Rapid target tracking method in dense clutter environment

Abstract

The invention discloses a rapid target tracking method in a dense clutter environment. The method comprises the following steps: determining a prediction target state set and a prediction measurement set at the current moment; constructing a cluster set, wherein the cluster set comprises a detection measurement set and a prediction measurement set corresponding to each cluster; constructing a weight matrix corresponding to each cluster, selecting a plurality of measurement and state association combinations by using the weight matrixes, and determining the weight of each measurement and state association combination; updating a state mean value and a state variance after mapping at the current moment according to each measurement and state association combination; performing weighted averaging on each measurement and state association combination by utilizing the weight of each measurement and state association combination and combining the mapped state mean value and state variance to obtain a target state after state updating; and combining the target states of each cluster after state updating to obtain a target tracking result at the current moment. Compared with an existing method, the method is lower in calculation amount and easier in engineering implementation.

Description

Technical Field

[0001] The present invention relates to the technical field of radar, and particularly to a method for fast target tracking in a dense clutter environment. Background Art

[0002] With the development of modern technology, it is required that radar not only has the ability to track single targets, but also has the ability to accurately track multiple targets. The detection environment of radar is not limited to air detection with a relatively simple environment, but also includes complex ground / sea detection. Especially when detecting sea targets, due to ocean movement, the sea clutter received by the radar is no longer fixed clutter, but has the same Doppler characteristics as moving targets, resulting in a large number of false measurements after the target echoes received by the radar are detected. The problems caused by a large number of false measurements for accurate radar multi-target tracking are mainly as follows:

[0003] 1) A large number of false measurements will lead to a large number of misassociations in the target tracking process. These misassociations include not only the misassociations between tracks and false measurements, but also the misassociations between different tracks and different target measurements, resulting in not only a lot of false tracks, but also a serious decrease in target tracking error.

[0004] 2) A large number of false measurements will make the calculation of the likelihood function for track-measurement association very complex, resulting in a large hardware calculation requirement for specific engineering implementation, thus significantly increasing the hardware cost.

[0005] Traditional multi-target tracking algorithms are based on data association. The key step is to determine the allocation relationship between measurements and target tracks, and to achieve tracking filtering on the basis of track management. The multi-target tracking algorithm based on multi-Bernoulli random set filtering transforms the explicit association process between the measurement set and the elements in the target state set to the set level. Although it avoids the data association process, since the elements in the random set are unordered, it is difficult to estimate the number and state of targets at each moment when there are many targets and false measurements, and it is impossible to accurately give the track information of targets at consecutive moments. At the same time, there are difficulties in subsequent processing such as target identity recognition, situation and threat assessment, and it is difficult to meet the actual engineering needs. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for fast target tracking in a dense clutter environment to overcome the problems of large computational amount, many false tracks and poor target tracking accuracy of traditional multi-Bernoulli algorithms in a dense clutter environment.

[0007] To achieve the above task, the present invention adopts the following technical solutions:

[0008] A method for fast target tracking in a dense clutter environment includes:

[0009] Based on the target state set at the previous moment, determine the predicted target state set and the predicted measurement set at the current moment;

[0010] Perform association clustering on the detection measurement set of the radar at the current moment and the predicted measurement set to construct a clustering set; where the clustering set contains the detection measurement set and the predicted measurement set corresponding to each cluster;

[0011] Construct the weight matrix corresponding to each cluster in the clustering set, and use the index sets corresponding to the detection measurement set and the predicted measurement set to determine the values of the elements in the weight matrix;

[0012] Use the weight matrix to select multiple measurement and state association combinations, and determine the weights of each measurement and state association combination;

[0013] Update the mean and variance of the state after mapping at the current moment according to each measurement and state association combination;

[0014] Use the weights of each measurement and state association combination, combined with the mean and variance of the state after mapping, to perform weighted averaging on each measurement and state association combination to obtain the target state after updating the state;

[0015] Merge the target states after updating the state of each cluster to obtain the target tracking result at the current moment.

