A method, device, equipment and medium for generating continuous tracks of multiple extended targets

Abstract

The present application discloses a method, device, equipment and medium for generating continuous tracks of multiple extended targets, which relates to the technical field of multiple target tracking, including: when an extended target is blocked in a target tracking scene, constructing a set of extinct targets at the previous moment and a corresponding first LMB parameter set for the extended target; obtaining a set of targets to be newly generated at the current moment and a corresponding second LMB parameter set based on the first LMB parameter set; judging whether each target to be newly generated belongs to a measured newly generated target based on the measurement set at the current moment, and determining a third LMB parameter set corresponding to each measured newly generated target; filtering based on the multi-target density including the third LMB parameter set at the current moment; obtaining target state information and target track information based on the filtering result, and obtaining the set of targets to be newly generated updated at the current moment through compilation. In this way, the continuity and accuracy of the track before and after the target is blocked are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-target tracking, and in particular to a method, device, equipment and medium for generating continuous tracks of multiple extended targets. Background Art

[0002] With the widespread application of high-resolution sensors, a single target will generate multiple measurement points. At this time, the traditional point target model assumption is no longer valid. In order to obtain good tracking performance, an extended target tracking method should be used. For complex application scenarios where the number of targets in the monitoring area changes dynamically, the target tracking method based on random finite sets (RFS) has more advantages than the traditional tracking method based on data association. Combining the extended target modeling method with the multi-target tracking filter can successfully achieve multi-extended target tracking. However, in more complex actual scene conditions such as near ports and near shores, the target may be temporarily blocked by a cross-river bridge during movement, resulting in inaccurate target state estimation and discontinuous tracks of the same target. This makes the traditional extended target tracking method face practical challenges.

[0003] Although algorithms such as Gaussian Inverse Wishart Probability Hypothesis Density (GIW-PHD) filters have the ability to accurately estimate the kinematic states and extended forms of multiple extended targets, the implementation process of GIW-PHD filtering shows that at each specific moment, the filter provides an unordered set of estimated results of each target state (covering kinematic states and extended forms). Due to the disorder of this set, although the number of targets can be estimated through the potential analysis of the set, there is a lack of clear correspondence between the estimated states of each target at different moments, so the target track cannot be effectively formed.

[0004] At present, there are two main methods to generate continuous tracks of targets. One feasible method is to introduce target labels based on state estimation methods such as probability hypothesis density (PHD) filters to distinguish different targets. Another approach is to use a target tracking method based on label RFS theory. The theoretical derivation of this method has a closed solution, so it is more popular in practical applications. However, in complex scenes such as near ports and near shores, extended targets may be obscured, resulting in the loss of measurements at some moments. This will lead to misjudgment of the tracking algorithm, causing the same target to present two different labels (corresponding to two different tracks) before and after the occlusion, which is also one of the common problems of the current tracking method based on label RFS. Therefore, how to generate continuous and accurate tracks when the extended target is obscured is an urgent problem to be solved in the current field. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for generating continuous tracks of multiple extended targets, which can effectively improve the continuity and accuracy of the tracks before and after target occlusion. The specific scheme is as follows:

[0006] In a first aspect, the present application provides a method for generating continuous tracks of multiple extended targets, comprising:

[0007] When an extended target is blocked in the target tracking scene, a set of eliminated targets at a moment before the current moment and a corresponding first LMB parameter set are constructed for the extended target;

[0008] Based on the first LMB parameter set and the corresponding target motion characteristics, a set of targets to be generated at the current moment and a corresponding second LMB parameter set are obtained;

[0009] Based on the measurement set at the current moment, determine whether each target to be newly created in the target to be newly created set belongs to a measurement newly created target, and determine a third LMB parameter set corresponding to each measurement newly created target based on the second LMB parameter set;

[0010] Based on the current moment, performing GGIW-LMB filtering using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result;

[0011] Based on each posterior density component and weight in the filtering result, the corresponding target state information and target track information are obtained, and the updated target set to be generated at the current moment is obtained through reorganization.

[0012] Optional,

[0013] The step of constructing a set of extinction targets at a moment before the current moment and a corresponding first LMB parameter set for the extended target includes:

[0014] Obtain a first extended target state estimation result and a second extended target state estimation result corresponding to two moments before the current moment respectively;

[0015] Utilizing the difference between the label sets at the first two moments, acquiring corresponding extended target information through the first extended target state estimation result and the second extended target state estimation result, so as to obtain the extinction target set at the previous moment and the corresponding first LMB parameter set;

[0016] The extended target information includes kinematic state information, extended morphology information and measurement rate information.

[0017] Optional,

[0018] The obtaining, based on the first LMB parameter set and the corresponding target motion characteristics, a target set to be generated at the current moment and a corresponding second LMB parameter set includes:

[0019] By merging the first LMB parameter set and the target LMB parameter set to be generated after being reorganized at the previous moment, a corresponding merged target set and a corresponding LMB parameter set are obtained;

[0020] Based on the motion characteristics of the targets in the merged target set, the LMB parameter set corresponding to the merged target set is preprocessed to obtain a target set to be newly generated at the current moment and a corresponding second LMB parameter set.

