence probability of the latent newborn target in model t i,k , and denote the predicted trajectory label and predicted model label, respectively, and denote the predicted mean vector and covariance of the latent newborn target in model t i,k , respectively; and have:

[0144]

[0145]

[0146] The predicted measurement vector and its error covariance matrix of the target in model t i,k are:

[0147]

[0148] In model t i,k Predicting measurement vectors for potential newborn targets and its error covariance matrix for:

[0149]

[0150] Among them, H k It is the observation matrix.

[0151] Step B: Calculate the association probability and Mahalanobis distance between the measured data and the predicted target under different models, as well as the updated mean vector and updated covariance matrix of the predicted target.

[0152] In one example, the probability of association between measured data and predicted targets under different models is expressed as:

[0153]

[0154] in, M represents the measurement data at time k. k z represents the number of measurement data. j,k p represents the j-th measurement data; D It is the detection probability, R k It is the observation noise covariance;

[0155] Model t i,k The measurement data z j,k The Mahalanobis distance between the target and the target is expressed as:

[0156]

[0157] and, It follows a chi-square distribution with n degrees of freedom, where n is the observed data z. j,k The dimension; given a confidence level α, a reception threshold q can be obtained. α ,if Determine the measurement data z j,k It fell into the receiving gate of the predicted target i;

[0158] Model t i,k The updated mean vector and covariance matrix of the target i to be predicted are expressed as follows:

[0159]

[0160]

[0161] in,

[0162] Step C: The Mahalanobis distance corresponding to the model with the maximum association probability of the measurement data z

[0163] In an example, the model with the maximum association probability of the prediction target i and the measurement data z j,k is represented as:

[0164]

[0165] The Mahalanobis distance corresponding to the model with the maximum association probability is:

[0166]

[0167] The cost matrix of the two-dimensional assignment is established with the Mahalanobis distance as the cost, and is represented as:

[0168] C = [C1 C2]

[0169] Wherein:

[0170]

[0171] The optimal solution of the cost matrix is obtained by using the optimal Murty algorithm, and is represented as:

[0172] Step D: Based on the optimal solution obtained in step C, the updated target at the current time is determined.

[0173] In an example, if 1≤θ i ≤M k , it indicates that the measurement data z j,k is associated with the prediction target i, and j = θ i is taken, and the mean vector and the covariance matrix corresponding to the model with the maximum association probability are selected as the updated mean vector and the covariance matrix of the prediction target i at the current time, and are respectively represented as:

[0174]

[0175] The updated existence probability, the updated model label and the updated trajectory label of the prediction target i are respectively represented as:

[0176]

[0177] The binary variable g j,k (j = 1, 2, L, M k ) with an initial value of 0 is set to 1, i.e. g j,k = 1, which is used to mark that the measurement data z j,k has been used;

[0178] If θ i > M k, the prediction target i is considered to be missed, and the updated mean vector and the updated covariance matrix of the prediction target i are respectively taken as:

[0179]

[0180] The updated existence probability, the updated model label and the updated trajectory label of the prediction target i are respectively represented as:

[0181]

[0182] The updated target at the current moment is represented as:

[0183]

[0184] Step E: calculating the association probability, the Mahalanobis distance between the unused measurement data and the prediction potential new target under different models, and the updated mean vector and the covariance matrix of the prediction potential new target.

[0185] In an example, the unused measurement data at the kth moment is sorted to obtain the unused measurement data represented as:

[0186]

[0187] wherein, is the total number of unused measurement data, is the gth unused measurement data;

[0188] The Mahalanobis distance between the prediction potential new target i and the unused measurement data under different models is calculated and represented as:

[0189]

[0190] The association probability of the measurement data and the prediction potential new target under different models is:

[0191]

[0192] The updated mean vector and the covariance matrix of the prediction potential new target i under different models are respectively represented as:

[0193]

[0194]

[0195] wherein,

[0196] Step F: Select the Mahalanobis distance corresponding to the model with the maximum association probability in step E to establish the cost matrix of two-dimensional assignment, and use Murty algorithm to obtain the optimal solution.

[0197] In an example, the model with the maximum association probability of the predicted potential new target i and the measurement data is represented as:

[0198]

[0199] The Mahalanobis distance corresponding to the model with the maximum association probability is:

[0200]

[0201] The cost matrix of two-dimensional assignment is established with the Mahalanobis distance as the cost, and is represented as:

[0202] C = [C1 C2]

[0203] Wherein:

[0204]

[0205] The cost matrix C is solved by Murty algorithm to obtain the optimal solution Wherein, And

[0206] Step G: Based on the optimal solution obtained in step F, determine the new target at the current time.

