CN113296089B - LMB density fusion method and device for multi-early-warning-machine target tracking system - Google Patents

LMB density fusion method and device for multi-early-warning-machine target tracking system Download PDF

Info

Publication number
CN113296089B
CN113296089B CN202110852130.8A CN202110852130A CN113296089B CN 113296089 B CN113296089 B CN 113296089B CN 202110852130 A CN202110852130 A CN 202110852130A CN 113296089 B CN113296089 B CN 113296089B
Authority
CN
China
Prior art keywords
tag
sensor
local
fusion
association
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110852130.8A
Other languages
Chinese (zh)
Other versions
CN113296089A (en
Inventor
吴卫华
郑茂
黄伟平
孙合敏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Air Force Early Warning Academy
Original Assignee
Air Force Early Warning Academy
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Air Force Early Warning Academy filed Critical Air Force Early Warning Academy
Priority to CN202110852130.8A priority Critical patent/CN113296089B/en
Publication of CN113296089A publication Critical patent/CN113296089A/en
Application granted granted Critical
Publication of CN113296089B publication Critical patent/CN113296089B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/66—Radar-tracking systems; Analogous systems
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention provides an LMB density fusion method and device for a multi-early warning machine target tracking system, wherein the method comprises the following steps: 1) performing iteration initialization; 2) performing sequential LMB association on each node and adjacent nodes thereof, and performing coarse association and then fine association; 3) sequentially fusing each node and adjacent nodes thereof by LMB according to the results of the coarse correlation and the fine correlation; 4) after performing pair-by-pair sequential fusion on each node and adjacent nodes thereof according to the step 2) and the step 3), storing a final fusion result as input of next iteration; 5) when all nodes and their adjacent nodes are fused in sequence, the number of iterations is increased by 1, and the next iteration is entered. The invention comprehensively considers the problems of multi-view field and DBZ shielding, designs an effective LMB density fusion method, realizes the target of multi-early warning machine cooperative blind removal, and obviously improves the tracking performance compared with a local tracker.