[0021] Optionally, judging whether each target to be newly created in the target to be newly created set belongs to a measurement newly created target based on the measurement set at the current moment, and determining a third LMB parameter set corresponding to each measurement newly created target based on the second LMB parameter set includes:

[0022] Acquire the measurement set at the current moment, and respectively calculate the Mahalanobis distance between the pseudo measurement of the kinematic state of each target in the target set to be newly generated and each measurement value in the measurement set to obtain corresponding first distance information;

[0023] Obtaining a first preset distance threshold, and comparing the obtained plurality of first distance information corresponding to each of the to-be-generated targets with the first preset distance threshold respectively, to obtain threshold comparison results corresponding to each of the to-be-generated targets respectively;

[0024] The corresponding measurement quantity threshold is calculated through the measurement rate information of each of the targets to be newly generated, and when the measurement quantity less than the first preset distance threshold in the comparison result is greater than the measurement quantity threshold, it is determined that the corresponding target to be newly generated belongs to the measurement newly generated target, and the corresponding third LMB parameter set is obtained.

[0025] Optionally, performing GGIW-LMB filtering using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result includes:

[0026] Acquire the multi-target prior density at the current moment; wherein the multi-target prior density includes the corresponding new target density, the surviving target density and the third LMB parameter set used to measure the new target density;

[0027] Convert the LMB distribution of the multi-objective prior density into a δ-GLMB distribution, and perform a GGIW-LMB filter update step based on the converted multi-objective prior density and the measurement set at the current moment to obtain the multi-objective posterior density at the current moment;

[0028] By performing corresponding LMB distribution transformation on the multi-objective posterior density, the transformed multi-objective posterior density is obtained as a filtering result.

[0029] Optionally, the obtaining of corresponding target state information and target track information based on each posterior density component and weight in the filtering result, and obtaining the updated target set to be generated at the current moment by reorganization, includes:

[0030] Obtaining a preset weight threshold, eliminating the posterior density components whose weights are less than the preset weight threshold in the filtering results, and obtaining the corresponding eliminated results;

[0031] Based on each of the posterior density components in the eliminated results, corresponding state information, covariance matrix and inverse Wishart distribution parameters are obtained;

[0032] Arrange the posterior density components in the elimination result in order from small to large according to the weights to obtain the current component arrangement result;

[0033] Based on the component arrangement result, the posterior density component with the largest current weight is taken as the merging center, and the Mahalanobis distance between the merging center and the current component to be merged is calculated by extracting the state information, the covariance matrix and the inverse Wishart distribution parameter corresponding to the merging center to obtain the corresponding second distance information;

[0034] Comparing each of the second distance information with a second preset distance threshold, respectively, and based on the comparison result obtained, merging the posterior density components corresponding to the second distance information that is smaller than the second preset distance threshold;

[0035] For each of the unmerged posterior density components in the elimination result, arrange them in order from small to large according to the weight to obtain a new component arrangement result, and jump again to the step of taking the posterior density component with the largest current weight as the merging center based on the component arrangement result, until the merging operation of all the posterior density components is completed to obtain the corresponding component merging result;

[0036] A preset number of target components with the largest weights in the component merging results are retained, and corresponding target state information and target track information are obtained through the target components.

[0037] Optionally, the updated target set to be newly generated at the current moment obtained by reorganization includes:

[0038] In the target set to be generated, for each target to be generated that does not belong to the measured target, targets whose existence time is longer than a preset time threshold are eliminated to obtain the target set to be generated updated at the current moment and the corresponding LMB parameter set.

[0039] In a second aspect, the present application provides a device for generating continuous tracks of multiple extended targets, comprising:

[0040] A parameter set construction module, for constructing a set of extinct targets at a moment before the current moment and a corresponding first LMB parameter set for the extended target when an extended target is blocked in the target tracking scene;

[0041] A set acquisition module, used to obtain a target set to be generated at the current moment and a corresponding second LMB parameter set based on the first LMB parameter set and the corresponding target motion characteristics;

[0042] a target judgment module, configured to judge whether each target in the target set to be newly generated belongs to a measurement newly generated target based on the measurement set at the current moment, and determine a third LMB parameter set corresponding to each of the measurement newly generated targets based on the second LMB parameter set;

[0043] A filtering result acquisition module, used to perform GGIW-LMB filtering based on the current moment using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result;

[0044] The track generation module is used to obtain the corresponding target state information and target track information based on the posterior density components and weights in the filtering results, and obtain the updated target set to be generated at the current moment through reorganization.

[0045] In a third aspect, the present application provides an electronic device, including:

[0046] Memory, used to store computer programs;

[0047] The processor is used to execute the computer program to implement the steps of the aforementioned method for generating continuous tracks of multiple extended targets.

[0048] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of the aforementioned method for generating continuous tracks of multiple extended targets.

[0049] It can be seen that in the present application, when there is an extended target that is occluded in the target tracking scene, a set of extinct targets at the previous moment before the current moment and the corresponding first LMB parameter set are constructed for the extended target; based on the first LMB parameter set and the corresponding target motion characteristics, a set of targets to be newly generated at the current moment and the corresponding second LMB parameter set are obtained; based on the measurement set at the current moment, it is determined whether each target to be newly generated in the set of targets to be newly generated belongs to a measured newly generated target, and based on the second LMB parameter set, a third LMB parameter set corresponding to each measured newly generated target is determined; based on the current moment, GGIW-LMB filtering is performed using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result; based on each posterior density component and weight in the filtering result, corresponding target state information and target track information are obtained, and the updated set of targets to be newly generated at the current moment is obtained through compilation. That is to say, in the present application, in view of the situation where the target of the extended target is obscured, the first LMB parameter set of the previous moment of the constructed current moment is processed to obtain the target set to be newly generated at the current moment and the corresponding second LMB parameter set, and then the new target judgment is measured based on the measurement set, and the third LMB parameter set corresponding to each of the measured new targets is obtained, and then filtering is performed, and the final target state information and target track information are obtained according to each posterior density component and weight in the filtering result. In this way, the continuity and accuracy of the track before and after the target is obscured can be effectively improved, so that the consistency of the track before and after the target is obscured can be maintained while ensuring the accuracy of the estimation of the extended target state. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0051] Figure 1 A flow chart of a method for generating continuous tracks of multiple extended targets provided in this application;

