A distributed radar fusion tracking method based on adaptive birth model and dynamic field of view partitioning

Abstract

This paper discloses a distributed radar fusion tracking method based on an adaptive birth model and dynamic field of view partitioning. This method, based on active and passive distributed radars, solves the problems of target loss and track interruption for active radars and difficulty initiating tracks for passive radars in active interference scenarios through the adaptive birth model and dynamic field of view partitioning track fusion method, achieving stable target fusion tracking throughout the entire process. The method first determines the radar operating state and divides the field of view space. Then, based on the feedback of the fused target state, it sets a Bernoulli birth distribution. Simultaneously, when the radar field of view space changes, it adaptively adjusts the prediction model to achieve robust target tracking throughout the entire process and improve the target track integrity.

Description

Technical Field

[0001] The present invention relates to the field of distributed radar target tracking, and in particular to a distributed radar fusion tracking method based on an adaptive birth model and dynamic field of view division. Background Art

[0002] The electromagnetic environment of modern battlefields is becoming increasingly complex. Multi-sensor systems have been widely adopted due to their rich sensory information, excellent survivability, and strong anti-interference capabilities. However, multi-sensor systems still have limited perception capabilities in some detection scenarios, such as electromagnetic interference and limited detection power. In particular, when sensors are subject to active jamming, the target track is interrupted, and it is difficult to initiate a new track after the interruption, seriously affecting normal detection and early warning. Active and passive distributed radars, through the joint positioning technology of passive radar arrays, can perceive the target state under active jamming. At the same time, by combining global prior information about the target, they can effectively initiate the target track. Therefore, by utilizing global information to stabilize the target track in the radar field of view (FOV), and designing fusion weights based on the different FOVs of different radars, the track interruption problem under active jamming can be solved. This opens up the possibility of continuously tracking targets through track fusion under limited perception capabilities.

[0003] To address the problem of target track interruption in distributed radars with limited sensing capabilities, traditional track fusion methods are limited by the different influences of different radar fields of view, resulting in poor results. In view of this, W.Yi, G.Li and G.Battistelli ("Distributed Multi-Sensor Fusion of PHD Filters With Different Sensor Fields of View," in IEEE Transactions on Signal Processing, vol. 68, pp. 5204-5218, 2020.) proposed a track fusion method based on field of view partitioning for fixed radar fields of view, and a track fusion method based on target state clustering for obstructed radar fields of view. This method can achieve stable output of target tracks in distributed radar scenarios with different fields of view.

[0004] To address the challenge of target track initiation in distributed radars with limited sensing capabilities, traditional random finite set-based methods simply model the birth distribution using single-station measurements without considering global information. Furthermore, in active jamming scenarios, high radar clutter leads to high computational complexity, inaccurate birth distributions, and poor track initiation. In response to this, Li T, Sun S, Corchado JM, et al. (Random finite set-based Bayesian filters using magnitude-adaptive target birth intensity [C] / / 17th International Conference on Information Fusion (FUSION). IEEE, 2014: 1-8.) designed the magnitude and distribution of the birth distribution through measurement feedback, addressing the inaccurate birth distribution and achieving stable target initiation and tracking. Trezza et al. (Trezza, Anthony, Donald J. Bucci, and Pramod K. Varshney. "Multi-sensor joint adaptive birth sampler for labeled random finite set tracking." IEEE Transactions on Signal Processing 70 (2022): 1010-1025.) also achieved stable target track initiation and tracking by effectively utilizing global measurements from multiple radars through Gibbs sampling. These studies demonstrate that active and passive distributed radars have the potential to improve target track integrity within the radar's field of view by utilizing global prior information to initiate tracks and designing non-co-viewing track fusion methods.