Description

LMB density fusion method and device for multi-early-warning-machine target tracking system
Technical Field
The invention relates to the field of target tracking, in particular to an LMB density fusion method and device for a multi-early-warning-machine target tracking system.
Background
The presence of a Doppler Blind Zone (DBZ) is a major challenge for early warning aircraft radars, since early warning aircraft radars cannot obtain measurements from targets that are masked DBZ. Therefore, a series of missed detections may cause frequent switching of the track labels for a plurality of track segments, which seriously affects the tracking performance.
There are many ways to address the DBZ shadowing problem. However, most of these works are based on the framework of a conventional Random Vector (RV). In recent years, some studies have attempted to solve this problem under a Random Finite Set (RFS) framework. RFS-based methods can be broadly divided into two categories: unlabeled RFS (ULRFS unlabeled random finite set) and the nearest labeled RFS (LRFS random finite set). The former include Probability Hypothesis Density (PHD), potential PHD (cphd), and multi-bernoulli (MB) filters, and the latter include tag MB (LMB tag multi-bernoulli), generalized LMB (glmb), and marginalized glmb (mglmb) filters.
Since the relative velocity of the target is affected by the target-sensor relative geometry, a target hidden in DBZ of one sensor will not normally be masked by DBZ of another sensor at the same time. Therefore, complementary information from a plurality of early warning machine radars at different positions can be combined to cooperatively remove blindness, so that adverse effects caused by a visual field blind zone (VBZ) in a multi-view scene and a Doppler Blind Zone (DBZ) inherent to the early warning machine radars are reduced, and the tracking performance of the early warning machine is remarkably improved. Thus, using multi-sensor fusion techniques, the adverse effects of BZ (including VBZ and DBZ) shadowing can be greatly mitigated.
In LRFS density fusion, the problem of tag inconsistency is an unsolved problem. This problem is particularly acute in fully distributed fusion without a fusion center, which is further exacerbated when the VBZ and DBZ problems coexist.
To address this problem, the prior art provides a tag matching solution for geometric mean (GA) fusion. However, this approach is not suitable for multi-field fusion. Furthermore, the fusion result is influenced by the probability of presence (EP). To solve these two problems, the prior art proposes a multi-field AA fusion method by density correlation and density fusion. But when performing density correlation, an empirical value set manually is used as the correlation threshold. Furthermore, the problem of tag inconsistency is not considered in the implementation of density fusion.
Furthermore, the sensor fusion method described above does not consider the DBZ occlusion problem. The applicant has previously proposed a PHD fusion method suitable for DBZ occlusion, however, the fusion in this method is centralized, not distributed, and does not consider multi-field fusion. Since multi-view fusion is also a practical problem for multi-forewarning machine tracking, it is necessary to combine multi-view fusion and DBZ occlusion for multi-forewarning machine tracking. Based on the above consideration, the invention provides a full-distributed LRFS density fusion method comprehensively considering multi-field of view and DBZ shading.
Disclosure of Invention
The invention aims to provide an LMB density fusion method and device for a multi-early warning machine target tracking system, and aims to solve the problem that multi-field and DBZ shielding are not comprehensively considered in the conventional multi-early warning machine target tracking.
The invention is realized by the following steps:
in a first aspect, the present invention provides an LMB density fusion method for a multi-warning-machine target tracking system, where the system includes a sensor network composed of multiple sensor nodes, each node is equipped with a warning-machine radar, and the warning-machine radar of each node operates a multi-target tracker, the method includes the following steps:
1) performing iteration initialization;
2) sequentially associating each node with the adjacent node thereof to an LMB, performing coarse association and then performing fine association, wherein the coarse association is used for screening out unique components which are not associated with a remote sensor in a local sensor and unique components which are not associated with the local sensor in the remote sensor, and the fine association is used for screening out common components which are associated with the local sensor and the remote sensor and determining an association relationship;
3) and sequentially fusing each node and the adjacent nodes thereof by the LMB according to the results of the coarse correlation and the fine correlation, wherein the steps comprise: for the associated public components, adopting consistency fusion processing; for the unique component which is not associated, judging whether the unique component falls into the BZ of another sensor, if the unique component is out of the BZ of another sensor, not participating in subsequent association fusion, and if the unique component is in the BZ of another sensor, directly adding the unique component into a fusion result;
4) after performing pair-by-pair sequential fusion on each node and adjacent nodes thereof according to the step 2) and the step 3), storing a final fusion result as input of next iteration;
5) when all nodes and their adjacent nodes are fused in sequence, the number of iterations is increased by 1, and the next iteration is entered.
Further, each node of the system runs a tag random finite set filter, the output of the tag random finite set filter is in a Gaussian mixture form, and the specific process of the coarse association is as follows:
local sensor
Figure 943157DEST_PATH_IMAGE001
Middle LBC tag
Figure 125877DEST_PATH_IMAGE002
And a remote sensor
Figure 108877DEST_PATH_IMAGE003
Middle LBC tag
Figure 972927DEST_PATH_IMAGE004
The spatial density of the GLMB components is in a Gaussian mixture form, and the spatial density is respectively as follows:
Figure 52879DEST_PATH_IMAGE005
wherein the content of the first and second substances,
Figure 344183DEST_PATH_IMAGE006
representing local sensors
Figure 814478DEST_PATH_IMAGE007
Middle LBC tag
Figure 278958DEST_PATH_IMAGE008
The spatial density of the GLMB component of (a),
Figure 416678DEST_PATH_IMAGE009
representing spatial density
Figure 613304DEST_PATH_IMAGE006
The total number of gaussian components required,
Figure 633213DEST_PATH_IMAGE010
represents the weight of the mixture of gaussians,
Figure 839066DEST_PATH_IMAGE011
represents a mean and a covariance of
Figure 565714DEST_PATH_IMAGE012
Is distributed over the positive-negative ratio of (c),
Figure 464400DEST_PATH_IMAGE013
representing remote sensors
Figure 971604DEST_PATH_IMAGE014
Middle LBC tag
Figure 449990DEST_PATH_IMAGE015
The spatial density of the GLMB component of (a),
Figure 562303DEST_PATH_IMAGE016
representing spatial density
Figure 163048DEST_PATH_IMAGE013
The total number of gaussian components required,
Figure 95232DEST_PATH_IMAGE017
represents the weight of the mixture of gaussians,
Figure 642888DEST_PATH_IMAGE018
represents a mean and a covariance of
Figure 406445DEST_PATH_IMAGE019
Normal distribution of (2);
the mean and covariance of the weighted Gaussian components are respectively
Figure 646933DEST_PATH_IMAGE020
And
Figure 800834DEST_PATH_IMAGE021
；
if the difference between the two mean values
Figure 948919DEST_PATH_IMAGE022
Obey card square
Figure 504665DEST_PATH_IMAGE023
Distribution, namely:
Figure 712793DEST_PATH_IMAGE024
wherein
Figure 353989DEST_PATH_IMAGE025
Is the probability of the door,
Figure 243448DEST_PATH_IMAGE026
is the dimension of the state vector and,
Figure 981597DEST_PATH_IMAGE027
；
then correspond to
Figure 298309DEST_PATH_IMAGE028
Is marked with a label
Figure 489119DEST_PATH_IMAGE002
And correspond to
Figure 916689DEST_PATH_IMAGE029
Is marked with a label
Figure 447027DEST_PATH_IMAGE004
They are associated, otherwise they are not.
Further, the specific process of the fine correlation is as follows:
binary allocation variable as defined below
Figure 996957DEST_PATH_IMAGE030
：
Figure 878326DEST_PATH_IMAGE031
In calculating the association cost, the following association metrics are utilized
Figure 109587DEST_PATH_IMAGE032
：
Figure 556749DEST_PATH_IMAGE033
Label (R)
Figure 949684DEST_PATH_IMAGE002
And a label
Figure 318348DEST_PATH_IMAGE004
Is expressed as
Figure 150038DEST_PATH_IMAGE034
Which is a correlation metric
Figure 654969DEST_PATH_IMAGE035
The negative logarithm of (d), i.e.:
Figure 281122DEST_PATH_IMAGE036
solving for optimal distribution variable
Figure 137083DEST_PATH_IMAGE037
Cause local sensors
Figure 710147DEST_PATH_IMAGE001
Each track of (a) is assigned to a remote sensor
Figure 866322DEST_PATH_IMAGE003
A certain track or virtual track, and a remote sensor
Figure 866639DEST_PATH_IMAGE003
Is assigned to a local sensor
Figure 944316DEST_PATH_IMAGE001
To a certain track or virtual track, i.e. solving the following two-dimensional distribution problem:
Figure 383388DEST_PATH_IMAGE038
in the formula:
Figure 597331DEST_PATH_IMAGE039
and
Figure 830866DEST_PATH_IMAGE040