[0052] Figure 2 A specific flow chart of a method for generating continuous tracks of multiple extended targets provided in this application;

[0053] Figure 3 A target tag transfer schematic diagram provided for this application;

[0054] Figure 4 A flowchart for obtaining target track information based on filtering results provided by this application;

[0055] Figure 5 A schematic diagram of a real motion trajectory of a target provided in this application;

[0056] Figure 6 A schematic diagram of a target kinematic state and track estimation result provided in this application;

[0057] Figure 7 A schematic diagram of a target expansion form and track estimation result provided in this application;

[0058] Figure 8 A schematic diagram of the structure of a device for generating continuous tracks of multiple extended targets provided in this application;

[0059] Fig. 9 A structural diagram of an electronic device provided for this application. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] At present, there are two main methods for generating the continuous track of the target. A feasible method is to introduce the target label on the basis of the state estimation method such as the probability hypothesis density filter to distinguish different targets. Another approach is to adopt the target tracking method based on the label RFS theory. The theoretical derivation of this method has a closed solution, so it is more favored in practical applications. However, in complex scenes such as near ports and near shores, the extended target may be blocked and the measurement of some moments may be lost. This will lead to misjudgment of the tracking algorithm, so that the same target presents two different labels (corresponding to two different tracks) before and after the occlusion, which is also one of the common problems of the tracking method based on the label RFS. Therefore, how to generate continuous and accurate tracks when the extended target is blocked is a problem to be solved in the current field. To this end, the present application provides a multi-extended target continuous track generation scheme, which can effectively improve the continuity and accuracy of the track before and after the target is blocked.

[0062] See also Figure 1 As shown, an embodiment of the present invention discloses a method for generating continuous tracks of multiple extended targets, including:

[0063] Step S11: when an extended target is blocked in the target tracking scene, a set of eliminated targets at a moment before the current moment and a corresponding first LMB parameter set are constructed for the extended target.

[0064] Combination Figure 2 As shown, in this embodiment, the state of the extended target is first modeled as in, is the measurement rate parameter, modeled as a gamma distribution; is the kinematic state vector (including the position coordinate x in the x direction and the velocity The position coordinate y and speed in the y direction T represents the transposed sign), modeled as a Gaussian distribution; X ε Expand the morphological matrix for the target, modeled as an inverse Wishart distribution.

[0065] Specifically, the construction of the first LMB (Labeled Multi-Bernoulli) parameter set for the extended target at the previous moment (i.e., moment k-1, k represents the observation time period sequence number) includes: obtaining the first extended target state estimation result and the second extended target state estimation result corresponding to the two moments before the current moment (i.e., moment k-1 and moment k-2), respectively; utilizing the difference between the label sets of the first two moments, obtaining the corresponding extended target information through the first extended target state estimation result and the second extended target state estimation result, so as to obtain the extinct target set at the previous moment and the corresponding first LMB parameter set; wherein the extended target information includes kinematic state information, extended morphology information and measurement rate information. Specifically by comparing the first and second extended target state estimation results. It should be understood that the posterior density of the extended target that disappears at moment k-1 can be obtained to include the label set Parameter Set (in, ), that is, the first LMB parameter set. represents the existence probability of the target with label l, represents the posterior density function of the target with label l (including the kinematic state information, extended morphological information, measurement rate information, etc. of the extended target), is a gamma-Gaussian-inverse Wishart mixture distribution, expressed as:

[0066]

[0067] In the formula, Representation variables It follows a gamma distribution with shape parameter α and scale parameter β; Represents vector The mean is m and the covariance matrix is ( represents the Kronecker product) of the Gaussian distribution; Represents the matrix X ε It follows an inverse Wishart distribution with v degrees of freedom and a parameter matrix V.

[0068] Step S12: Based on the first LMB parameter set and the corresponding target motion characteristics, a target set to be generated at the current moment and a corresponding second LMB parameter set are obtained.

[0069] In this embodiment, the obtaining of the target set to be generated at the current moment and the corresponding second LMB parameter set based on the first LMB parameter set and the corresponding target motion characteristics includes: obtaining the corresponding merged target set and the corresponding LMB parameter set by merging the first LMB parameter set and the LMB parameter set of the target to be generated after being reorganized at the previous moment; and preprocessing the LMB parameter set corresponding to the merged target set based on the motion characteristics of the targets in the merged target set to obtain the target set to be generated at the current moment and the corresponding second LMB parameter set. Specifically, assuming that the target set to be generated after being reorganized at the k-1 moment is represented as a label set (k d represents the disappearance moment of the target, that is, the moment of occlusion), parameter set (in, ), and merge it with the extinction target set at time k-1 (that is, the first LMB parameter set). The merged set is defined as the label space Parameter Set (in, )’s LMB distribution.