[0005] The previous methods mentioned above have solved many of the challenges of distributed radar target tracking in conventional scenarios. However, they still have certain limitations for radar target tracking under active jamming. On the one hand, active jamming determines the radar's detection capability and, by extension, its field of view. However, the activation and deactivation of non-cooperative active jammers is unpredictable, leading to dynamic and unknown changes in the radar's field of view. The traditional method of dividing the field of view based on the radar's observation range is no longer feasible. It is necessary to dynamically divide the radar's field of view based on the radar's state and design fusion weights. On the other hand, due to factors such as active jamming and multipath effects, the radar will detect a large number of clutter points. At the track initiation stage, processing all points will result in an explosion in computational complexity, seriously affecting the performance of real-time processing. Therefore, it is necessary to design an adaptive birth model based on global prior information to quickly and accurately initiate the track.

[0006] Therefore, in the active and passive distributed radar scenarios under active interference, studying a fusion tracking method based on adaptive birth model and dynamic field of view division can effectively improve the track integrity, which has important practical significance and application value. Summary of the Invention

[0007] This invention provides a robust track initiation and fusion method for active and passive distributed radars under active jamming conditions. Based on active and passive distributed radars, this method, through an adaptive birth model and dynamic field-of-view tracking fusion method, addresses the issues of target loss and track interruption in active radars and difficulty in track initiation in passive radars, achieving stable target fusion tracking throughout the entire process.

[0008] The technical solution of the present invention is: a distributed radar fusion tracking method based on an adaptive birth model and dynamic field of view division, comprising the following steps:

[0009] Step 1: Initialize the node radar field of view division, judge the radar working status based on characteristic information such as echo amplitude, signal-to-noise ratio, and clutter rate, divide the radar field of view, calculate the field of view indicator function, and assign the fusion weight of each node radar;

[0010] Step 2: Node radar target tracking. Based on the Bernoulli filter, each node uses local measurements to track the target.

[0011] Step 3: Track fusion at the fusion center: Set the distribution amplitude threshold for nodes to communicate to the fusion center, filter the Gaussian terms in the Bernoulli distribution to be fused, and then transmit the target state distribution to the fusion center for target state fusion based on dynamic field of view partitioning.

[0012] Step 4: Set up adaptive birth distribution. After the target state is fused in Step 3, if there is a target at the fusion center, compare the number of targets at the fusion center with the number of targets at each node. Add an adaptive birth distribution to the nodes where no target is filtered out. The amplitude is the set value, and the distribution is determined by the fusion target state.

[0013] Step 5: Detection and processing of radar field of view changes. When the radar's echo amplitude, signal-to-noise ratio, clutter rate and other characteristic information change significantly, the field of view indicator function is recalculated based on the radar working state, and it is considered that a field of view switching event has occurred at this time; the current time frame is recorded as k. c , and add a protection time k for the radar view switching event p , and immediately proceed to "Step 6";

[0014] Step 6: Modify the radar parameters for field of view switching. After the operation is completed, return to "Step 2" to perform node radar target tracking.

[0015] Furthermore, in step 1, the node radar field of view division initialization method is: in the active and passive distributed radar tracking scenario, the active radar field of view is represented as The passive radar field of view is expressed as The distributed radar observation space is All radar fields of view belong to this observation space, i.e. The radar working status is defined as The radar working state is divided into normal and and abnormal Right now:

[0016]

[0017]

[0018] in, represents the empty set;

[0019] Radar field of view indicator function F i (x) is as follows:

[0020]

[0021] in, for The indicator function is defined as shown above, Γ(·) is the target state space To the observation space The mapping satisfies

[0022] Furthermore, in the step 1, the radar working state is divided into two states: normal working state and abnormal working state, and the radar field of view is adjusted accordingly, the field of view indicator function is calculated, and the fusion weight of each node radar is assigned.