are respectively a sensor
Figure 395840DEST_PATH_IMAGE001
And a sensor
Figure 310706DEST_PATH_IMAGE003
A labelset of LMB density of;
then, two tag sets are implemented by finding the best allocation
Figure 441473DEST_PATH_IMAGE039
And
Figure 783593DEST_PATH_IMAGE040
a fine association between them.
Further, the consistency fusion processing is adopted for the associated common components, and specifically includes:
performing GCI fusion on the fused spatial densities for the LMB densities associated with the local sensors and the remote sensors;
the fused EP is processed according to whether the associated LMB density falls within the BZ of another sensor as follows: if the local LBC falls within the BZ of the remote sensor and the remote LBC does not fall within the BZ of the local sensor, the EP of the fused LBC is taken as the EP of the local LBC; conversely, if the local LBC does not fall within the BZ of the remote sensor, and the remote LBC falls within the BZ of the local sensor, then the EP of the fused LBC is taken as the EP of the remote LBC.
Further, the method further comprises:
for the associated common component, the fusion tag employs a local tag.
Further, the method further comprises:
for the associated common component, the corresponding tag association history is the tag and corresponding sensor index of the remote LBC added to the tag association history of the local LBC.
Further, the determining whether the unassociated unique component falls into the BZ of another sensor, if it is outside the BZ of another sensor, then not participating in subsequent association fusion, and if it is inside the BZ of another sensor, then directly adding to the fusion result, specifically includes:
for each unassociated local LBC obtained from the coarse association, determining whether it falls within the BZ of the remote sensor, and if it falls within the BZ of the remote sensor and has a high EP, adding it directly to the local fusion result;
for each unassociated remote LBC obtained from the coarse association, it is determined whether it falls within the BZ of the local sensor, if it falls within the BZ of the local sensor and has a high EP, it is also added to the local fusion result, and the corresponding fusion tag and tag association history are rewritten according to local criteria that the fused track tag is on the local platform.
Further, the rewriting of the corresponding fusion tag and tag association history according to the local criterion specifically includes:
retrieving a tag association history of the remote LBC, if tags related to the local sensor index in the tag association history of the remote LBC are not empty, using the tags most associated with the local sensor index as final fusion tags, and attaching the currently associated tags and the remote sensor index as a new tag association history to the current tag association history; otherwise, if the label related to the local sensor index in the label association history of the remote LBC is null, retrieving the label library of all the local LBCs, and according to the time component in the label of the remote LBC
Figure 632600DEST_PATH_IMAGE041
The maximum number of targets generated at this time
Figure 351158DEST_PATH_IMAGE042
Increase by 1 to obtain the final fusion tag
Figure 539694DEST_PATH_IMAGE043
I.e. by
Figure 318294DEST_PATH_IMAGE044
Further, the currently associated tag and the remote sensor index are combined into a first tag association history.
In a second aspect, the present invention further provides an LMB density fusion apparatus for a target tracking system of multiple early warning machines, where the apparatus includes:
a processor;
a memory having stored thereon a computer program operable on the processor;
wherein the computer program when executed by the processor implements the steps of the multi-forewarning machine target tracking LMB density fusion method as described in any one of the above.
In a third aspect, the present invention further provides a computer-readable storage medium, where a data processing program is stored, and when the data processing program is executed by a processor, the data processing program implements the steps of the LMB density fusion method for multi-warning-machine target tracking described in any one of the above.
Compared with the prior art, the invention has the following beneficial effects:
the LMB density fusion method and device for the multi-early warning machine target tracking system comprehensively consider the problems of multi-view field and DBZ shielding, and an effective LMB density fusion method is designed, and comprises two important parts: the method comprises the following steps of first coarse association and then fine association, wherein the coarse association is used for screening out unique components which are not associated with other sensors, and the fine association is used for selecting common components (association components) which are associated with other sensors and determining association relations; and secondly, consistency fusion of combination and fusion, wherein in the fusion step, related public components are fused by adopting an improved GCI fusion rule, and in the combination step, unique components which do not belong to BZ are added into the fusion result so as to avoid attenuation caused by the GCI fusion rule. Simulation results show that the fusion method effectively achieves the goal of cooperative blind removal of multiple early warning machines, and compared with a local tracker, the tracking performance is remarkably improved.
Drawings
Fig. 1 is a flowchart of an LMB density fusion method for a target tracking system of multiple early warning machines according to an embodiment of the present invention;
FIG. 2 is a graph of a target slope distance and corresponding VBZ versus time provided by an embodiment of the present invention;
FIG. 3 is a graph of absolute Doppler and corresponding DBZs versus time for a target provided by an embodiment of the present invention;
FIG. 4 is a schematic view of statistical OSPA-T of various sensors according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The multi-early warning machine target tracking system comprises a sensor network consisting of a plurality of sensor nodes, wherein each node is provided with a maximum detection range
Figure 654597DEST_PATH_IMAGE045
The same early warning machine radar, each node runs a multi-target tracker for tracking the target of interest, and each node has a range
Figure 911266DEST_PATH_IMAGE046
An isotropic wireless communication module, provided that the module is the same for all nodes and is
Figure 16625DEST_PATH_IMAGE047
。
For each node
Figure 700548DEST_PATH_IMAGE001
The adjacent nodes with communication relation are called nodes
Figure 461830DEST_PATH_IMAGE001
Remote node of (2), node
Figure 584507DEST_PATH_IMAGE001
Known per se as local node, node
Figure 482056DEST_PATH_IMAGE001
The early warning machine radar of (1) is called a local sensor, and the early warning machine radar of the remote node is called a remote sensor. Distributed fusion is considered here, and therefore, each node
Figure 602459DEST_PATH_IMAGE048
Is transmitted to all its remote nodes and the node is
Figure 913354DEST_PATH_IMAGE001
Local posterior probabilities for these remote nodes are also collected. Then, the fusion engine is used for fusing the remote posterior values with the local posterior values to obtain fused posterior values which serve as nodes
Figure 511826DEST_PATH_IMAGE001
Prior information for the next iteration of filtering.
More specifically, each node independently performs the following functions in a parallel, fully distributed and scalable manner:
1) joint prediction and update of the GLMB filter;
2) GLMB2LMB, i.e., GLMB density is converted to LMB density;
3) performing consistent GA fusion on the converted LMB density;
4) LMB2GLMB, i.e. LMB density is converted to GLMB density;
5) multi-target state estimation (as system output);
6) adaptive track initiation based on multi-target state estimation and current measurements.
The transformed GLMB density obtained in step 4) and the birth GLMB density obtained in step 6) are sent to step 1) for entering the next cycle at the next time using imperfect measurements (i.e. presence of missing detections and clutter).
The above fusion framework shows that the main body of the local tracker is the GLMB filter. In the fusion process, firstly, the GLMB density output by the filter is converted into an LMB density form, and then the LMB densities of different nodes are fused by utilizing a GCI rule. The fused LMB is converted back to the GLMB form so that the local GLMB filter continues.
As shown in fig. 1, an embodiment of the present invention provides an LMB density fusion method for a target tracking system of multiple early warning machines, where the method includes the following steps:
1) iteration initialization:
Figure 326198DEST_PATH_IMAGE049
；
2) sequentially associating each node with the adjacent node thereof to an LMB, performing coarse association and then performing fine association, wherein the coarse association is used for screening out unique components which are not associated with a remote sensor in a local sensor and unique components which are not associated with the local sensor in the remote sensor, and the fine association is used for screening out common components which are associated with the local sensor and the remote sensor and determining an association relationship;