[0070] The LMB parameter set corresponding to the merged target set is preprocessed according to the motion characteristics of the extended target. Generally, assuming that the motion characteristics of the extended target in motion will not change significantly before and after passing through the occluder, the state transfer function of the preprocessing process is expressed as:

[0071] P MB (x,l)φ(x + |x,l);

[0072] In the formula, is the state transfer kernel, x and x + They represent the extended target states before and after the state transition, P MB (x, l) represents the evolution probability. The LMB parameter set of the target to be generated at the current moment (i.e., moment k) calculated in this way, i.e., the second LMB parameter set, is expressed as (in, ), the expression in this parameter set can be calculated by the following two formulas:

[0073]

[0074]

[0075] Step S13: Based on the measurement set at the current moment, determine whether each target to be newly generated belongs to a measurement newly generated target, and determine a third LMB parameter set corresponding to each measurement newly generated target.

[0076] In this embodiment, the method of determining whether each target to be newly generated in the target to be newly generated set belongs to a measurement newly generated target based on the measurement set at the current moment, and determining a third LMB parameter set corresponding to each measurement newly generated target based on the second LMB parameter set, includes: obtaining the measurement set at the current moment, and respectively calculating the Mahalanobis distance between the pseudo-measurement of the kinematic state of each target in the target to be newly generated set and each measurement value in the measurement set to obtain corresponding first distance information; obtaining a first preset distance threshold (which can be obtained by chi-square distribution (χ 2 ) determines), and compares the obtained multiple first distance information corresponding to each of the to-be-generated targets with the first preset distance threshold respectively, to obtain the threshold comparison results corresponding to each of the to-be-generated targets respectively; calculates the measurement quantity threshold through the measurement rate information of each of the to-be-generated targets, and when the measurement quantity less than the preset distance threshold in the comparison result is greater than the measurement quantity threshold, it is determined that the corresponding to-be-generated target belongs to the measurement new target, and obtains the corresponding third LMB parameter set. It should be understood that from the probability density function of the second LMB parameter set (in, ) can be used to extract the measurement rate estimate for each target and kinematic state Specifically, assume that the measurement set obtained by the sensor at time k is:

[0077]

[0078] Calculate each measurement With each new target kinematic state Pseudo-measurement Mahalanobis distance (S is the new information covariance matrix). Under the extended target condition, when the Mahalanobis distance between a target to be newly created and multiple measurements is within the threshold, the target to be newly created is determined to be a measurement target. The specific number of measurements can be estimated by the measurement rate of each target to be newly created. The resulting measurement target is defined as the label space containing The corresponding third LBM parameter set is (in, ).in, represents the existence probability of a new target with label l, Represents the density function of the measured new target with label l (including the kinematic state information, extended morphology information, measurement rate information, etc. of the extended target).

[0079] Step S14: Based on the current moment, GGIW-LMB filtering is performed using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result.

[0080] In this embodiment, the use of the multi-target density including the third LMB parameter set to perform GGIW-LMB filtering to obtain the corresponding filtering result includes: obtaining the multi-target prior density at the current moment; wherein the multi-target prior density includes the corresponding new target density, the surviving target density and the third LMB parameter set as the measurement of the new target density; converting the LMB distribution of the multi-target prior density into a δ-GLMB distribution, and performing the GGIW-LMB filter update step based on the converted multi-target prior density and the measurement set at the current moment to obtain the multi-target posterior density at the current moment; by performing the corresponding LMB distribution conversion on the multi-target posterior density, the converted multi-target posterior density is obtained as the filtering result. wherein the new target density at time k is given by the prior information, expressed as a label space containing Parameter Set (in, ) of the LMB distribution, where ε γ (l) represents the probability of existence of a new target with label l, p γ (l) represents the density function of the new target with label l. The measured new target density at time k is generated by step S13 and is represented as The parameter set is (in, )’s LMB distribution.

[0081] It should be understood that the survival target density at time k is evolved from the posterior density of the target at time k-1. Assuming that the survival probability of the target is p s (·,l), then the predicted density of the surviving target is expressed as the label space containing Parameter Set (in, ) is the LMB distribution of

[0082]

[0083]

[0084] in, is a Gamma-Gaussian-Inverse Wishart mixed distribution. Then the multi-target prior density is expressed as the union of the three LMB distributions of the new target density, the measured new target density and the survival target density:

[0085]

[0086] At the same time, combined Figure 3As shown, in this embodiment, in order to calculate the multi-target posterior density, the LMB distribution of the prior density is first converted to a δ-GLMB (δ-Generalized Labeled Multi Bernoulli) distribution before subsequent calculations. Assume that the measurement set obtained by the sensor at time k is Z k (abbreviated as Z), the measurement set is divided into several subsets W, each subset W consists of several independent measurements z, that is, W = {z 1 ,…,z |W|}(|·| means to find the cardinality of the set, that is, the number of elements contained in the set). Update all prediction targets one by one as follows:

[0087] For each subset W of the measurement set, the single-target likelihood function is:

[0088]

[0089] In the formula, Indicates that the variable follows a Poisson distribution with a mean of λ. Assume that the prior density of a target is the GGIW mixed distribution defined above:

[0090]

[0091] Then, according to the Bayesian filtering formula, the posterior density of the target is calculated by the following formula:

[0092]

[0093] In the formula, represents the Kronecker product.

[0094] Assuming that the target measurement rate is independent of the target's kinematic state and extended form, the target measurement rate can be calculated separately from the target's kinematic state and extended form.

[0095] The updated calculation expression of the target measurement rate is:

[0096]

[0097] In the formula, the two terms on the right side of the equal sign are obtained after calculation on the left side of the equal sign, where It means that the updated measurement rate follows the shape parameter α W , the scale parameter is β W Gamma distribution of is the coefficient obtained after the operation, which is a function of the measurement subset W and the parameters ζ = (α, β, m, P, v, V).