[0207] In an example, if Take g = θ i , the measurement data is assigned to the predicted potential new target i, a new target h is detected, and the mean vector and covariance of the new target h are represented as:

[0208]

[0209] The existence probability, trajectory label and model label of the new target h are represented as:

[0210] r h,k = 1,

[0211] The mean vectors of the new target h at the k-2 time and the k-1 time are represented as:

[0212]

[0213] The used measurement data and are removed from the measurement data and That is,

[0214]

[0215] Through the above steps, the unused measurement data at time k-1 and time k is obtained And And the new target at time k is:

[0216]

[0217] Wherein, represents the number of new targets, and the mean vector and trajectory label set of the new target at time k-2, time k-1 and time k are:

[0218]

[0219] Step H: filtering the targets with existence probability greater than a given threshold from the updated targets as surviving targets, and merging the surviving targets and the new targets to form the targets at the current time, and taking the targets at the current time as the input of the filter at the next time.

[0220] In an example, given a pruning threshold τ r ∈ [0.001, 0.009], the targets with existence probability are filtered from the updated targets at the current time as surviving targets, and the surviving targets at time k are represented as:

[0221]

[0222] Wherein, is the number of surviving targets, and the mean vector and trajectory label set of the surviving targets at time k are:

[0223]

[0224] Merging the surviving targets and the new targets, the targets at time k are obtained as:

[0225]

[0226] Wherein, is the number of targets at time k;

[0227] Merging the set and the set the mean vector and trajectory label set of the targets at time k are obtained The set X k is the output of the filter at time k; the set and are used to supplement the output of the filter at time k-2 and time k-1, and are represented as:

[0228]

[0229] Step I: using the measurement data unused at the previous two time instants, the potential new-born target at the current time instant is obtained by using least square method, and the potential new-born target at the current time instant is also taken as the input of the filter at the next time instant.

[0230] In an example, the measurement data and are selected from the measurement data unused at the time instants k-1 and k respectively. and It is judged whether the speed threshold criterion is satisfied:

[0231]

[0232] wherein v min and v max represent the minimum speed and the maximum speed respectively, and T is the scanning period; the potential new-born target is formed according to the measurement data unused at the two continuous time instants satisfying the speed threshold criterion;

[0233] If the measurement data and satisfy the speed threshold criterion, a potential new-born target i is detected; the mean vector of the potential new-born target i at the time instants k-1 and k can be respectively represented as:

[0234]

[0235] wherein:

[0236]

[0237] The error covariance of the potential new-born target i at the time instant k is:

[0238]

[0239] The potential new-born target i is represented as:

[0240]

[0241] wherein g represents the index of the measurement data , and ε i,k =g represents that the measurement data is associated with the potential new-born target i.

[0242] The above steps are repeated, and the potential new-born target set is obtained as:

[0243]

[0244] wherein, represents the number of potential new targets.

[0245] In the related art, when target tracking is performed, the calculation complexity is high, the calculation amount is large, and prior parameters such as clutter density and survival probability are needed. Through the above method, the Mahalanobis distance between the predicted target and the measurement data, and the Mahalanobis distance between the potential new target and the unused measurement data are constructed to avoid the prior parameters such as the clutter density and the survival probability, and to reduce the calculation complexity and the calculation amount in the target tracking process. At the same time, the measurement data, the predicted target and the potential new target are used to determine the targets at the current time, and the unused measurement data at two continuous time points is used to determine the potential new target at the current time, so that the measurement data is fully used in the target tracking, and the accuracy of the target tracking is higher.

[0246] Embodiment 2

[0247] In order to better illustrate the content of the present application, the effect of the present application is explained by a specific example. In the embodiment of the present application, there are 8 maneuvering targets, and the motion trajectories of all the maneuvering targets are as shown in Figure 2 The state of each target is composed of position and velocity, and is represented as wherein, and represent the position components, and represent the velocity components, and the superscript T represents the transpose of the vector.