[0023] Furthermore, in step 2, the method for performing one-step prediction of the existing target state based on the target motion model is:

[0024] r k|k-1 =p b (1-r k-1|k-1 )+p s r k-1|k-1

[0025]

[0026] Among them, r k|k-1 represents the probability of the target existing at time k, p k|k-1 (x) represents the predicted target space state distribution at time k, r k-1|k-1 and p k-1|k-1(x) represent the target posterior existence probability and target posterior spatial distribution at time k-1 respectively. In the formula, p b is the target birth probability, b k|k-1 (x) is the target birth distribution, p s is the target survival probability, π k|k-1 (x|x′) represents the target state transition probability density function.

[0027] Furthermore, in step 2, when local measurement arrives, the target state is updated as follows:

[0028]

[0029]

[0030] Where z is the radar receiving measurement, Z k is the radar received measurement set, p d (x) represents the radar target detection probability, g k (z|x) is the target measurement likelihood function, λ is the clutter rate, and c(z) represents the probability distribution of the measurement being clutter, which is a uniform distribution here.

[0031] Furthermore, in step 3, the distribution amplitude threshold T of the node communicating to the fusion center is set. a , after filtering the Gaussian items in the Bernoulli distribution to be fused, the target state distribution is transmitted to the fusion center for target state fusion; suppose the radar sequence number set to be fused is , the Bernoulli arithmetic mean fusion can be expressed as follows:

[0032]

[0033]

[0034] Among them, r f and p f (x) represents the target existence probability and target space state distribution after Bernoulli arithmetic mean fusion; ω i By the view indicator function F i (x) Impact:

[0035]

[0036] in, Represents the original fusion weight of radar i, satisfying That is, the fusion weight depends on the radar field of view indicator function. When the radar field of view indicator function is 0, the fusion weight is 0.

[0037] Furthermore, in step 4, after the target state fusion in step 3, if there is a target in the fusion center, the number of targets in the fusion center is compared with the number of targets in each node, and the following two-step conditional judgment is performed in sequence:

[0038] Judgment condition 1: The radar's local target number estimate should be consistent with the fusion center, which can be expressed as r in Bernoulli filtering. i >T d , where T d is the decision threshold of the Bernoulli estimator, usually set to 0.5;

[0039] Judgment condition 2: The radar's local target state estimate should be consistent with the fusion center, which can be expressed in Bernoulli filtering as:

[0040]

[0041] in, represents the critical value of the chi-square distribution with the degree of freedom being the target state dimension n and the significance level being α, and They are Gaussian approximations of the target spatial state estimation for the radar local and fusion center respectively, and the approximate results are the first and second order moments of the Gaussian mixture distribution.

[0042] Furthermore, in step 4, if any step in the comparison does not meet the conditions, an adaptive birth distribution is added for the node that does not filter out the target; its target birth probability p b is the set value The target space state distribution b(x) is obtained by the one-step prediction of the fusion center target state p(x′) according to the target motion model:

[0043] b(x)=∫π(x|x′)p(x′)dx′

[0044] Where π(x|x′) is the target state transition density; if the condition is met at each step, the node does not add the adaptive birth distribution.

[0045] Furthermore, in step 6, the current time frame k is within the view switching protection time, that is, k<k c +k p , do the following:

[0046] The first step is to detect the limited radar capability, that is, the radar working state For radars, the target survival probability in the Bernoulli filter gradually decays to prevent incorrect estimation of the target state:

[0047] p s =max(εp s ,p d )

[0048] Among them, ε is the target survival probability attenuation coefficient, p s is the target survival probability, p d is the target detection probability;

[0049] The second step is to modify the target birth probability p of the adaptive birth distribution in "Step 4". b , increasing the target birth probability is equivalent to confirming the prior information, which will make the track start faster. In the Bernoulli filter, there is no need to consider the overestimation of the number of targets caused by too many Bernoulli distributions, then:

[0050]

[0051] in, is the maximum target birth probability;

[0052] The third step is to detect radar with limited sensing capability, that is, the radar working state The radar maintains its field of view indicator function value as 1, and robust fusion is achieved at this time, that is, the target information of the radar with limited perception capability is retained to participate in the fusion.