[0016] Furthermore, based on the target state set at time k - 1, predict the predicted target state set and the predicted measurement set at the current time k, expressed as:

[0017] r k|k-1 l = p s r k-1 l

[0018] m k|k-1 l = Fm k-1 l

[0019] P k|k-1 l = FP k-1 l F T + Q

[0020]

[0021] In the formula, r k-1 l 、m k-1 l 、P k-1 lis the existence probability, state mean, and state variance of the l-th target at time k-1; r k|k-1 l , m k|k-1 l , P k|k-1 l respectively represent the predicted existence probability, state mean, and state variance of the l-th target at time k; is the position of the l-th predicted measurement, p s is the preset target survival probability, F is the state transition matrix, H is the measurement matrix, Q is the state noise; the superscript T represents the matrix transpose.

[0022] Furthermore, the correlation clustering of the current radar detection measurement set and the predicted measurement set to construct a clustering set includes:

[0023] Calculate the spatial distance d between each detection measurement in the detection measurement set and each predicted measurement in the predicted measurement set n,l ;

[0024]

[0025] where z k n is the position of the n-th detection measurement, N is the number of detection measurements obtained by the radar at time k, and L is the number of targets at time k-1;

[0026] Connect the detection measurement and the predicted measurement with a spatial distance d n,l less than the preset threshold ξ;

[0027] Cluster the detection measurements and predicted measurements with edge connections to obtain a clustering set where represents the set of detection measurements in the s-th cluster at time k, represents the index set of the detection measurements in the s-th cluster, represents the set of predicted measurements in the s-th cluster, θ s represents the index set of the predicted measurements in the s-th cluster, and S represents the number of clusters.

[0028] Furthermore, for the weight matrix corresponding to each cluster in the constructed clustering set, use the index sets corresponding to the detection measurement set and the predicted measurement set to determine the values of the elements in the weight matrix, including:

[0029] Construct the weight matrix C of the s-th cluster, where C is a dimensional matrix, where respectively represent the index sets The number of elements in the middle, the weight matrix C before The values of each element C(i,j) in the

[0030]

[0031] where θ s (i) represents the i-th element in the index set θ s in the middle, represents the index set the j-th element in the middle, represents the index set θ s the target existence probability of the i-th element in the middle, represents the index set the detection measurement position of the j-th element in the middle, represents the index set θ s the predicted measurement position of the i-th element in the middle, P D represents the preset target detection probability, represents the state variance of the i-th element in the index set θ at time k, H is the measurement matrix, R represents the measurement noise covariance matrix, s represents the mean and the variance is when the Gaussian distribution at the probability density at;

[0032] The weight matrix C the values of each element in the

[0033]

[0034] Furthermore, the method of selecting multiple measurement and state association combinations by using the weight matrix specifically adopts an improved Gibbs sampling method:

[0035] First, set g = 0, and then use random sampling to obtain the initial measurement and state association combination γ 0 the mapping relationship of is:

[0036] γ 0 (i) = r 0,i i = 1,...|θ s |

[0037] where, γ 0 (i) represents the i-th measurement in the initial measurement and state association combination γ 0 in the middle, r 0,i represents the measurement index number obtained by randomly sampling the detection measurements from 1 to in the middle; ​

[0038] Set g = g + 1, and obtain the g-th measurement and state association combination γ as follows: g The mapping is:

[0039] γ g (i) = r g,i i=1,…|θ s |

[0040] Among them, γ g (i) represents the g-th measurement and state association combination γ g The i-th measurement in r g,i satisfy:

[0041]

[0042] Calculate the gth measurement and state association combination γ g The weight w g for:

[0043]

[0044] Determine whether g is less than the preset number G of measurement and state association combinations. If so, update g = g + 1 and continue to calculate γ g .