3) and sequentially fusing each node and the adjacent nodes thereof by the LMB according to the results of the coarse correlation and the fine correlation, wherein the steps comprise: for the associated public components, adopting consistency fusion processing; for the unique component which is not associated, judging whether the unique component falls into the BZ of another sensor, if the unique component is out of the BZ of another sensor, not participating in subsequent association fusion, and if the unique component is in the BZ of another sensor, directly adding the unique component into a fusion result;
4) after performing pair-by-pair sequential fusion on each node and adjacent nodes thereof according to the step 2) and the step 3), storing a final fusion result as input of next iteration;
5) when all nodes and their neighbors are fused in sequential pairs, the number of iterations is increased by 1, (i.e., the number of iterations increases
Figure 617502DEST_PATH_IMAGE050
) And entering the next iteration.
The above steps are explained in detail below.
To determine tag associations in LMB density, GCI divergence is typically used as a tag inconsistency indicator. Tag association cost for LMB density for two sensor case
Figure 353377DEST_PATH_IMAGE051
Comprises the following steps:
Figure 817856DEST_PATH_IMAGE052
wherein the content of the first and second substances,
Figure 424418DEST_PATH_IMAGE053
Figure 948940DEST_PATH_IMAGE054
Figure 906532DEST_PATH_IMAGE055
wherein the content of the first and second substances,
Figure 112386DEST_PATH_IMAGE056
representative sensor
Figure 901350DEST_PATH_IMAGE007
Middle LBC tag
Figure 534457DEST_PATH_IMAGE008
The probability of the existence of (a) is,
Figure 776082DEST_PATH_IMAGE057
representative sensor
Figure 520047DEST_PATH_IMAGE007
The fusion weight of (a) is calculated,
Figure 620641DEST_PATH_IMAGE058
representative sensor
Figure 221387DEST_PATH_IMAGE014
Middle LBC tag
Figure 153570DEST_PATH_IMAGE015
The probability of the existence of (a) is,
Figure 497964DEST_PATH_IMAGE059
representative sensor
Figure 199204DEST_PATH_IMAGE060
The fusion weight of (a) is calculated,
Figure 174113DEST_PATH_IMAGE061
representing local sensors
Figure 655910DEST_PATH_IMAGE007
Middle LBC tag
Figure 741678DEST_PATH_IMAGE008
The spatial density of the GLMB component of (a),
Figure 359741DEST_PATH_IMAGE062
representing remote sensors
Figure 505551DEST_PATH_IMAGE014
Middle LBC tag
Figure 412328DEST_PATH_IMAGE015
The spatial density of the GLMB component of (a).
From this point forward, for convenience of presentation,
Figure 98524DEST_PATH_IMAGE001
and
Figure 774356DEST_PATH_IMAGE003
respectively, a local sensor index and a remote sensor index, and
Figure 887805DEST_PATH_IMAGE002
or
Figure 16298DEST_PATH_IMAGE004
Representing tags from the local sensor or the tag bernoulli component (LBC) in the remote sensor.
The method works well when all nodes have the same field of view and a higher probability of detection. However, it is not applicable to multi-sensor fusion with different fields of view, because in this case, the Bernoulli Component (BC) within one sensor's unique field of view would not be associated with any BC in the other sensors; furthermore, since the GCI divergence contains the probability of presence (EP), it is strongly influenced by the probability of presence of BC. On the other hand, the EP is closely related to whether or not the target is detected. Therefore, its performance under low detection probability and BZ masking will deteriorate significantly.
In view of the above problems, the embodiment of the present invention provides a two-step association method that performs "coarse association" and then "fine association". In this method, the correlation metric used to calculate the correlation cost matrix is related only to spatial density and not to EP and fusion weights, and therefore the correlation threshold is automatically calculated.
In the step (2), the specific process of coarse association is as follows:
each node runs a tag random finite set filter, which in this embodiment is a MM-GLMB-DBZ (multi-model generalized tag-multi-bernoulli) filter, the output of which is in the form of a Gaussian Mixture (GM). Local sensor
Figure 709448DEST_PATH_IMAGE001
Middle LBC tag
Figure 302103DEST_PATH_IMAGE002
And a remote sensor
Figure 789716DEST_PATH_IMAGE003
Middle LBC tag
Figure 467822DEST_PATH_IMAGE004
The spatial density of the GLMB components of (a) is in the form of Gaussian Mixture (GM) and is:
Figure 699084DEST_PATH_IMAGE005
formula (1)
Wherein the content of the first and second substances,
Figure 83929DEST_PATH_IMAGE061
representing local sensors
Figure 804760DEST_PATH_IMAGE007
Middle LBC tag
Figure 907845DEST_PATH_IMAGE008
The spatial density of the GLMB component of (a),
Figure 739535DEST_PATH_IMAGE063
representing spatial density
Figure 244466DEST_PATH_IMAGE061
The total number of gaussian components required,
Figure 73881DEST_PATH_IMAGE064
represents the weight of the mixture of gaussians,
Figure 726579DEST_PATH_IMAGE065
represents a mean and a covariance of
Figure 34064DEST_PATH_IMAGE066
Is distributed over the positive-negative ratio of (c),
Figure 659080DEST_PATH_IMAGE062
representing remote sensors
Figure 721714DEST_PATH_IMAGE014
Middle LBC tag
Figure 533813DEST_PATH_IMAGE015
The spatial density of the GLMB component of (a),
Figure 972884DEST_PATH_IMAGE067
representing spatial density
Figure 186828DEST_PATH_IMAGE062
The total number of gaussian components required,
Figure 420363DEST_PATH_IMAGE068
represents the weight of the mixture of gaussians,
Figure 985337DEST_PATH_IMAGE069
represents a mean and a covariance of
Figure 634624DEST_PATH_IMAGE019
Is normally distributed.
Furthermore, if the GLMB is in GM form, after GLMB2LMB conversion, the converted LMB is also in GM form.
For the GM form of LMB density, the remote sensor is roughly screened for all tag bernoulli components (LBCs) that are not related to LBCs. This serves two purposes: firstly, the dimensionality of assignment problems involved in subsequent fine correlation is reduced, so that the computational complexity is reduced; second, it is easy to screen out the unique components in each sensor caused by BZ (including DBZ and VBZ).
The phenomenon of non-association is common due to the presence of BZ. For example, if the target is within the BZ of sensor 1, but outside the BZ of sensor 2, there may be LBC corresponding to the target within the LMB density of sensor 2, and there may be no component corresponding to the target in sensor 1, at which time a non-correlation phenomenon will occur.
For this reason, without loss of generality, it is assumed that the weights of the two are the largestThe mean and covariance of the Gaussian components are respectively
Figure 30970DEST_PATH_IMAGE020
And
Figure 373090DEST_PATH_IMAGE021
；
if both correspond to the same target, the difference between the two mean values
Figure 222097DEST_PATH_IMAGE022
Will obey the chi-square
Figure 940654DEST_PATH_IMAGE023
Distribution, namely:
Figure 129190DEST_PATH_IMAGE024
formula (2)
Wherein
Figure 704528DEST_PATH_IMAGE025
Is the probability of the door,
Figure 978514DEST_PATH_IMAGE026
is the dimension of the state vector and,
Figure 297500DEST_PATH_IMAGE027
；
therefore, if the above condition is satisfied, it corresponds to
Figure 340543DEST_PATH_IMAGE028
Is marked with a label
Figure 290044DEST_PATH_IMAGE002
And correspond to
Figure 848064DEST_PATH_IMAGE029
Is marked with a label
Figure 908424DEST_PATH_IMAGE004
They are associated, otherwise they are not. The above association process is referred to as "coarse association".
After the coarse correlation, the resulting correlated LMB density is further processed by a so-called fine correlation. The fine correlation uses a classical allocation algorithm. Typically, density and density correlation is performed between all neighboring nodes. To avoid the combinatorial complexity of n-dimensional (n-D) assignments, pairwise associations (i.e., two-dimensional assignments) are employed.
In the step (2), the precise association process is as follows:
binary allocation variable as defined below
Figure 868290DEST_PATH_IMAGE030
：
Figure 988693DEST_PATH_IMAGE031
Formula (3)
Furthermore, association metrics need to be defined. As previously mentioned, the GCI divergence correlation metric is influenced by the EP and the fusion weight, affecting the correctness of the correlation. In fact, the core indicator of whether the densities of different sensors are correlated should relate only to the spatial (location) density. Thus, in calculating the association cost, the following association metric is utilized
Figure 237272DEST_PATH_IMAGE032
：
Figure 898060DEST_PATH_IMAGE033
Formula (4)
Substitution of formula (1) into formula (4) yields:
Figure 650115DEST_PATH_IMAGE070
formula (5)
Therefore, the density correlation metric is not only suitable for equally processing densities from different sensors due to its symmetry, but also has significant advantages: the GM form density transmitted over the radio link may be analytically calculated.
Label (R)
Figure 3736DEST_PATH_IMAGE002
And a label
Figure 739611DEST_PATH_IMAGE004
Is expressed as
Figure 876194DEST_PATH_IMAGE034
Which is a correlation metric
Figure 810652DEST_PATH_IMAGE035
The negative logarithm of (d), i.e.:
Figure 272858DEST_PATH_IMAGE036
formula (6)
The final aim being to solve for the optimal distribution variable