[0053] To summarize, "Step 1" is the initialization operation, the initial radar field of view information; "Step 5" and "Step 6" are operations after the special event is triggered, which can effectively deal with active interference; the normal processing flow is: first perform the "Step 1" initialization operation, and then perform the "Step 2", "Step 3" and "Step 4" loop operations to achieve conventional target fusion tracking, and each loop detects whether the "Step 5" and "Step 6" operations are required.

[0054] The distributed radar fusion tracking method based on the adaptive birth model and dynamic field of view division proposed in the present invention has the following beneficial effects:

[0055] 1. Each radar node selectively sets an adaptive birth distribution based on the target state fed back by the fusion center. When active jamming occurs, the birth distribution parameters and survival probability parameters are modified. This fully utilizes the global information detected by the active and passive distributed radars and achieves rapid initiation and termination of target tracks for the active and passive distributed radars in active jamming scenarios.

[0056] 2. The fusion center modifies the fusion weights in real time based on the dynamic radar field of view, making full use of the field of view information to effectively prevent misestimation of target status and number, achieving stable fusion tracking of active and passive distributed radar targets in active jamming scenarios, and improving the target's track integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 Detection diagram of a typical active and passive distributed radar system;

[0058] Figure 2 Schematic diagram of distributed radar field of view division;

[0059] Figure 3 Adaptive birth model Bernoulli filter processing flow chart;

[0060] Figure 4 Radar target track fusion result diagram

[0061] Figure 5 Schematic diagram of target presence probability changes in active and passive radars;

[0062] Figure 6 Target OSPA error estimated by radar fusion;

[0063] Figure 7 Target potential error estimated by radar fusion. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0065] Typical active and passive distributed radar system detection scenarios include: Figure 1 As shown. The active and passive distributed radar system consists of a main radar and passive radars arranged around the main radar, and its detection mode is as follows. When there is no interference signal, the active radar can detect and track the target normally, and the echo of the passive radar is mostly clutter points. At this time, the active radar is mainly used to detect the target. When there is an interference signal, the active radar is interfered with, and a large amount of clutter appears in the echo, making it impossible to detect and track normally. At this time, the passive radar can be combined to jointly locate the status information of the target with poor accuracy, that is, the passive radar is mainly used to detect the target. When the active radar turns on the anti-interference measures, both the active and passive radars can detect the target. At this time, the active radar with higher accuracy is mainly used for detection. Based on this, the specific steps of a distributed radar fusion tracking method based on adaptive birth model and dynamic field of view division are as follows:

[0066] Step 1: Initialize the node radar field of view division, judge the radar working status based on characteristic information such as echo amplitude, signal-to-noise ratio, and clutter rate, divide the radar field of view, calculate the field of view indicator function, and assign the fusion weight of each node radar.

[0067] The radar field of view can be understood as the radar detection area, which requires the radar to be able to measure the status of the target within the area. Figure 2 As shown in the figure, the conventional radar field of view and its division method are shown. The single radar field of view is as follows and That is, the area within the radar detection boundary. represents the area where two radar fields of view intersect, Represents the area outside the viewshed.

[0068] However, in complex electromagnetic scenarios, the radar field of view cannot be correctly divided based on the detection area alone. For example, when the active radar is subject to electromagnetic interference, the active radar beam can still cover the area it scans, but it cannot measure the target's status information. Therefore, in the active and passive distributed radar tracking scenario with limited perception capabilities, the following definition is declared. The total field of view of the active radar is expressed as The total field of view of the passive radar is expressed as At the same time, we define the distributed radar observation space All radar fields of view belong to this observation space, i.e. The radar working status is defined as For ease of understanding, the radar working status is divided into normal and abnormal Then we have:

[0069]

[0070]

[0071] Radar field of view indicator function F i (x) is as follows:

[0072]

[0073] in, for The indicator function is defined as shown in the above formula. Γ(·) is the target state space To the observation space The mapping satisfies

[0074] Step 2: Node radar target tracking. Based on the Bernoulli filter, each node uses local measurements to track the target.