[0045] Furthermore, the updating of the state mean and state variance after mapping at the current moment according to each measurement and state association combination includes:

[0046] when hour:

[0047]

[0048] when hour:

[0049]

[0050] In the formula The predicted index set θ for the k-time prediction s The state mean of the i-th element in , Represents the g-th measurement and state association combination γ g The detection measurement corresponding to the i-th measurement in , Represents the index set θ s The predicted measurement corresponding to the i-th element in , I represents the identity matrix, H is the measurement matrix, represents the index set θ at time k s The state variance of the i-th element in , R represents the measurement noise covariance matrix, G i is the gain matrix corresponding to the i-th measurement, and its expression is:

[0051]

[0052] Further, using the weights of each measurement and state association combination, combining the mapped state mean and state variance, and performing weighted averaging on each measurement and state association combination to obtain the target state after updating the state, includes:

[0053] Obtaining the target state after updating the state is:

[0054]

[0055] where w g is the weight of the g-th measurement and state association combination γ g , G is the preset number of measurement and state association combinations, r k|k-1 i represents the existence probability of the i-th element in the prediction index set θ s at time k, represents the state variance after mapping at time k - 1, represents the target existence probability after weighted average update corresponding to the i-th element in the current prediction index set θ s at time k, represents the state mean after weighted average update corresponding to the i-th element in the current prediction index set θ s at time k, represents the state variance after weighted average update corresponding to the i-th element in the current prediction index set θ s at time k.

[0056] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, the fast target tracking method in the dense clutter environment is implemented.

[0057] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the fast target tracking method in the dense clutter environment is implemented.

[0058] Compared with the prior art, the present invention has the following technical characteristics:

[0059] Compared with the traditional Bernoulli algorithm, the present invention can effectively reduce false tracks and improve the multi-target tracking accuracy, with lower computational complexity and easier engineering implementation. By designing a weight matrix in each interval and performing multi-hypothesis set association, the present invention can effectively reduce the misassociation between tracks and false measurements, as well as between tracks and other target measurements, thereby effectively reducing false tracks and improving the multi-target tracking accuracy. Description of the Drawings

[0060] Figure 1 is a schematic flow chart of the method of the present invention;

[0061] Figure 2 is the cumulative result of multi-frame measurements of a radar in a complex environment in an embodiment

[0062] Figure 3 is the result of multi-target track and complex measurement interval division in a dense clutter environment in an embodiment

[0063] Figure 4 is the multi-target tracking result in a dense clutter environment in an embodiment;

[0064] Figure 5 is the comparison result of the target tracking position accuracy and the traditional method in an embodiment. Detailed Embodiment

[0065] The present invention provides a fast target tracking method in a dense clutter environment, which is mainly applied to low / seasurface or low-altitude monitoring and multi-target tracking in a dense clutter environment, reducing the false track rate and calculation amount of target tracking by a radar in a complex environment, and improving the target tracking accuracy. The specific implementation process of the present invention is as follows:

[0066] Step 1, set the current time as k and the previous time as k-1; set the target state set at time k-1 as {r k-1 l ,m k-1 l ,P k-1 l}, l＝1 L , where L is the number of targets at time k-1, r k-1 l is the existence probability of the l-th target at time k-1, m k-1 l is the state mean of the l-th target at time k-1, P k-1 l is the state variance of the l-th target at time k-1; the detection measurement set of the radar at time k is where N is the number of detection measurements obtained by the radar at time k, z k n is the position of the n-th detection measurement; among which the detection measurement includes targets and interferences that are non-targets but are misdetected.

[0067] Step 2, first predict the target state at time k-1, and respectively obtain the predicted target state set at time k as {r k|k-1 l ,mk|k-1 l , P k|k-1 l} l＝1 L and the predicted measurement set

[0068] r k|k-1 l = p s r k-1 l

[0069] m k|k-1 l = Fm k-1 l

[0070] P k|k-1 l = FP k-1 l F T + Q

[0071]

[0072] where r k|k-1 l , m k|k-1 l , P k|k-1 l respectively represent the existence probability, state mean, and state variance of the l-th target at the k-th moment obtained by prediction; is the position of the l-th predicted measurement, p s is the preset target survival probability, F is the state transition matrix, H is the measurement matrix, and Q is the state noise; the superscript T represents the matrix transpose.