Figure 27187DEST_PATH_IMAGE037
Cause local sensors
Figure 498620DEST_PATH_IMAGE001
Each track of (a) is assigned to a remote sensor
Figure 225267DEST_PATH_IMAGE003
A certain track or virtual track, and a remote sensor
Figure 655112DEST_PATH_IMAGE003
Is assigned to a local sensor
Figure 99999DEST_PATH_IMAGE001
To a certain track or virtual track, i.e. solving the following two-dimensional distribution problem:
Figure 906281DEST_PATH_IMAGE038
formula (7)
In the formula:
Figure 753015DEST_PATH_IMAGE039
and
Figure 557023DEST_PATH_IMAGE040
are respectively a sensor
Figure 285944DEST_PATH_IMAGE001
And a sensor
Figure 833600DEST_PATH_IMAGE003
A labelset of LMB density of;
then, two tag sets are implemented by finding the best allocation
Figure 597157DEST_PATH_IMAGE039
And
Figure 306487DEST_PATH_IMAGE040
a fine association between them. The optimization problem can be solved in polynomial time using the hungarian algorithm.
It should be noted that the correlation cost matrix is extended to accommodate the non-correlated case. Virtual (counterfeit) labels
Figure 788284DEST_PATH_IMAGE071
Or
Figure 139631DEST_PATH_IMAGE072
Has been added to
Figure 695377DEST_PATH_IMAGE073
And
Figure 637925DEST_PATH_IMAGE074
to accommodate disassociations to allow sensors
Figure 279122DEST_PATH_IMAGE001
Is marked with a label
Figure 230898DEST_PATH_IMAGE002
Unassociated sensor
Figure 906730DEST_PATH_IMAGE003
And vice versa. For the unassociated case, assume that one track is not detected and the other track is in volume
Figure 223441DEST_PATH_IMAGE075
Are uniformly distributed in the monitored area, then
Figure 414251DEST_PATH_IMAGE076
Wherein the content of the first and second substances,
Figure 841822DEST_PATH_IMAGE077
which represents the probability of detection by the sensor,
Figure 434477DEST_PATH_IMAGE078
representing the volume of the surveillance zone.
The density on the correlation needs to be fused. Without loss of generality, considering the two sensor case, a fused EP can be obtained by the following method
Figure 922090DEST_PATH_IMAGE079
And spatial density
Figure 537879DEST_PATH_IMAGE080
Figure 831457DEST_PATH_IMAGE081
Formula (8)
Figure 216302DEST_PATH_IMAGE082
Wherein the content of the first and second substances,
Figure 937134DEST_PATH_IMAGE083
Figure 40219DEST_PATH_IMAGE084
formula (9)
Figure 871909DEST_PATH_IMAGE085
(symbol)
Figure 111260DEST_PATH_IMAGE086
Presentation label
Figure 206255DEST_PATH_IMAGE002
And
Figure 593374DEST_PATH_IMAGE004
corresponding to the same target, otherwise
Figure 166438DEST_PATH_IMAGE087
。
And performing different fusion processing on the non-correlation component obtained by coarse correlation and the correlation component obtained by fine correlation. Specifically, a "fusion" plus "combination" strategy is adopted. The general idea is as follows: for the relevant components (i.e. the common part of the two sensors), a consistent fusion process is used, i.e. a "fusion" method is used; on the other hand, for non-correlated components (i.e. unique parts of both sensors), in order to ensure that the sensors do not lose targets, they do not participate in the consistent fusion, but rather are added directly to the resulting fusion result, i.e. using a "combined" approach. For the sensor
Figure 588192DEST_PATH_IMAGE001
Repeating the strategy of 'fusing' and 'combining' with the fusing density of a certain adjacent node until all the adjacent nodes are processed.
Notwithstanding the above factsThe real track is slightly ensured not to be missed as much as possible, but some side effects are brought: more false tracks may be generated. However, loss of track poses a greater hazard than false alarms. Moreover, for a general false alarm, it will automatically filter out and disappear after a period of time. Therefore, it makes sense to adopt such a strategy in order to avoid the loss of the target as much as possible. On the other hand, as a remedy, in order to reduce side effects, before the unassociated components are added directly to the fused result, they first need to be identified according to whether they fall into the BZ (DBZ + VBZ) of the "other" sensor, in order to avoid being added erroneously to the fused result, thereby reducing false tracks as much as possible. In particular, when the sensor is used
Figure 854088DEST_PATH_IMAGE001
When a density is not relevant, it will be distinguished according to whether it falls into BZ: if it is outside the BZ, then it does not participate in the subsequent association fusion; if it is within BZ, it is added directly to the fused result.
In the step (3), for the associated common component, a consistency fusion process is adopted, which specifically includes:
performing GCI fusion on the LMB density associated with the local sensor and the remote sensor and the fused space density, wherein the fusion weight is Metropolis weight;
specifically, for correlated LMB densities, assuming that the LB density from the local tracker is in the form of a gaussian mixture, a fused spatial density may be obtained
Figure 666186DEST_PATH_IMAGE088
Is composed of
Figure 839679DEST_PATH_IMAGE090
Wherein the content of the first and second substances,
Figure 53622DEST_PATH_IMAGE091
Figure 287158DEST_PATH_IMAGE092
Figure 586552DEST_PATH_IMAGE093
Figure 766998DEST_PATH_IMAGE094
Figure 163344DEST_PATH_IMAGE095
processing the fused EP according to whether the associated LMB density falls within the BZ of another sensor, as follows: if the local LBC falls within the BZ of the remote sensor and the remote LBC does not fall within the BZ of the local sensor, the EP of the fused LBC is taken as the EP of the local LBC; conversely, if the local LBC does not fall within the BZ of the remote sensor, and the remote LBC falls within the BZ of the local sensor, then the EP of the fused LBC is taken as the EP of the remote LBC; for other cases, the calculation is based on the fusion of local EP's.
The fused EP is calculated specifically according to the following strategy:
(a) if there is only local LBC
Figure 505463DEST_PATH_IMAGE002
At remote sensor
Figure 88892DEST_PATH_IMAGE003
In BZ (meaning with local LBC)
Figure 807449DEST_PATH_IMAGE002
Corresponding remote LBC
Figure 58302DEST_PATH_IMAGE004
Will be smaller) or if there is only a remote LBC
Figure 836902DEST_PATH_IMAGE004
At local sensors
Figure 845309DEST_PATH_IMAGE001
In BZ (meaning with remote LBC)
Figure 164295DEST_PATH_IMAGE004
Corresponding local LBC
Figure 219056DEST_PATH_IMAGE002
Will be smaller), in order to avoid the attenuation caused by the fusion of the EP (equation (8)), the fused EP is modified to
Figure 965295DEST_PATH_IMAGE096
Or
Figure 992157DEST_PATH_IMAGE097
Wherein
Figure 786938DEST_PATH_IMAGE098
And
Figure 12383DEST_PATH_IMAGE099
respectively the average of the gaussian components with the largest weight,
Figure 867206DEST_PATH_IMAGE100
an EP representing the fusion after the modification,
Figure 912523DEST_PATH_IMAGE101
the indication function is represented by a representation of,
Figure 510994DEST_PATH_IMAGE102
represents
Figure 794208DEST_PATH_IMAGE098
Drop-in remote sensor
Figure 882250DEST_PATH_IMAGE003
In the above-mentioned BZ, the above-mentioned,
Figure 352545DEST_PATH_IMAGE103
represents
Figure 20287DEST_PATH_IMAGE099
Drop-in local sensor
Figure 954745DEST_PATH_IMAGE001
In BZ of (1).
(b) Otherwise, according to equation (9), the fused EP is
Figure 151371DEST_PATH_IMAGE104
Unlike equation (9), the effect of the Chernoff coefficient on the fused EP is avoided.
Thirdly, the fusion tag adopts a local tag;
it should be noted that for the fusion tag
Figure 171280DEST_PATH_IMAGE105
The local platform is taken as the standard, and the following local standard is followed. In other words, for an associated local tag
Figure 377133DEST_PATH_IMAGE002
And remote tag
Figure 369360DEST_PATH_IMAGE004
The fusion label adopts a local label
Figure 799204DEST_PATH_IMAGE002
I.e. by
Figure 509671DEST_PATH_IMAGE106
. This is the label fusion rule of the associated label.
And fourthly, adding the label of the remote LBC and the corresponding sensor index into the label association history of the local LBC.
In addition, to determine subsequent fused tags corresponding to unassociated LBCs, a new tag association history is added for the fused LBCs. Tag association history
Figure 253636DEST_PATH_IMAGE107
Is to be associated with the local tag
Figure 100370DEST_PATH_IMAGE002
Associated remote tags
Figure 966694DEST_PATH_IMAGE004
And remote sensor indexing
Figure 633299DEST_PATH_IMAGE003
Appending a previous tag association history from a previous time
Figure 243272DEST_PATH_IMAGE108
After that, namely
Figure 678933DEST_PATH_IMAGE109
Executing the trimming and merging steps to reduce the storage and calculation capacity required by the subsequent fusion operation.
In the step (3), adding the non-associated unique component to the obtained fusion result specifically includes:
for each unassociated local LBC obtained by the coarse association (i.e., it is unassociated with all LBCs of the remote sensor), it is determined whether it falls within the BZ of the remote sensor, if it falls within the BZ of the remote sensor, and has a high EP (e.g., greater than some threshold value)
Figure 919421DEST_PATH_IMAGE110
) This means that it may be a high-lying track in the remote sensor BZ, and then straightens itAdding the fusion result into a local fusion result;