[0075] The target state modeled by Bernoulli random finite set is expressed as follows:

[0076]

[0077] Among them, x is a single target state, X is the target state set, f(X) represents the target state density function, r represents the target existence probability, p(x) represents the target space state distribution, Represents the empty set.

[0078] The Bernoulli filter is based on the Bayesian framework, and the processing flow includes two steps: prediction and update, which requires recursive target existence probability r and target space state distribution p(x).

[0079] The method for making a one-step prediction of the existing target state based on the target motion model is:

[0080] r k|k-1 =p b (1-r k-1|k-1 )+p s r k-1|k-1

[0081]

[0082] Among them, r k|k-1 represents the probability of the target existing at time k, p k|k-1 (x) represents the predicted target space state distribution at time k. k-1|k-1 and p k-1|k-1 (x) represent the target posterior existence probability and target posterior spatial distribution at time k-1 respectively. In the formula, p b is the target birth probability, b k|k-1 (x) is the target birth distribution, p s is the target survival probability, π k|k-1 (x|x′) represents the target state transition probability density function.

[0083] When local measurements arrive, the target state is updated as follows:

[0084]

[0085]

[0086] Where z is the radar receiving measurement, Z k is the radar receive measurement set. p d (x) represents the radar target detection probability, g k (z|x) is the target measurement likelihood function, λ is the clutter rate, and c(z) represents the probability distribution of the measurement being clutter, which is a uniform distribution here.

[0087] When implemented on a computer, the Gaussian mixture distribution is used to approximate the target space state distribution, namely:

[0088]

[0089] in, Indicates that the mean is m i , covariance is P i Gaussian probability density function, w i is the weight of the i-th Gaussian component, and N represents the total number of Gaussian components. Then, the recursive parameter is replaced by the weight w i , mean m i and the covariance matrix Pi .

[0090] Step 3: Track fusion at the fusion center. Set the distribution amplitude threshold for nodes to communicate with the fusion center. After screening the Gaussian items in the Bernoulli distribution to be fused, transmit the target state distribution to the fusion center for target state fusion based on dynamic field of view division.

[0091] After the radar local target tracking in "step 2", the amplitude threshold T of the radar node communication to the fusion center is set. a , filter the Gaussian items in the Bernoulli distribution to be fused. Only the Gaussian component weight w i Greater than the amplitude threshold T a The Gaussian components of are transmitted to the fusion center for target state arithmetic average fusion. Assume that the radar sequence number set to be fused is The Bernoulli arithmetic mean fusion can be expressed as follows:

[0092]

[0093]

[0094] Among them, r f and p f (x) represents the target existence probability and target space state distribution after Bernoulli arithmetic mean fusion. i By the view indicator function F i (x) Impact:

[0095]

[0096] in, Represents the original fusion weight of radar i, satisfying

[0097] Step 4: Set up adaptive birth distribution. After the target state is fused in Step 3, if there is a target at the fusion center, compare the number of targets at the fusion center with the number of targets at each node. Add an adaptive birth distribution for nodes that do not filter out targets. The amplitude is the set value, and the distribution is determined by the fusion target state.

[0098] Figure 3 The following diagram shows the Bernoulli filter processing flow for adaptive birth distribution. The following details the adaptive birth distribution setting method. After the target state fusion in "Step 3," if there is a target at the fusion center, the number of targets at the fusion center is compared with the number of targets at each node. If there is no target at the fusion center, no operation is performed. The comparison method between the central node and each node is as follows:

[0099] First, the radar's local target number estimate should be consistent with the fusion center. In Bernoulli filtering, it can be expressed as ri >T d , where T d is the decision threshold of the Bernoulli estimator, usually set to 0.5. If they are consistent, the comparison will continue. The radar's local target state estimate should be consistent with the fusion center, which can be expressed in Bernoulli filtering as:

[0100]

[0101] in, It represents the critical value of the chi-square distribution with the target state dimension n as the degree of freedom and the significance level α as the significance level. and The Gaussian approximation of the target spatial state estimates for the radar local and fusion centers is possible because the target spatial state distribution described by the Gaussian mixture distribution in the Bernoulli filter is for the same target. The approximation method is to calculate the first and second moments of the Gaussian mixture distribution. Take the radar local target spatial state estimate approximation as an example:

[0102]

[0103]

[0104] Among them, N l Indicates the total number of local Gaussian components.