[0098] The updated calculation expression of the joint distribution of the target kinematic state and the extended morphology is:

[0099]

[0100] In the formula, the three terms on the left side of the equal sign are calculated to obtain the three terms on the right side of the equal sign, among which, Indicates that the updated target kinematic state follows the mean value m W , the covariance is Gaussian distribution of It means that the updated extended form has degrees of freedom v W , the parameter matrix is ​​V W The inverse Wishart distribution of is the coefficient obtained after the operation, which is a function of the measurement subset W and the parameters ζ = (α, β, m, P, v, V).

[0101] By calculating the measurement update of each target one by one through the previous two equations and the subset of each measurement set, the parameters and weights of the GGIW mixed distribution can be obtained, which are brought into the update step expression of the extended target form δ-GLMB filter. The calculated multi-target posterior density at time k is expressed as:

[0102]

[0103] Among them, △(X) represents the label difference indicator, represents a discrete index set, represents all possible partitioning results of dividing the measurement set Z into i subsets, represents one of the possible partition results, Indicates the result of dividing multiple targets into one Finally, in order to transfer the posterior density to the next moment, the multi-target posterior density is converted into LMB distribution, which is expressed as:

[0104]

[0105] in, represents the updated target label set, ε (l) represents the probability of existence of the target with label l, p (l) Represents the posterior density function of the target with label l.

[0106] Step S15, based on each posterior density component and weight in the filtering result, corresponding target state information and target track information are obtained, and the updated target set to be generated at the current moment is obtained through reorganization.

[0107] Specifically, in this embodiment, (1) the posterior density components less than the preset weight threshold Th in the filtering result are first eliminated to obtain the corresponding eliminated result; (2) the components in the eliminated result are merged according to the density function of each component; (3) the component with the largest weight J is retained. max The posterior density components are used to obtain the estimated target state information and target track information. Further, regarding step (2), ① obtain the corresponding state x according to the density function of each component k|k , covariance matrix P k|k 、Inverse Wishart distribution parameter v k|k and V k|k (The subscript k|k indicates that the parameter is extracted from the posterior density function); ② Arrange the components from small to large according to the weight; ③ Take the component with the largest weight as the current merging center, and calculate the Mahalanobis distance between the merging center and other (here "other" refers to all the components to be merged after excluding the previously merged components) posterior density components based on the parameters obtained in step ②, and determine whether to merge by judging whether the distance is less than the corresponding threshold U; ④ The components merged in step ③ are no longer considered, and step ③ is repeated for each posterior density component in the remaining elimination result until all components have been merged. In other words, each merging operation selects the one with the largest weight as the merging center from the unmerged components remaining in the elimination result until all components are merged.

[0108] It should be understood that, in this embodiment, the set of targets to be generated after being updated at the current moment by reorganization includes: in the set of targets to be generated, for each target to be generated that does not belong to the measured new target determined in step S13, the existence time (i.e., kk d ) is greater than the preset time threshold (which can be used with k mb The target of (represented by) is eliminated to obtain the set of targets to be regenerated after the current moment update. Among them, the targets to be regenerated whose existence time is longer than the preset time threshold are the real extinct targets.

[0109] It can be seen that, in the case of extended targets being obscured in the target tracking scenario, the present application processes the first LMB parameter set constructed at the previous moment to obtain the set of new targets to be generated at the current moment and the corresponding second LMB parameter set, and then measures the new targets based on the measurement set, and obtains the third LMB parameter set corresponding to each of the measured new targets, and then performs filtering, and obtains the final target state information and target track information based on each posterior density component and weight in the filtering result. In this way, the continuity and accuracy of the track before and after the target is obscured can be effectively improved, so that the consistency of the track before and after the target is obscured can be maintained while ensuring the accuracy of the extended target state estimation.

[0110] As can be seen from the previous embodiment, after performing GGIW-LMB filtering, the present application obtains the corresponding track information based on the filtering result. To this end, the present embodiment will describe in detail the process of obtaining the track information. Figure 4 As shown, an embodiment of the present invention discloses a method for generating continuous tracks of multiple extended targets, including:

[0111] Step S21, obtaining a preset weight threshold, eliminating the posterior density components whose weights are less than the preset weight threshold in the filtering result, and obtaining the corresponding eliminated result.

[0112] Step S22: Obtain corresponding state information, covariance matrix and inverse Wishart distribution parameters based on each posterior density component in the elimination result.

[0113] Step S23: Arrange the posterior density components in the elimination result from small to large according to the weights to obtain the current component arrangement result.

[0114] Step S24, based on the component arrangement result, take the posterior density component with the largest current weight as the merging center, and calculate the Mahalanobis distance between the merging center and the current component to be merged by extracting the state information, the covariance matrix and the inverse Wishart distribution parameters corresponding to the merging center to obtain the corresponding second distance information.

[0115] Specifically, in this embodiment, when calculating the distance, it is assumed that the merging center is the i-th component, and the remaining components can be represented by j. The required formula is as follows:

[0116]

[0117]

[0118] Step S25: Compare each piece of the second distance information with a second preset distance threshold, and based on the comparison result, merge the posterior density components corresponding to the second distance information that is smaller than the second preset distance threshold.

[0119] Step S26: for each of the unmerged posterior density components in the eliminated result, arrange them in order from small to large according to the weight to obtain a new component arrangement result, and jump again to the step of taking the posterior density component with the largest current weight as the merging center based on the component arrangement result, until the merging operation of all the posterior density components is completed to obtain the corresponding component merging result.

[0120] Step S27: retain a preset number of target components with the largest weights in the component merging result, and obtain corresponding target state information and target track information through the target components.

[0121] In this embodiment, the number of targets in the scene at the current moment is estimated based on the weights of all LMB components. Then from the largest weight The target state and track (label) information are extracted from the posterior density components.

[0122] Among them, the specific process of the above steps S21 to S27 can refer to the corresponding content disclosed in the above embodiments, and will not be repeated here.

[0123] It can be seen that in the embodiment of the present application, the corresponding merging operation is completed by the weight of each posterior density component in the filtering result, which further reduces the number of components and improves the efficiency of track generation.

[0124] Combine the following Figures 5 to 7 The disclosed schematic diagram specifically illustrates the technical solution of the embodiment of the present application by setting up a typical simulation experiment scenario.

[0125] In this embodiment, a monitoring area of ​​[-1000, 1000](m)×[-1000, 1000](m) is considered. Figure 5 There are four moving extended targets in the scene shown, among which target 3 and target 4 pass through the occlusion area at the same time at time k=39-43, and then reappear in the monitoring area. In this embodiment, the parameters of the new targets with the same parameters are set as follows: α=10, β=1, v=10, V=diag([100,100]), P=diag([10,10]) 2 The kinematic states of the new targets are set as: m 1 =[-800,400,0,0],m 2 =[-800,550,0,0], m 3 =[-800,0,0,0], m 4 =[-550,-240,0,0].

[0126] In the current scenario, the target's survival probability is p S,k =0.99; the detection probability of the target is p D,k =0.98; the target measurement rate follows a Poisson distribution with a mean of γ = 15, the clutter in the scene is evenly distributed in the monitoring area, and the number at each moment follows a Poisson distribution with a mean of 20. Regarding the threshold for measuring the new typicality: the evolution probability is set to P MB =0.98, the threshold τ of the Mahalanobis distance is Measurement quantity The maximum allowed existence time of the extended target of extinction is set to k mb =10. In the pruning and merging operation, the pruning and merging thresholds of the components are Th=10 respectively. -5 and U = 4, the maximum number of components allowed is J max =10.

[0127] Figure 6 The track and centroid position estimation results of the disappeared extended target in this embodiment are given. Figure 7 The estimation result of the target extension form of this embodiment is given. Since the target identity information remains unchanged after the target is restarted due to short-term occlusion and measurement loss, it is not tracked as a new target. Therefore, the track remains continuous and unchanged with that before the occlusion. It can be seen from both figures that the tracks of the two targets at the bottom of the monitoring area are continuous and consistent before and after the occlusion, indicating that the algorithm does not misjudge the same target before and after the occlusion, forming a correct and continuous track.

[0128] See also Figure 8 As shown, the embodiment of the present application also discloses a device for generating continuous tracks of multiple extended targets, including:

[0129] A parameter set construction module 11 is used for constructing a set of extinct targets at a moment before the current moment and a corresponding first LMB parameter set for the extended target when an extended target is blocked in the target tracking scene;

[0130] A set acquisition module 12 is used to obtain a target set to be generated at a current moment and a corresponding second LMB parameter set based on the first LMB parameter set and the corresponding target motion characteristics;

[0131] The target judgment module 13 is used to judge whether each target in the target set to be newly generated belongs to a measurement newly generated target based on the measurement set at the current moment, and determine a third LMB parameter set corresponding to each of the measurement newly generated targets based on the second LMB parameter set;

[0132] A filtering result acquisition module 14 is used to perform GGIW-LMB filtering based on the current moment using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result;

[0133] The track generation module 15 is used to obtain corresponding target state information and target track information based on each posterior density component and weight in the filtering result, and obtain the updated target set to be generated at the current moment through reorganization.

[0134] Among them, for more specific working processes of the above-mentioned modules, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0135] It can be seen that the embodiment of the present application, in view of the situation where the extended target in the target tracking scene is blocked, processes the first LMB parameter set of the moment before the current moment to obtain the set of new targets to be generated at the current moment and the corresponding second LMB parameter set, and then measures the new targets based on the measurement set to determine, and obtains the third LMB parameter set corresponding to each of the measured new targets, and then performs filtering, and obtains the final target state information and target track information based on each posterior density component and weight in the filtering result. In this way, the continuity and accuracy of the track before and after the target is blocked can be effectively improved, so that the consistency of the track before and after the target is blocked can be maintained while ensuring the accuracy of the estimation of the extended target state.

[0136] In some specific embodiments, the parameter set construction module 11 may specifically include:

[0137] A state estimation result acquisition unit, used to acquire a first extended target state estimation result and a second extended target state estimation result corresponding to two moments before the current moment respectively;

[0138] a first LMB parameter set acquisition unit, configured to utilize the difference between the label sets at the first two moments to acquire corresponding extended target information through the first extended target state estimation result and the second extended target state estimation result, so as to obtain the extinction target set at the previous moment and the corresponding first LMB parameter set;

[0139] The extended target information includes kinematic state information, extended morphology information and measurement rate information.

[0140] In some specific embodiments, the set acquisition module 12 may specifically include:

[0141] a set merging unit, configured to obtain a corresponding merged target set and a corresponding LMB parameter set by merging the first LMB parameter set and the target LMB parameter set to be generated after being reorganized at a previous moment;

[0142] A preprocessing unit is used to preprocess the LMB parameter set corresponding to the merged target set based on the motion characteristics of the targets in the merged target set to obtain a target set to be newly generated at a current moment and a corresponding second LMB parameter set.

[0143] In some specific embodiments, the target determination module 13 may specifically include:

[0144] A first distance calculation unit is used to obtain a measurement set at a current moment, and respectively calculate the Mahalanobis distance between the pseudo measurement of the kinematic state of each target in the target set to be newly generated and each measurement value in the measurement set to obtain corresponding first distance information;

[0145] A threshold comparison unit, used to obtain a first preset distance threshold, and compare the obtained plurality of first distance information corresponding to each of the to-be-generated targets with the first preset distance threshold respectively, to obtain threshold comparison results corresponding to each of the to-be-generated targets respectively;

[0146] The third LMB parameter set acquisition unit is used to calculate the corresponding measurement quantity threshold value through the measurement rate information of each of the targets to be newly generated, and when the measurement quantity less than the preset distance threshold in the comparison result is greater than the measurement quantity threshold value, it is determined that the corresponding target to be newly generated belongs to the measurement newly generated target, and the corresponding third LMB parameter set is obtained.

[0147] In some specific embodiments, the filtering result acquisition module 14 may specifically include:

[0148] A multi-target prior density acquisition unit, used to acquire the multi-target prior density at the current moment; wherein the multi-target prior density includes the corresponding new target density, the surviving target density and the third LMB parameter set used to measure the new target density;

[0149] A filtering unit, configured to convert the LMB distribution of the multi-objective prior density into a δ-GLMB distribution, and perform a GGIW-LMB filter update step based on the converted multi-objective prior density and the measurement set at the current moment to obtain the multi-objective posterior density at the current moment;

[0150] The LMB distribution conversion unit is used to obtain the converted multi-objective posterior density as a filtering result by performing corresponding LMB distribution conversion on the multi-objective posterior density.

[0151] In some specific embodiments, the track generation module 15 may specifically include:

[0152] The target set updating unit is used to remove the targets whose existence time is longer than a preset time threshold from the target set that is not a measured target, and obtain the target set that is updated at the current moment and the corresponding LMB parameter set.

[0153] In some specific embodiments, the track generation module 15 may specifically include:

[0154] A component elimination unit, used to obtain a preset weight threshold, eliminate the posterior density components whose weights are less than the preset weight threshold in the filtering result, and obtain the corresponding eliminated result;

[0155] A parameter information acquisition unit, used to obtain corresponding state information, covariance matrix and inverse Wishart distribution parameters based on each of the posterior density components in the elimination result;

[0156] A component arrangement unit, used for arranging the posterior density components in the elimination result from small to large according to the weight, so as to obtain the current component arrangement result;

[0157] A second distance calculation unit is used to take the posterior density component with the largest current weight as the merging center based on the component arrangement result, and calculate the Mahalanobis distance between the merging center and the current component to be merged by extracting the state information, the covariance matrix and the inverse Wishart distribution parameter corresponding to the merging center to obtain the corresponding second distance information;

[0158] a component merging unit, configured to compare each piece of the second distance information with a second preset distance threshold, and based on the comparison result, merge the posterior density components corresponding to the second distance information that is smaller than the second preset distance threshold;

[0159] A circulation unit is used to arrange the unmerged posterior density components in the elimination result in order from small to large according to the weight to obtain a new component arrangement result, and jump to the step of taking the posterior density component with the largest current weight as the merging center based on the component arrangement result, until the merging operation of all the posterior density components is completed to obtain the corresponding component merging result;

[0160] The track information acquisition unit is used to retain a preset number of target components with the largest weights in the component merging result, and obtain corresponding target state information and target track information through the target components.

[0161] Furthermore, the present application also discloses an electronic device. Fig. 9 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

[0162] Fig. 9A schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the method for generating continuous tracks of multiple extended targets disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0163] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0164] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0165] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, and can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the method for generating continuous tracks of multiple extended targets executed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks.

[0166] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned disclosed method for generating continuous tracks of multiple extended targets is implemented. The specific steps of the method can be referred to the corresponding contents disclosed in the aforementioned embodiments, and will not be repeated here.

[0167] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0168] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0169] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0170] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0171] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for generating continuous tracks of multiple extended targets, characterized in that: include: When an extended target is blocked in the target tracking scene, a set of eliminated targets at a moment before the current moment and a corresponding first LMB parameter set are constructed for the extended target; Based on the first LMB parameter set and the corresponding target motion characteristics, a set of targets to be generated at the current moment and a corresponding second LMB parameter set are obtained; Based on the measurement set at the current moment, determine whether each target to be newly created in the target to be newly created set belongs to a measurement newly created target, and determine a third LMB parameter set corresponding to each measurement newly created target based on the second LMB parameter set; Based on the current moment, performing GGIW-LMB filtering using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result; Based on each posterior density component and weight in the filtering result, the corresponding target state information and target track information are obtained, and the updated target set to be generated at the current moment is obtained through reorganization.

2. The method for generating continuous tracks of multiple extended targets according to claim 1, characterized in that: The step of constructing a set of extinction targets at a moment before the current moment and a corresponding first LMB parameter set for the extended target includes: Obtain a first extended target state estimation result and a second extended target state estimation result corresponding to two moments before the current moment respectively; Utilizing the difference between the label sets at the first two moments, acquiring corresponding extended target information through the first extended target state estimation result and the second extended target state estimation result, so as to obtain the extinction target set at the previous moment and the corresponding first LMB parameter set; The extended target information includes kinematic state information, extended morphology information and measurement rate information.