[0073] Step 3, perform association clustering on the detection measurement set and the predicted measurement set to obtain the clustering set as The specific steps of the association clustering are as follows:

[0074] 3.1) Calculate the spatial distance d between each detection measurement in the detection measurement set and each predicted measurement in the predicted measurement set n,l ;

[0075]

[0076] 3.2) Connect the detection measurement and the predicted measurement with a spatial distance d n,l less than the preset threshold ξ;

[0077] 3.3) Cluster the detected measurements and predicted measurements with edge connections to obtain a cluster set where represents the set of detected measurements in the s-th cluster at time k, represents the index set of the detected measurements in the s-th cluster, represents the set of predicted measurements in the s-th cluster, θ s represents the index set of the predicted measurements in the s-th cluster, and S represents the number of clusters.

[0078] Step 4, construct the weight matrix C of the s-th cluster, where C is a dimensional matrix, where respectively represent the number of elements in the index set . The value of each element C(i,j) in the first dimensional column vector of the weight matrix C is as follows:

[0079]

[0080] where θ s (i) represents the i-th element in the index set θ s , represents the j-th element in the index set , represents the target existence probability of the i-th element in the index set θ s , represents the detected measurement position of the j-th element in the index set , represents the predicted measurement position of the i-th element in the index set θ s , P D represents the preset target detection probability, represents the state variance of the i-th element in the index set θ s at time k, H is the measurement matrix, R represents the measurement noise covariance matrix, represents a Gaussian distribution with mean and variance at the probability density.

[0081] The value of each element C(i,j) in the dimensional column vector of the weight matrix C is as follows:

[0082]

[0083] Step 5, use the improved Gibbs sampling method to select G measurement and state association combinations γ g ; where g = 1, 2,..., G, γg Denote the g-th state association combination; the specific sampling method is as follows:

[0084] 5.1) First, set g = 0, and then use random sampling to obtain the initial measurement and state association combination γ 0 The mapping relationship is:

[0085] γ 0 (i) = r 0,i i = 1, … |θ s |

[0086] where γ 0 (i) represents the i-th measurement in the initial measurement and state association combination γ 0 and r 0,i represents the measurement index number obtained by randomly sampling the detection measurements from 1 to .

[0087] 5.2) Set g = g + 1, and obtain the g-th measurement and state association combination γ g according to the following method. The mapping is:

[0088] γ g (i) = r g,i i = 1, … |θ s |

[0089] where γ g (i) represents the i-th measurement in the g-th measurement and state association combination γ g and r g,i represents the index number of the detection measurement that does not belong to the vector [γ g (1), …, γ g (i - 1), γ g-1 (i + 1), …, γ g-1 (|θ s |)] and has the maximum weight, that is:

[0090]

[0091] 5.3) Calculate the weight w g of the g-th measurement and state association combination γ g as:

[0092]

[0093] 5.4) Judge whether g is less than G. If it is less, go back to step 5.2) to continue running; otherwise, end the operation.

[0094] Step 6, according to each measurement and state association combination γ g obtain the state mean value after mapping at the current k moment With the update of the state variance , we get:

[0095] When :

[0096]

[0097] When :

[0098]

[0099] In the formula is the state mean of the i-th element in the predicted prediction index set θ s at time k, represents the measured measurement corresponding to the i-th measurement in the g-th measurement and state association combination γ g , represents the predicted measurement corresponding to the i-th element in the index set θ s . I represents the identity matrix, H is the measurement matrix, represents the state variance of the i-th element in the index set θ s at time k. R represents the measurement noise covariance matrix, G i is the gain matrix corresponding to the i-th measurement, and its expression is:

[0100]

[0101] Step 6, using the weight w g of the g-th measurement and state association combination γ g , combined with the mapped state mean and the state variance , perform weighted averaging on each measurement and state association combination γ g to obtain the target state after updating the state as:

[0102]

[0103] where r k|k-1 i represents the existence probability of the i-th element in the predicted prediction index set θ s at time k, represents the state variance after mapping at time k-1, represents the target existence probability after weighted average update corresponding to the i-th element in the current predicted prediction index set θ s , represents the state mean after weighted average update corresponding to the i-th element in the current predicted prediction index set θ s at time k. Denoted as the predicted index set θ at the current k-th moment s The i-th element in corresponds to the state variance after weighted average update.