[0248] The state transition matrix is:

[0249] The covariance matrix is:

[0250] wherein, w i,k-1 is the turning rate, Δt k =t k -t k-1 is the time difference between the current time and the previous time, σ v is the process noise standard deviation, the observation matrix and the observation noise covariance matrix are and wherein, σ r and σ α are the observation noise standard deviations, σ r =2.5m,

[0251] In the embodiment of the present application, three motion models are adopted to describe the maneuverability of the target: model 1 is subject to a turning rate of ω i,k-1 =0°s-1 , process noise standard deviation σ υ = 2ms -2 uniform motion; model 2 subject to a turning rate ω i,k-1 = 5°s -1 , process noise standard deviation σ υ = 3ms -2 uniform turning motion; model 3 subject to a turning rate ω i,k-1 = -5°s -1 , process noise standard deviation σ υ = 3ms -2 uniform turning motion. The Markov one-step transition probability matrix among the three models is The initial state vectors of the eight maneuvering targets used are [-800, 12, -100, 16] T , [-800, 12, -500, 16] T , [50, 10, -500, 12] T , [-400, 8, 400, -16] T , [-150, -20, -600, 16] T , [500, -8, -300, 14] T , [670, -10, -650, 18] T , [100, 10, -50, -18] T , which appear at t = 1s, 1s, 1s, 6s, 6s, 15s, 15s, 15s, and disappear at t = 71s, 71s, 60s, 71s, 60s, 71s, 60s, 60s, respectively. Other parameters are set as: target detection probability p D = 0.90, average clutter number N c = 12. σ υ = 2ms -2 , v min = 5ms -1 , v max = 50ms -1 , q α = 7.824 (α = 0.98).

[0252] Table 1

[0253]

[0254] Table 1 shows the average OSPA (2)(Optimal Sub-pattern Assignment, OSPA) error and average execution time.

[0255] Figure 3 For the average OSPA (2) Error comparison diagram. In Figure 3 , the target tracking method provided by the embodiment of the application and the AGLMB filter tracking algorithm are respectively shown in the average OSPA (2) error obtained by 150 experiments. Figure 3 The experimental results in Table 1 show that, compared with the AGLMB filter tracking algorithm, the target tracking method provided by the embodiment of the application can obtain more accurate and reliable target state estimation, and the OSPA (2) error of the target tracking method provided by the embodiment of the application is smaller than the OSPA (2) error of the existing method; at the same time, the target tracking method provided by the embodiment of the application has a high running speed, and the average execution time is much smaller than the average execution time of the AGLMB filter tracking algorithm.

[0256] Embodiment 3

[0257] The embodiment of the application provides a target tracking device without clutter density requirement, as shown in Figure 4 , the device comprises:

[0258] A prediction module 1 is configured to obtain predicted targets and predicted potential new-born targets at a current time based on each target and each potential new-born target at a previous time; for details, refer to the related description of step A in the method embodiment.

[0259] A first calculation module 2 is configured to calculate the association probability, Mahalanobis distance between the measurement data and the predicted targets under different models, and the updated mean vector and the updated covariance matrix of the predicted targets; for details, refer to the related description of step B in the method embodiment.

[0260] A first cost matrix establishment module 3 is configured to select the Mahalanobis distance corresponding to the model with the maximum association probability in step B to establish a two-dimensional assignment cost matrix, and use the Murty algorithm to obtain an optimal solution; for details, refer to the related description of step C in the method embodiment.

[0261] A target determination module 4 is configured to determine the updated target at the current time based on the optimal solution obtained in step C; for details, refer to the related description of step D in the method embodiment.

[0262] The second calculation module 5 is configured to calculate the association probability, the Mahalanobis distance, and the updated mean vector and covariance matrix of the potential new target under different models based on the unused measurement data; for details, refer to the description of step E in the method embodiment.

[0263] The second cost matrix establishment module 6 is configured to establish a two-dimensional allocation cost matrix based on the Mahalanobis distance corresponding to the model with the maximum association probability in step E, and solve the optimal solution by using the Murty algorithm; for details, refer to the description of step F in the method embodiment.

[0264] The new target determination module 7 is configured to determine the new target at the current time based on the optimal solution obtained in step F; for details, refer to the description of step G in the method embodiment.

[0265] The screening module 8 is configured to screen the target with a probability greater than a given threshold value from the updated target as a surviving target, and combine the surviving target and the new target to form the target at the current time, and simultaneously take the target at the current time as the input of the filter at the next time; for details, refer to the description of step H in the method embodiment.

[0266] The potential new target determination module 9 is configured to obtain the potential new target at the current time by using the least square method based on the unused measurement data at the previous two times, and simultaneously take the potential new target at the current time as the input of the filter at the next time; for details, refer to the description of step I in the method embodiment.