for each unassociated remote LBC obtained by the coarse association (i.e. it is not associated with all LBCs of the local sensor), it is determined whether it falls within the BZ of the local sensor, if it falls within the BZ of the local sensor and has a high EP, which means that it may be a high-lying track in the BZ of the local sensor, it is also added to the local fusion result, however, unlike the processing of the local unassociated components, the "unseen" local sensor receives the tags sent by the other "seen" remote sensors, which need to be redefined in the local sensor, which is the local criterion, and therefore the corresponding fused tag and tag association history need to be rewritten according to the local criterion, which means that the fused track tag is on the local platform.
According to local criteria, for the tags of each remote LBC which are not currently associated with all local LBCs, fusing the tag naming rule and the corresponding tag association history determination rule as follows:
retrieving a tag association history of the remote LBC, if tags related to the local sensor index in the tag association history of the remote LBC are not empty, using the tags most associated with the local sensor index as final fusion tags, and attaching the currently associated tags and the remote sensor index as a new tag association history to the current tag association history; otherwise, if the tag related to the local sensor index in the tag association history of the remote LBC is empty, which means that the remote LBC is associated with the local sensor for the first time, the tag library of all local LBCs is retrieved, according to the time component in the tag of the remote LBC
Figure 135639DEST_PATH_IMAGE041
The maximum number of targets generated at this time
Figure 221406DEST_PATH_IMAGE042
Increase by 1 to obtain the final fusion tag
Figure 839470DEST_PATH_IMAGE043
I.e. by
Figure 250859DEST_PATH_IMAGE044
Further, the currently associated tag and the remote sensor index are combined into a first tag association history.
The embodiment of the invention also provides an LMB density fusion device for a multi-early warning machine target tracking system, which comprises:
a processor;
a memory having stored thereon a computer program operable on the processor;
wherein the computer program, when executed by the processor, implements the steps of the LMB density fusion method for multi-early warning machine target tracking of the above embodiments.
The embodiment of the invention also provides a computer-readable storage medium, wherein a data processing program is stored on the computer-readable storage medium, and when the data processing program is executed by a processor, the steps of the LMB density fusion method for target tracking of multiple early warning machines in the embodiment are realized.
The embodiment of the invention also carries out the following simulation experiment:
and evaluating and verifying the effectiveness of the provided sensor fusion method by taking an unfused local tracker as a reference. Hereinafter, the former and the latter are referred to as local tracking and fusion tracking, respectively. They are compared by using a performance metric called optimal sub-pattern assignment for track (OSPA-T), which includes tag error, potential error, and positioning error. The order parameter and the cutoff parameter are set to
Figure 892056DEST_PATH_IMAGE111
And
Figure 843832DEST_PATH_IMAGE112
another metric parameter is set to
Figure 254085DEST_PATH_IMAGE113
。
A. Experimental setup
For ease of illustration, consider a scenario in which 4 targets are tracked with 4 early warning machines in a 2-dimensional x-y plane. Assume that each object is moving according to a Constant Velocity (CV) model
Figure 836376DEST_PATH_IMAGE114
Wherein the target state becomes
Figure 27185DEST_PATH_IMAGE115
The state transition matrix is given by
Figure 454756DEST_PATH_IMAGE116
Wherein the content of the first and second substances,
Figure 250673DEST_PATH_IMAGE117
s is the time interval between the start of the cycle,
Figure 535024DEST_PATH_IMAGE118
is a zero mean white Gaussian process noise with covariance
Figure 150813DEST_PATH_IMAGE119
Wherein the content of the first and second substances,
Figure 647654DEST_PATH_IMAGE120
and
Figure 94816DEST_PATH_IMAGE121
is the variance of the process noise, modeling the acceleration along the x-axis and y-axis, respectively. The initial state of the target is shown in table 1.
TABLE 14 initial states of the targets
Figure 753330DEST_PATH_IMAGE122
Survival per target profileA rate of
Figure 653153DEST_PATH_IMAGE123
Each sensor detecting a probability of
Figure 422526DEST_PATH_IMAGE124
. For each sensor, the detection range is defined by the distance range
Figure 193036DEST_PATH_IMAGE125
(m) and the angular ranges given in table 2 and the position determination of each sensor.
Furthermore, the standard deviation of the measurement error is
Figure 84768DEST_PATH_IMAGE126
、
Figure 409571DEST_PATH_IMAGE127
And, and
Figure 44951DEST_PATH_IMAGE128
. Assuming that the clutter is uniformly distributed in the detection area, the potential distribution thereof obeys the mean value of
Figure 404388DEST_PATH_IMAGE129
Poisson distribution of (a). Since the polar coordinates are transformed into a Cartesian coordinate system, the transformed clutter is spatially distributed as
Figure 467022DEST_PATH_IMAGE130
Wherein, in the step (A),
Figure 544700DEST_PATH_IMAGE131
in order to monitor the volume of the region,
Figure 655875DEST_PATH_IMAGE132
to detect the range. Thus, the spatial intensity of the clutter is
Figure 932136DEST_PATH_IMAGE133
. Suppose that the maximum speed that the sensor can detect is
Figure 368933DEST_PATH_IMAGE134
The intensity of the clutter Doppler is
Figure 668328DEST_PATH_IMAGE135
. When BZ masking is involved, set
Figure 645511DEST_PATH_IMAGE136
And
Figure 713961DEST_PATH_IMAGE137
。
TABLE 24 parameters of sensors
Figure 118398DEST_PATH_IMAGE138
The local tracker uses the MM-GLMB-DBZ tracker. To avoid duplication, the common parameters of the local trackers are the same as in the unfused local trackers.
For the fusion engine, Metropolis weights are selected as consistent fusion weights. At each time, two consistency iterations are applied, i.e.
Figure 905088DEST_PATH_IMAGE139
. Estimated trajectories with lifetimes greater than 3 will be displayed and used to calculate performance statistics.
B. Experiment and results
The typical results of a single run with BZ masking and the statistical performance of 100 Monte Carlo runs are presented in turn.
Fig. 2 and 3 show the time when the target falls into VBZ and DBZ of the different sensors. For example, object 1 (T1) falls into the VBZ of sensors 2, 3, and 4, and DBZ of sensor 1 (S1) and sensor 3. More specifically, the time of falling into the VBZ of sensors 2, 3, and 4 is 1 to 43 seconds, 1 to 100 seconds, and 59 to 100 seconds, respectively, while the time period of falling into DBZ of sensors 1 and 3 is 46 to 58 seconds and 37 to 67 seconds. From fig. 2 and 3, the time period in which each target falls in the BZ of each sensor can be obtained. It can be seen that for each sensor, within 1 to 43 seconds, 46 to 58 seconds, 59 to 100 seconds, there are always two targets (e.g., T3 and (T4 or T1 or T2) that fall BZs for S1), and only one target (e.g., T3) falls into BZ for a short period of 44 to 45 seconds. Further, it is also noted that the time for the target to fall into BZ between 46 seconds and 58 seconds is relatively short, being 13 time steps.
FIG. 4 shows a comparison of the performance of the corresponding OSPA-T. It can be seen that the local tracker can track only two targets most of the time due to the presence of BZ, with some tracking capability for targets that are obscured by DBZ for a relatively short period of time (46 s-58 s). Therefore, during this time period, the target estimates are close to 3, and the corresponding OSPA-T curves drop somewhat. In contrast, all targets can be stably tracked over time using the proposed sensor fusion. Therefore, the OSPA-T obtained by the sensor fusion is obviously lower than that of the local tracker, and the validity of the sensor fusion is verified.
In summary, the LMB density fusion method and apparatus for the target tracking system of multiple early warning machines provided by the embodiments of the present invention comprehensively consider the problems of multiple fields of view and DBZ occlusion, and design an effective LMB density fusion method, which includes two important parts: the method comprises the following steps of first coarse association and then fine association, wherein the coarse association is used for screening out unique components which are not associated with other sensors, and the fine association is used for selecting common components (association components) which are associated with other sensors and determining association relations; and secondly, consistency fusion of combination and fusion, wherein in the fusion step, related public components are fused by adopting an improved GCI fusion rule, and in the combination step, unique components which do not belong to BZ are added into the fusion result so as to avoid attenuation caused by the GCI fusion rule. Simulation results show that the fusion method effectively achieves the goal of cooperative blind removal of multiple early warning machines, and compared with a local tracker, the tracking performance is remarkably improved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (10)