[0104] Step 7, merge the target states after each cluster update to obtain the target tracking result at the k-th moment as

[0105] Example:

[0106] 1. Experimental verification conditions.

[0107] Continuous wave radar is used for data acquisition in the experiment. The data acquisition environment is the urban suburbs, etc. The targets of this experiment are mainly slow small targets such as pedestrians and drones.

[0108] 2. Experimental results.

[0109] Figure 2 It is the measurement result after pulse compression, MTD, and constant false alarm detection of the collected echoes when the radar detects low-altitude and ground targets in the urban suburbs. It can be seen from the figure that due to a lot of clutter in the urban suburbs, including many urban buildings, as well as woods and hilly terrains, etc., and there are also many vehicle targets on the ground. Therefore, after the radar constant false alarm detection, a large number of measurements are included. These measurements not only include real target measurements, but also include a large number of clutter false measurements, thus having a serious impact on target track tracking.

[0110] Figure 3 It is the result of interval division of the measurements obtained by radar detection and the target tracks. It can be seen from the figure that after interval division, each target track and measurement only need to be associated with the components within their own intervals, thus greatly reducing the association calculation amount and reducing the engineering implementation difficulty of this embodiment.

[0111] Figure 4 This is the target track obtained after tracking the measurements obtained by radar detection in this embodiment. It can be seen from the result after using the tracking method of this embodiment that the figure includes multiple target tracks. It can be seen that the tracks of the targets are continuous and without breakpoints after tracking, indicating that the method of this embodiment can achieve stable tracking of multiple targets.

[0112] Figure 5 It is the comparison of the target tracking and positioning accuracy results obtained by using the method of this embodiment and the traditional method respectively. It can be seen from the figure that the method proposed in this aspect has higher positioning accuracy than the traditional method.

[0113] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included within the protection scope of the present application.

Claims

1. A fast target tracking method in a dense clutter environment, characterized in that: include: Based on the target state set at the previous moment, determine the predicted target state set and predicted measurement set at the current moment; Associating and clustering the detection measurement set of the radar at the current moment and the predicted measurement set to construct a cluster set; The cluster set includes a detection measurement set and a prediction measurement set corresponding to each cluster; Constructing a weight matrix corresponding to each cluster in the cluster set, and determining the value of each element in the weight matrix using the index set corresponding to the detection measurement set and the prediction measurement set; Selecting a plurality of measurement-state association combinations using the weight matrix, and determining a weight for each measurement-state association combination; Update the state mean and state variance after mapping at the current moment according to each measurement and state association combination; Using the weight of each measurement and state association combination, combined with the mapped state mean and state variance, each measurement and state association combination is weighted averaged to obtain the target state after the state is updated; The target states after each cluster update are merged to obtain the target tracking result at the current moment.

2. The fast target tracking method in a dense clutter environment according to claim 1, characterized in that: Based on the target state set at time k-1, the predicted target state set and predicted measurement set at time k are predicted, which can be expressed as: r k|k-1 l ＝p s r k-1 l m k|k-1 l =Fm k-1 l P k|k-1 l =FP k-1 l F T +Q In the formula, r k-1 l 、m k-1 l , P k-1 l is the existence probability, state mean and state variance of the lth target at time k-1; r kk-1 l 、m kk-1 l , P kk-1 l They represent the predicted existence probability, state mean, and state variance of the lth target at time k respectively; is the position of the lth predicted measurement, p s is the preset target survival probability, F is the state transfer matrix, H is the measurement matrix, Q is the state noise; the superscript T represents the matrix transpose.

3. The fast target tracking method in a dense clutter environment according to claim 1, characterized in that: The associating and clustering the detection measurement set of the radar at the current moment and the prediction measurement set to construct a cluster set includes: Calculate the detection measurement set separately The detection measurement and prediction measurement set in The spatial distance d between the predicted measurements in n,l ; Among them, z k n is the position of the nth detection measurement, N is the number of detection measurements obtained by the radar at time k, and L is the number of targets at time k-1; The spatial distance d n,l The detected measurements that are less than the preset threshold ξ are edge-connected with the predicted measurements; Cluster the edge-connected detection and prediction measurements to obtain a cluster set in represents the set of detection measurements in the s-th cluster at time k, represents the index set of detection measurements in the sth cluster, represents the set of predicted measurements in the sth cluster, θ s represents the set of indices of the predicted measurements in the sth cluster, and S represents the number of clusters.