[0267] The function description of the target tracking device with the requirement of clutter density elimination provided in the embodiment of the present application is described in detail in the description of the target tracking method with the requirement of clutter density elimination in the above embodiment.

[0268] It should be noted that, in this document, the terms such as “first” and “second” are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between the entities or operations. Moreover, the terms “include”, “contain” or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement “including a…” does not exclude the presence of another identical element in the process, method, article or device including the element.

[0269] The foregoing is considered as illustrative of the principles of the application. Numerous modifications and changes will readily occur to those skilled in the art, and it is intended to embody these modifications and changes into the scope of the application insofar as they fall within the scope of the appended claims.

Claims

1. A clutter-free density demand target tracking method, characterized by, The method comprises: Step A: based on each target and each potential new target at the previous moment, respectively obtaining predicted targets and predicted potential new targets at the current moment under different models; Step B: calculating the association probability, Mahalanobis distance between the measurement data and the predicted targets under different models, and the updated mean vector and the updated covariance matrix of the predicted targets; Step C: selecting the Mahalanobis distance corresponding to the model with the maximum association probability in step B to establish a cost matrix of two-dimensional assignment, and using Murty algorithm to obtain an optimal solution; Step D: determining the updated target at the current moment based on the optimal solution obtained in step C; Step E: calculating the association probability, Mahalanobis distance between the unused measurement data and the predicted potential new targets under different models, and the updated mean vector and the covariance matrix of the predicted potential new targets; Step F: selecting the Mahalanobis distance corresponding to the model with the maximum association probability in step E to establish a cost matrix of two-dimensional assignment, and using Murty algorithm to obtain an optimal solution; Step G: determining the new target at the current moment based on the optimal solution obtained in step F; Step H: selecting the target with a probability greater than a given threshold from the updated target as a surviving target, and merging the surviving target and the new target to form the target at the current moment, and taking the target at the current moment as the input of the filter at the next moment; Step I: using the least square method to obtain the potential new target at the current moment by using the unused measurement data of the previous two moments, and taking the potential new target at the current moment as the input of the filter at the next moment.

2. The method of claim 1, wherein the target and the potential new target at the previous moment are respectively represented as: the predicted targets at the current moment under different models are calculated and represented as: where k-1 represents the previous time, and k represents the current time, and respectively represent the existence probability of the target, the track label, the model label and the number of targets at k-1 time; ε i,k-1 , and respectively represent the measurement index associated with the potential new target i at k-1 time, the existence probability of the potential new target, the track label, the model label and the number of potential new targets at k-1 time; and represent the mean vector and the covariance matrix of the target at k-1 time, represent the mean vector of the potential new target i at k-2 time, and represent the mean vector and the covariance matrix of the potential new target at k-1 time; the predicted potential new targets at the current moment under different models are calculated and represented as: where t i,k = 1, 2, L, M r denotes the model label, M r denotes the total number of models used, denotes the predicted existence probability of the prediction target in model t i,k and denote the predicted trajectory label and the predicted model label, respectively, and denote the predicted mean vector and the covariance matrix of the prediction target in model t i,k , respectively; and have:​ Wherein, ρ(t) i,k |t i,k-1 ) is the Markov one-step transition probability, Φ k-1 (t i,k ) is model t i,k The corresponding state transition matrix, Q k-1 (t i,k ) is model t i,k Covariance under; 3. The method of claim 2, wherein the association probability between the measurement data and the predicted targets under different models is represented as: wherein, represents the predicted existence probability of a potential new target at model t i，k under the predicted trajectory label, and represent the predicted trajectory label and the predicted model label, and represent the predicted mean vector and the covariance of a potential new target at model t i,k under the predicted trajectory label; and At model t i,k The prediction measurement vector of the prediction target and its error covariance matrix is: At model t i,k The predicted measurement vector of a potential new target is predicted and its error covariance matrix is: where H k is the observation matrix.