1. An LMB density fusion method for a multi-early warning machine target tracking system, the system comprises a sensor network consisting of a plurality of sensor nodes, each node is provided with an early warning machine radar, and the early warning machine radar of each node runs a multi-target tracker, and the method is characterized by comprising the following steps:
1) performing iteration initialization;
2) performing sequential label multi-Bernoulli LMB association on each node and adjacent nodes thereof, performing coarse association and then performing fine association, wherein the coarse association is used for screening out unique components which are not associated with a remote sensor in a local sensor and unique components which are not associated with the local sensor in the remote sensor, and the fine association is used for screening out public components which are associated with the local sensor and the remote sensor and determining an association relation;
3) and according to the results of the coarse correlation and the fine correlation, sequentially fusing each node and the adjacent nodes by the label multi-Bernoulli LMB, wherein the fusion comprises the following steps: for the associated public components, adopting consistency fusion processing; for the unique component not associated, whether it falls into the doppler shadow DBZ or the view-field shadow VBZ of another sensor is judged, if it is outside the doppler shadow DBZ or the view-field shadow VBZ of another sensor, it does not participate in the subsequent association fusion, if it is inside the doppler shadow DBZ or the view-field shadow VBZ of another sensor, it is directly added to the fusion result;
4) after performing pair-by-pair sequential fusion on each node and adjacent nodes thereof according to the step 2) and the step 3), storing a final fusion result as input of next iteration;
5) when all nodes and their adjacent nodes are fused in sequence, the number of iterations is increased by 1, and the next iteration is entered.
2. The LMB density fusion method for the multi-early warning machine target tracking system according to claim 1, wherein each node of the system runs a tag random finite set filter, the output of the tag random finite set filter is in a Gaussian mixture form, and the specific process of the coarse correlation is as follows:
local sensor
Figure DEST_PATH_IMAGE002
Middle label Bernoulli component LBC label
Figure DEST_PATH_IMAGE004
And a remote sensor
Figure DEST_PATH_IMAGE006
Middle label Bernoulli component LBC label
Figure DEST_PATH_IMAGE008
The space density of the generalized label multi-Bernoulli GLMB component is in a Gaussian mixture form, and the space density is respectively as follows:
Figure DEST_PATH_IMAGE010
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE012
representing local sensors
Figure DEST_PATH_IMAGE014
Middle label Bernoulli component LBC label
Figure DEST_PATH_IMAGE016
The spatial density of the generalized label multi-bernoulli GLMB component of (a),
Figure DEST_PATH_IMAGE018
representing spatial density
Figure DEST_PATH_IMAGE012A
The total number of gaussian components required,
Figure DEST_PATH_IMAGE020
represents the weight of the mixture of gaussians,
Figure DEST_PATH_IMAGE022
represents a mean and a covariance of
Figure DEST_PATH_IMAGE024
The normal distribution of (c),
Figure DEST_PATH_IMAGE026
representing remote sensors
Figure DEST_PATH_IMAGE028
Middle label Bernoulli component LBC label
Figure DEST_PATH_IMAGE030
The spatial density of the generalized label multi-bernoulli GLMB component of (a),
Figure DEST_PATH_IMAGE032
representing spatial density
Figure DEST_PATH_IMAGE026A
The total number of gaussian components required,
Figure DEST_PATH_IMAGE034
represents the weight of the mixture of gaussians,
Figure DEST_PATH_IMAGE036
represents a mean and a covariance of
Figure DEST_PATH_IMAGE038
Normal distribution of (2);
the mean and covariance of the weighted Gaussian components are respectively
Figure DEST_PATH_IMAGE040
And
Figure DEST_PATH_IMAGE042
；
if the difference between the two mean values
Figure DEST_PATH_IMAGE044
Obey card square
Figure DEST_PATH_IMAGE046
Distribution, namely:
Figure DEST_PATH_IMAGE048
wherein
Figure DEST_PATH_IMAGE050
Is the probability of the door,
Figure DEST_PATH_IMAGE052
is the dimension of the state vector and,
Figure DEST_PATH_IMAGE054
；
then correspond to
Figure DEST_PATH_IMAGE056
Is marked with a label
Figure DEST_PATH_IMAGE004A
And correspond to
Figure DEST_PATH_IMAGE058
Is marked with a label
Figure DEST_PATH_IMAGE008A
They are associated, otherwise they are not.
3. The LMB density fusion method for the multi-early warning machine target tracking system according to claim 2, wherein the precise association is performed by the following specific process:
binary allocation variable as defined below
Figure DEST_PATH_IMAGE060
：
Figure DEST_PATH_IMAGE062
In calculating the association cost, the following association metrics are utilized
Figure DEST_PATH_IMAGE064
：
Figure DEST_PATH_IMAGE066
Label (R)
Figure DEST_PATH_IMAGE004AA
And a label
Figure DEST_PATH_IMAGE008AA
Is expressed as
Figure DEST_PATH_IMAGE068
Which is a correlation metric
Figure DEST_PATH_IMAGE070
The negative logarithm of (d), i.e.:
Figure DEST_PATH_IMAGE072
solving for optimal distribution variable
Figure DEST_PATH_IMAGE074
Cause local sensors
Figure DEST_PATH_IMAGE075
Each track of (a) is assigned to a remote sensor
Figure DEST_PATH_IMAGE006A
A certain track or virtual track, and a remote sensor
Figure DEST_PATH_IMAGE006AA
Is assigned to a local sensor
Figure DEST_PATH_IMAGE075A
To a certain track or virtual track, i.e. solving the following two-dimensional distribution problem:
Figure DEST_PATH_IMAGE077
in the formula:
Figure DEST_PATH_IMAGE079
and
Figure DEST_PATH_IMAGE081
are respectively a sensor
Figure DEST_PATH_IMAGE075AA
And a sensor
Figure DEST_PATH_IMAGE006AAA
(ii) a tagset of the label poly-bernoulli LMB densities of (a);
then, two tag sets are implemented by finding the best allocation
Figure DEST_PATH_IMAGE079A
And
Figure DEST_PATH_IMAGE081A
a fine association between them.
4. The LMB density fusion method for the multi-early warning machine target tracking system according to claim 1, wherein the consistency fusion processing is adopted for the associated common components, and specifically comprises:
performing GCI fusion on the fused spatial densities for the label multi-bernoulli LMB densities associated with the local sensor and the remote sensor;
the fused existence probability EP is processed according to whether the associated label multi-bernoulli LMB density falls within the doppler dead zone DBZ or the field of view dead zone VBZ of another sensor, as follows: if the local tag bernoulli component LBC falls in the doppler blind zone DBZ or the field of view blind zone VBZ of the remote sensor, and the remote tag bernoulli component LBC does not fall in the doppler blind zone DBZ or the field of view blind zone VBZ of the local sensor, the existence probability EP of the fused tag bernoulli component LBC is taken as the existence probability EP of the local tag bernoulli component LBC; conversely, if the local tag bernoulli component LBC does not fall within the remote sensor's doppler blind zone DBZ or field of view blind zone VBZ, and the remote tag bernoulli component LBC falls within the local sensor's doppler blind zone DBZ or field of view blind zone VBZ, then the existence probability EP of the fused tag bernoulli component LBC takes the existence probability EP of the remote tag bernoulli component LBC.
5. The LMB density fusion method for a multi-early warning machine target tracking system of claim 4, further comprising:
for the associated common component, the fusion tag employs a local tag.
6. The LMB density fusion method for a multi-early warning machine target tracking system of claim 4, further comprising:
for the associated common component, the corresponding tag association history is the tag and corresponding sensor index of the remote tag bernoulli component LBC added to the tag association history of the local tag bernoulli component LBC.
7. The LMB density fusion method for a multi-warning target tracking system according to claim 1, wherein the determining whether the non-associated unique component falls into the doppler shadow DBZ or the view shadow VBZ of another sensor, if it is outside the doppler shadow DBZ or the view shadow VBZ of another sensor, then it does not participate in the subsequent association fusion, if it is inside the doppler shadow DBZ or the view shadow VBZ of another sensor, then it is directly added to the fusion result, specifically comprising:
for each unassociated local tag bernoulli component LBC obtained by coarse association, determining whether it falls within the remote sensor's doppler blind zone DBZ or field of view blind zone VBZ, and if it falls within the remote sensor's doppler blind zone DBZ or field of view blind zone VBZ and has a high probability of existence EP, adding it directly to the local fusion result;
for each unassociated remote tag bernoulli component LBC obtained by coarse association, it is determined whether it falls within the local sensor's doppler blind zone DBZ or field of view blind zone VBZ, and if it falls within the local sensor's doppler blind zone DBZ or field of view blind zone VBZ and has a high probability of existence EP, it is also added to the local fusion result and the corresponding fused tag and tag association history are rewritten according to local criteria that the fused track tag is on the local platform.
8. The LMB density fusion method for a multi-early warning machine target tracking system according to claim 7, wherein rewriting the corresponding fusion tag and tag association history according to local criteria specifically comprises:
retrieving a tag association history of the remote tag Bernoulli component LBC, if the tags related to the local sensor index in the tag association history of the remote tag Bernoulli component LBC are not empty, using the tags most associated with the local sensor index as final fusion tags, and using the currently associated tags and the remote sensor index as new tag associationsThe association history is appended to the current tag association history; otherwise, if the labels related to the local sensor index in the label association history of the remote label Bernoulli component LBC are null, retrieving the label library of all the local label Bernoulli component LBCs, and according to the time component in the labels of the remote label Bernoulli component LBCs, searching the label library of all the local label Bernoulli component LBCs
Figure DEST_PATH_IMAGE083
The maximum number of targets generated at this time
Figure DEST_PATH_IMAGE085
Increase by 1 to obtain the final fusion tag
Figure DEST_PATH_IMAGE087
I.e. by
Figure DEST_PATH_IMAGE089
Further, the currently associated tag and the remote sensor index are combined into a first tag association history.
9. An LMB density fusion device for a multi-early warning machine target tracking system, the device comprising:
a processor;
a memory having stored thereon a computer program operable on the processor;
wherein the computer program when executed by the processor implements the steps of the LMB density fusion method of multi-forewarning machine target tracking of any of claims 1 to 8.
10. A computer-readable storage medium, having stored thereon a data processing program which, when executed by a processor, implements the steps of the multi-forewarning machine target tracking LMB density fusion method of any one of claims 1-8.
CN202110852130.8A 2021-07-27 2021-07-27 LMB density fusion method and device for multi-early-warning-machine target tracking system Active CN113296089B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110852130.8A CN113296089B (en) 2021-07-27 2021-07-27 LMB density fusion method and device for multi-early-warning-machine target tracking system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110852130.8A CN113296089B (en) 2021-07-27 2021-07-27 LMB density fusion method and device for multi-early-warning-machine target tracking system