[0105] If any step in the comparison does not meet the conditions, the Bernoulli birth distribution is added to the radar node at the next moment, and its target birth probability p b is the set value The target space state distribution b(x) is predicted by the fusion center target state p(x′):

[0106] b(x)=∫π(x|x′)p(x′)dx′

[0107] Among them, π(x|x′) is the target state transition density.

[0108] Step 5: Detection and processing of radar field of view changes. When the radar's echo amplitude, signal-to-noise ratio, clutter rate and other characteristic information change significantly, it is considered that the radar field of view has changed, and the current time frame is set to k. c ; At this time, add protection time k for radar field switching p , and proceed immediately to "Step 6".

[0109] Step 6: Modify the radar parameters for field of view switching. After the operation is completed, return to "Step 2" to perform node radar target tracking.

[0110] The current time frame k is within the view switching protection time, that is, k<kc +k p , do the following:

[0111] Step 1: Radar working status The radar with the survival probability decaying in the Bernoulli filter is:

[0112] p s =max(εp s ,p d )

[0113] Among them, ε is the target survival probability attenuation coefficient, p s is the target survival probability, p d is the target detection probability.

[0114] The second step is to modify the target birth probability p of the Bernoulli birth distribution in "Step 4". b In Bernoulli filtering, there is no need to consider the overestimation of the number of targets due to excessive Bernoulli distributions, so:

[0115]

[0116] in, is the maximum target birth probability.

[0117] Step 3: Set the radar working status The radar's field of view indicator function value is still 1.

[0118] The present invention provides the following embodiments to illustrate the inventive method:

[0119] The verification conditions of the embodiment of the present invention are shown in Table 1. The interference type used in the embodiment is noise suppression. The relevant conditions of this embodiment do not affect the specific implementation and technical content of the present invention.

[0120] Table 1 Implementation example verification conditions

[0121]

[0122]

[0123] During the tracking process, the state vector of the target is expressed as The trace vector of the radar measurement is represented by z k =[x k ,y k ] T .

[0124] The state transition equation is x k+1 =F k x k +v k, state transition matrix F k As shown below, where I2 is the identity matrix of dimension 2:

[0125]

[0126] The measurement equation is z k =H k x k +w k , measurement matrix H k for:

[0127]

[0128] Among them, v k is a zero-mean Gaussian process noise vector with standard deviation σ v , take 5m; w k is a zero-mean Gaussian measurement noise vector with a standard deviation of σ w , the active radar and passive radar are taken as 100m and 500m respectively.

[0129] Set the target initial position x1 = [10km, 400m / s, 15km, 0m / s] T Among the algorithm parameters, the amplitude threshold T of the radar node communicating to the fusion center is a Take 0.3, the significance level of the chi-square distribution is α and take 0.95, the target birth probability of the Bernoulli birth distribution Take 0.05, Take 0.4, the view switching protection time k p The time interval is 3s, and the target survival probability attenuation coefficient ε is 0.9.

[0130] Figure 4 The radar target track fusion result diagram is given. It can be seen from the figure that the fused track has high integrity and realizes full target tracking.

[0131] Figure 5 The change of target existence probability in active and passive radars is given, and it can be seen that the proposed adaptive birth distribution method can quickly start the target track.