4. The method of claim 3, wherein the Mahalanobis distance corresponding to the model with the maximum association probability is: a cost matrix of two-dimensional assignment is established by taking the Mahalanobis distance as a cost, and is represented as: wherein, represents measurement data at time k, M k represents the number of measurement data, z j,k represents the jth measurement data; p D is the detection probability, R k is the observation noise covariance; Model t i,k Lower measurement data z j,k The Mahalanobis distance between the prediction target is expressed as: And, obeys a chi-square distribution with n degrees of freedom, where n is the dimension of the observation data z j,k Given a confidence level a, a reception threshold q α If The measurement data z j,k Falls within the reception gate of the predicted target i; Model t i,k The updated mean vector and covariance matrix of the prediction target i are denoted as follows: wherein C=[C1 C2] Predicted target i and measurement data z j,k The model with the highest association probability is represented as: wherein:

5. The method of claim 4, wherein the updated existence probability, the updated model label and the updated trajectory label of the predicted target i are respectively represented as: the updated existence probability, the updated model label and the updated trajectory label of the predicted target i are respectively represented as: the updated target at the current moment is obtained from the above steps and is represented as: The optimal solution of the cost matrix is obtained using the optimal Murty algorithm as 6. The method of claim 5, wherein the updated mean vector and the covariance matrix of the predicted potential new target i under different models are respectively represented as: If 1≤θ i ≤M k , it indicates that the measurement data z j,k is associated with the prediction target i, taking j=θ i , the mean vector and the covariance matrix corresponding to the model with the maximum association probability are selected as the updated mean vector and the covariance matrix of the prediction target i at the current time, denoted as:

7. The method of claim 6, wherein the Mahalanobis distance corresponding to the model with the maximum association probability is: A binary variable g is set to 1, i.e. g = 1, for marking that the measurement data z has been used. j,k (j = 1, 2, L, M k ) is set to 1, i.e. g j,k = 1, for marking that the measurement data z j,k has been used. If θ i > M k , it is considered that the prediction target i is missed, and the updated mean vector and the updated covariance matrix of the prediction target i are taken as: a cost matrix of two-dimensional assignment is established by taking the Mahalanobis distance as a cost, and is represented as: C=[C1 C2] wherein: organize the unused measurement data at the k time point to obtain unused measurement data is represented as: wherein, is the total number of unused measurement data, is the gthunused measurement data; The predicted potential newborn target i and unused measurement data under different models were calculated. The Mahalanobis distance between them is expressed as: Measurement data under different models The associated probability with predicting a potential new-born target is:

8. The method of claim 7, wherein the existence probability, the trajectory label and the model label of the new target h are respectively represented as: wherein, ​ Predicting the probability of a potential new target i being associated with the measured data is represented by the model that maximizes the likelihood ​ ​ ​ ​ Solve the cost matrix C with Murty algorithm to get the optimal solution wherein, and ​ If Take g = θ i Then the measurement data Is assigned to the predicted potential new target i, a new target h is detected, the mean vector and covariance of the new target h are represented as: ​ r h,k ＝1， The mean vector of the new target h at the k-2 time point and the k-1 time point is respectively represented as: from the measurement data and removing used measurement data from the measurement data and i.e.: Through the processing of the above steps, the unused measurement data at time k-1 and time k is obtained and and the new target at time k is: wherein, represents the newly born target, the mean vector and the trajectory label set of the newly born target at the k-2 time, the k-1 time and the k time are:

9. The method of claim 8, characterized in that, Given a pruning threshold τ r ∈ [0.001, 0.009], the target with the existence probability is filtered out as the surviving target from the updated target at the current time, and the surviving target at time k is denoted as: wherein, is the number of surviving targets, and the mean vector and set of trajectory labels of the surviving targets at time k are: The new target and the surviving target are combined to obtain the target at the k time point: wherein, is the target number at time k; merged set and set obtaining the mean vector and the trajectory label set of the target at time k set X k as the output of the filter at time k; set and for the output of the filter at time k-2 and k-1, denoted as:

10. The method of claim 9, wherein, The potential new target at the current time point is obtained by using the least square method by using the measurement data not used at the previous two time points, and the potential new target comprises: measurement data unused at time k-1 measurement data unused at time k selecting measurement data from among and determining whether a speed threshold criterion is met where v min and v max represent the minimum and maximum velocity, respectively, and T is the scan period; potential new targets are formed from unused measurement data at consecutive times satisfying the velocity threshold criterion; If the measurement data and A potential new-born target i is detected if the velocity threshold criterion is met; the mean vectors of the potential new-born target i at time instant k-1 and k can be represented as follows: Wherein: The error covariance of the potential new target i at the k time point is: The potential new target i is represented as: where g denotes the measurement data of index i, ε i,k = g denotes the measurement data associated with the potential new target i; The above steps are repeated to obtain a potential new target set: wherein, represents the number of potential newly born targets.

Images