4. The method for fast target tracking in a dense clutter environment according to claim 1, characterized in that: The step of constructing a weight matrix for each cluster pair in the cluster set and determining the value of each element in the weight matrix using the index set corresponding to the detection measurement set and the prediction measurement set includes: Construct the sth clustering weight matrix C, where C is a dimensional matrix, where Respectively represent index collections The number of elements in the weight matrix C The values ​​of each element C(i,j) in the dimension column vector are as follows: where θ s (i) represents the index set θ s The i-th element in Represents an index collection The jth element in Represents the index set θ s The target existence probability of the i-th element in is, Represents an index collection The detected measurement position of the jth element in , Represents the index set θ s The predicted measured position of the i-th element in P D represents the preset target detection probability, represents the index set θ at time k s The state variance of the i-th element in , H is the measurement matrix, R represents the measurement noise covariance matrix, Indicates the mean and the variance is The Gaussian distribution at The probability density of Weight matrix C The values ​​of each element in the dimension column vector C(i,j) are as follows:

5. The method for fast target tracking in a dense clutter environment according to claim 1, characterized in that: The weight matrix is ​​used to select a plurality of measurement and state association combinations, specifically using an improved Gibbs sampling method: First, set g = 0, and then use random sampling to obtain the initial measurement and state association combination γ0: γ0(i)=r 0,i i=1,…|θ s | Where γ0(i) represents the i-th measurement in the initial measurement and state association combination γ0, r 0,i Indicates 1 to The measurement index number obtained by randomly sampling the detection measurement between; Set g = g + 1, and obtain the g-th measurement and state association combination γ as follows: g The mapping is: c g (i)=r g,i i=1,…|θ s | Among them, γ g (i) represents the g-th measurement and state association combination γ g The i-th measurement in r g,i satisfy: Calculate the gth measurement and state association combination γ g The weight w g for: Determine whether g is less than the preset number G of measurement and state association combinations. If so, update g = g + 1 and continue to calculate γ g .

6. The fast target tracking method in a dense clutter environment according to claim 1, characterized in that: The updating of the state mean and state variance after mapping at the current moment according to each measurement and state association combination includes: when hour: when hour: In the formula The predicted index set θ for the k-time prediction s The state mean of the i-th element in , Represents the g-th measurement and state association combination γ g The detection measurement corresponding to the i-th measurement in , Represents the index set θ s The predicted measurement corresponding to the i-th element in , I represents the identity matrix, H is the measurement matrix, represents the index set θ at time k s The state variance of the i-th element in , R represents the measurement noise covariance matrix, G i is the gain matrix corresponding to the i-th measurement, and its expression is:

7. The fast target tracking method in a dense clutter environment according to claim 1, characterized in that: The method uses the weight of each measurement and state association combination, combines the mapped state mean and state variance, and performs weighted averaging on each measurement and state association combination to obtain the target state after the state is updated, including: Get the target state after updating the state for: Among them, w g is the gth measurement and state association combination γ g The weight of G is the preset number of measurement and state association combinations, r k|k-1 i Represents the k-time prediction index set θ s The probability of the existence of the i-th element in , represents the state variance after mapping at time k-1, Represented as the current k-time prediction index set θ s The i-th element in corresponds to the weighted average updated target existence probability, Represented as the current k-time prediction index set θ s The i-th element in corresponds to the weighted average updated state mean, P k θs(i) Represented as the current k-time prediction index set θ s The i-th element in corresponds to the weighted average updated state variance.

8. A terminal device comprising a processor, a memory and a computer program stored in the memory; characterized in that: When the processor executes the computer program, the method for fast target tracking in a dense clutter environment according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, wherein a computer program is stored in the medium; characterized in that: When the computer program is executed by a processor, the method for fast target tracking in a dense clutter environment according to any one of claims 1 to 7 is implemented.

Images