[0132] In the comparison method, the optimal birth distribution is to set the birth distribution at the moment the target appears, and the target birth probability is The target space state distribution is the same as the real target state distribution. This is a theoretically optimal method. In reality, accurate prior information is required, which is difficult to achieve. Fixed birth distribution means that the birth distribution is set at a fixed position at each moment, and the target birth probability is The target space state distribution is the same as the real target state distribution; the measurement driven birth distribution refers to the birth distribution of the measurement setting at the tracking moment. Figure 5It can be seen that, especially for passive radars with poor accuracy, the track initiation performance is greatly improved, and the track can be established 5-6 frames earlier than the measurement-driven birth distribution and fixed birth distribution methods.

[0133] Figure 6 and Figure 7 The target optimal subpattern assignment (OSPA) error and potential error of radar fusion estimation are shown respectively. The cutoff parameter of OSPA is set to 1000 and the order parameter is set to 2.

[0134] The compared methods are all based on the Arithmetic Average (AA) criterion. AA refers to the standard arithmetic average fusion, SD (State-Dependent)-AA refers to the arithmetic average fusion based on fixed view division, and B2B (Bernoulli-to-Bernoulli)-AA refers to the arithmetic average fusion based on Bernoulli term matching. Figure 6 and Figure 7 It can be seen that the proposed method effectively reduces tracking error. In particular, when passive radar tracking is used, the effective track initiation and fusion method significantly improves the estimation accuracy of the number of targets and effectively maintains the integrity of the track.

[0135] The above results show that after using the method proposed in the present invention for track fusion, the estimation of target number and status is more accurate, and the track integrity is improved, which shows the superiority of the method proposed in the present invention.

[0136] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A distributed radar fusion tracking method based on adaptive birth model and dynamic field of view partitioning, characterized in that: The steps include: Step 1: Initialize the node radar field of view division, judge the radar working status based on the echo amplitude, signal-to-noise ratio, and clutter rate characteristic information, divide the radar field of view, calculate the field of view indicator function, and assign the fusion weight of each node radar; Step 2: Node radar target tracking. Based on the Bernoulli filter, each node uses local measurements to track the target. Step 3: Track fusion at the fusion center: Set the distribution amplitude threshold for nodes to communicate to the fusion center, filter the Gaussian terms in the Bernoulli distribution to be fused, and then transmit the target state distribution to the fusion center for target state fusion based on dynamic field of view partitioning. Step 4: Adaptive birth distribution setting: After the target state fusion in "Step 3", if there is a target at the fusion center, compare the number of targets at the fusion center with the number of targets at each node; Add an adaptive birth distribution for nodes that do not filter out the target, with an amplitude set to a set value and a distribution determined by the fusion target state; Step 5: Detection and processing of radar field of view changes. When the radar echo amplitude, signal-to-noise ratio, and clutter rate characteristic information change significantly, the field of view indicator function is recalculated based on the radar working state, and it is considered that a field of view switching event has occurred at this time; the current time frame is recorded as , and add a protection time for the radar view switching event , and immediately proceed to "Step 6"; Step 6: Modify the radar parameters for field of view switching. After the operation is completed, return to "Step 2" to perform node radar target tracking.

2. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning according to claim 1, characterized in that: In the step 1, the node radar field of view division initialization method is: in the active and passive distributed radar tracking scenario, the active radar field of view is represented as , the passive radar field of view is expressed as , the distributed radar observation space is , all radar fields of view belong to this observation space, that is, , the radar working state is defined as The radar working state is divided into normal and and abnormal ,Right now: ； ； in, represents the empty set; Radar field of view indicator function As shown below: ； in, for The indicator function is defined as follows: The target state space To the observation space The mapping satisfies .

3. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning as claimed in claim 1, characterized in that: In the step 1, the radar working state is divided into two states: normal working state and abnormal working state, and the radar field of view is adjusted accordingly, the field of view indicator function is calculated, and the fusion weight of each node radar is assigned.

4. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning according to claim 1, characterized in that: In step 2, the method for performing one-step prediction of the existing target state based on the target motion model is: ； ； in, represents the probability of the target existing at time k, represents the predicted target space state distribution at time k, and They represent the target posterior existence probability and target posterior spatial distribution at time k-1 respectively. In the formula, is the target birth probability, is the target birth distribution, is the target survival probability, represents the target state transition probability density function.

5. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning as claimed in claim 4, characterized in that: In step 2, when local measurements arrive, the target state is updated as follows: ； ； in, is the radar receiving measurement, is the set of radar received measurements, represents the radar target detection probability, is the target measurement likelihood function, is the clutter rate, Indicates the probability distribution of the measurement being clutter, which is uniform distribution here. .

6. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning as claimed in claim 1, characterized in that: In step 3, the distribution amplitude threshold of the node communication to the fusion center is set. , after filtering the Gaussian items in the Bernoulli distribution to be fused, the target state distribution is transmitted to the fusion center for target state fusion; suppose the radar sequence number set to be fused is , the Bernoulli arithmetic mean fusion can be expressed as follows: ； ； in, and They represent the target existence probability and target space state distribution after Bernoulli arithmetic mean fusion respectively; By the viewshed indicator function Influence: ； in, Indicates radar The original fusion weights satisfy , that is, the fusion weight depends on the radar field of view indicator function. When the radar field of view indicator function is 0, the fusion weight is 0.

7. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning as claimed in claim 1, characterized in that: In step 4, after the target state fusion in step 3, if there is a target in the fusion center, the number of targets in the fusion center is compared with the number of targets in each node, and the following two-step condition judgment is performed in sequence: Judgment condition 1: The radar's local target number estimate should be consistent with the fusion center, which can be expressed in Bernoulli filtering as ,in is the decision threshold of the Bernoulli estimator, usually set to 0.5; Judgment condition 2: The radar's local target state estimate should be consistent with the fusion center, which can be expressed in Bernoulli filtering as: ； in, The degrees of freedom are the target state dimension n, and the significance level is The critical value of the chi-square distribution, and They are Gaussian approximations of the target spatial state estimation for the radar local and fusion center respectively, and the approximate results are the first and second order moments of the Gaussian mixture distribution.

8. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning as claimed in claim 1, characterized in that: In step 4, if any step in the comparison does not meet the conditions, then add an adaptive birth distribution for the node that does not filter out the target; its target birth probability is the set value , target space state distribution Target status by fusion center According to the one-step prediction of the target motion model, we can get: ； in, is the target state transition density; if the condition is met at each step, the node does not add adaptive birth distribution.

9. The distributed radar fusion tracking method based on adaptive birth model and dynamic view partitioning as claimed in claim 1, characterized in that: In step 6, the current time frame In the view switching protection time, that is, , do the following: The first step is to detect the limited radar capability, that is, the radar working state For radars, the target survival probability in the Bernoulli filter gradually decays to prevent incorrect estimation of the target state: ； in, is the target survival probability attenuation coefficient, is the target survival probability, is the target detection probability; Step 2: Modify the target birth probability of the adaptive birth distribution in "Step 4" , increasing the target birth probability is equivalent to confirming the prior information, which will make the track start faster. In the Bernoulli filter, there is no need to consider the overestimation of the number of targets caused by too many Bernoulli distributions, then: ； in, is the maximum target birth probability; The third step is to detect radar with limited sensing capability, that is, the radar working state The radar maintains its field of view indicator function value as 1, and robust fusion is achieved at this time, that is, the target information of the radar with limited perception capability is retained to participate in the fusion.

10. A distributed radar fusion tracking method based on an adaptive birth model and dynamic view partitioning according to any one of claims 1 to 9, characterized in that: First, perform the initialization operation of "Step 1", and then loop the operations of "Step 2", "Step 3" and "Step 4" to achieve conventional target fusion tracking. Each loop detects whether "Step 5" and "Step 6" operations are required.

Images