3. The method for generating continuous tracks of multiple extended targets according to claim 1, characterized in that: The obtaining, based on the first LMB parameter set and the corresponding target motion characteristics, a target set to be generated at the current moment and a corresponding second LMB parameter set includes: By merging the first LMB parameter set and the target LMB parameter set to be generated after being reorganized at the previous moment, a corresponding merged target set and a corresponding LMB parameter set are obtained; Based on the motion characteristics of the targets in the merged target set, the LMB parameter set corresponding to the merged target set is preprocessed to obtain a target set to be newly generated at the current moment and a corresponding second LMB parameter set.

4. The method for generating continuous tracks of multiple extended targets according to claim 1, characterized in that: The step of judging whether each target to be newly created in the target to be newly created set belongs to a measurement newly created target based on the measurement set at the current moment, and determining a third LMB parameter set corresponding to each measurement newly created target based on the second LMB parameter set, includes: Acquire the measurement set at the current moment, and respectively calculate the Mahalanobis distance between the pseudo measurement of the kinematic state of each target in the target set to be newly generated and each measurement value in the measurement set to obtain corresponding first distance information; Obtaining a first preset distance threshold, and comparing the obtained plurality of first distance information corresponding to each of the to-be-generated targets with the first preset distance threshold respectively, to obtain threshold comparison results corresponding to each of the to-be-generated targets respectively; The corresponding measurement quantity threshold is calculated through the measurement rate information of each of the targets to be newly generated, and when the measurement quantity less than the first preset distance threshold in the comparison result is greater than the measurement quantity threshold, it is determined that the corresponding target to be newly generated belongs to the measurement newly generated target, and the corresponding third LMB parameter set is obtained.

5. The method for generating continuous tracks of multiple extended targets according to claim 1, characterized in that: The performing GGIW-LMB filtering using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result includes: Acquire the multi-target prior density at the current moment; wherein the multi-target prior density includes the corresponding new target density, the surviving target density and the third LMB parameter set used to measure the new target density; Convert the LMB distribution of the multi-objective prior density into a δ-GLMB distribution, and perform a GGIW-LMB filter update step based on the converted multi-objective prior density and the measurement set at the current moment to obtain the multi-objective posterior density at the current moment; By performing corresponding LMB distribution transformation on the multi-objective posterior density, the transformed multi-objective posterior density is obtained as a filtering result.

6. The method for generating continuous tracks of multiple extended targets according to claim 1, characterized in that: The corresponding target state information and target track information are obtained based on each posterior density component and weight in the filtering result, and the target set to be newly generated after being updated at the current moment is obtained by reorganization, including: Obtaining a preset weight threshold, eliminating the posterior density components whose weights are less than the preset weight threshold in the filtering results, and obtaining the corresponding eliminated results; Based on each of the posterior density components in the eliminated results, corresponding state information, covariance matrix and inverse Wishart distribution parameters are obtained; Arrange the posterior density components in the elimination result in order from small to large according to the weights to obtain the current component arrangement result; Based on the component arrangement result, the posterior density component with the largest current weight is taken as the merging center, and the Mahalanobis distance between the merging center and the current component to be merged is calculated by extracting the state information, the covariance matrix and the inverse Wishart distribution parameter corresponding to the merging center to obtain the corresponding second distance information; Comparing each of the second distance information with a second preset distance threshold, respectively, and based on the comparison result obtained, merging the posterior density components corresponding to the second distance information that is smaller than the second preset distance threshold; For each of the unmerged posterior density components in the elimination result, arrange them in order from small to large according to the weight to obtain a new component arrangement result, and jump again to the step of taking the posterior density component with the largest current weight as the merging center based on the component arrangement result, until the merging operation of all the posterior density components is completed to obtain the corresponding component merging result; A preset number of target components with the largest weights in the component merging results are retained, and corresponding target state information and target track information are obtained through the target components.

7. The method for generating continuous tracks of multiple extended targets according to any one of claims 1 to 6, characterized in that: The target set to be newly generated after being updated at the current moment obtained by reorganization includes: In the target set to be generated, for each target to be generated that does not belong to the measured target, targets whose existence time is longer than a preset time threshold are eliminated to obtain the target set to be generated updated at the current moment and the corresponding LMB parameter set.

8. A device for generating continuous tracks of multiple extended targets, characterized in that: include: A parameter set construction module, for constructing a set of extinct targets at a moment before the current moment and a corresponding first LMB parameter set for the extended target when an extended target is blocked in the target tracking scene; A set acquisition module, used to obtain a target set to be generated at the current moment and a corresponding second LMB parameter set based on the first LMB parameter set and the corresponding target motion characteristics; a target judgment module, configured to judge whether each target in the target set to be newly generated belongs to a measurement newly generated target based on the measurement set at the current moment, and determine a third LMB parameter set corresponding to each of the measurement newly generated targets based on the second LMB parameter set; A filtering result acquisition module, used to perform GGIW-LMB filtering based on the current moment using the multi-target density including the third LMB parameter set to obtain a corresponding filtering result; The track generation module is used to obtain the corresponding target state information and target track information based on the posterior density components and weights in the filtering results, and obtain the updated target set to be generated at the current moment through reorganization.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method for generating continuous tracks of multiple extended targets as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the method for generating continuous tracks of multiple extended targets as described in any one of claims 1 to 7.

Images