Publications (2)

Publication Number Publication Date
CN113296089A CN113296089A (en) 2021-08-24
CN113296089B true CN113296089B (en) 2021-11-19

Family

ID=77331306

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110852130.8A Active CN113296089B (en) 2021-07-27 2021-07-27 LMB density fusion method and device for multi-early-warning-machine target tracking system

Country Status (1)

Country Link
CN (1) CN113296089B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115200544B (en) * 2022-07-06 2023-05-09 中国电子科技集团公司第三十八研究所 Method and device for target tracking of maneuvering measurement and control station
CN116777950B (en) * 2023-04-19 2024-05-03 长沙理工大学 Multi-target visual tracking method, device, equipment and medium based on camera parameters

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3833904A (en) * 1973-02-05 1974-09-03 Hughes Aircraft Co Airborne switched array radar system
EP3026458B1 (en) * 2014-11-26 2021-09-01 Maritime Radar Systems Limited A system for monitoring a maritime environment
CN106371091B (en) * 2016-08-24 2018-11-02 四川九洲空管科技有限责任公司 The data fusion method and device of ADS-B and secondary radar monitoring information
CN106680806A (en) * 2016-11-24 2017-05-17 清华大学 Multi-radar plot fusion method
CN107192998B (en) * 2017-04-06 2019-05-31 中国电子科技集团公司第二十八研究所 A kind of adapter distribution track data fusion method based on covariance target function
CN107831490A (en) * 2017-12-01 2018-03-23 南京理工大学 A kind of improved more extension method for tracking target
CN108882271B (en) * 2018-07-05 2020-04-07 电子科技大学 Non-common-view multi-sensor fusion method based on label multi-Bernoulli distribution

Also Published As

Publication number Publication date
CN113296089A (en) 2021-08-24

Similar Documents

Publication Publication Date Title
Zhou et al. GM-PHD-based multi-target visual tracking using entropy distribution and game theory
CN113296089B (en) LMB density fusion method and device for multi-early-warning-machine target tracking system
CN110780269B (en) Explicit multi-target tracking method based on GM-PHD filter under self-adaptive new growth intensity
Chen et al. Tracking multiple moving objects using unscented Kalman filtering techniques
Xia et al. Extended target Poisson multi-Bernoulli mixture trackers based on sets of trajectories
Liang-Qun et al. Intuitionistic fuzzy joint probabilistic data association filter and its application to multitarget tracking
CN103902829B (en) Target tracking method and system transmitting edge distribution and existence probability
CN107861123A (en) A kind of through-wall radar is under complex environment to the method for multiple mobile object real-time tracking
Yazdian-Dehkordi et al. Penalized Gaussian mixture probability hypothesis density filter for multiple target tracking
CN111562569A (en) Weighted group sparse constraint-based multi-target constant false alarm detection method under Weibull background
Baser et al. A novel auxiliary particle PHD filter
Yang et al. A probabilistic framework for multitarget tracking with mutual occlusions
Qin et al. Application of an efficient graph-based partitioning algorithm for extended target tracking using GM-PHD filter
CN113743509A (en) Incomplete information-based online combat intention identification method and device
Wen et al. Optimization of the occlusion strategy in visual tracking
CN115657008A (en) Multi-target tracking method and device for airborne terahertz radar
Bae Survey of amplitude‐aided multi‐target tracking methods
Böhler et al. Stochastic partitioning for extended object probability hypothesis density filters
Zhu et al. The modified probability hypothesis density filter with adaptive birth intensity estimation for multi-target tracking in low detection probability
Wu et al. Multi-GMTI decentralized tracking via consensus LMB density fusion
CN106199595A (en) A kind of multipath mitigation algorithms based on node degree
Dunne et al. Weight partitioned probability hypothesis density filters
Laghmara et al. Evidential object association using heterogeneous sensor data
Zhu et al. A random matrix based method for tracking multiple extended targets
Zhang et al. Infrared dim-small object detection algorithm based on saliency map combined with target motion feature

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant