A distributed information fusion method based on sample greedy whisper communication

Abstract

The application discloses a kind of distributed information fusion methods based on sample greedy backdoor communication, comprising the following steps: S1, multiple unmanned aerial vehicle groups are communication network;S2, each unmanned aerial vehicle obtains target position information;S3, select an unmanned aerial vehicle, obtain the information matrix of the unmanned aerial vehicle;S4, fuse the information matrix of the unmanned aerial vehicle and other unmanned aerial vehicles;S5, update unmanned aerial vehicle and obtain target position information, repeat steps S2-S5, continuously monitor target.The distributed information fusion method based on sample greedy backdoor communication disclosed in the application has many advantages such as good stability, high positioning accuracy, fast positioning convergence speed and low communication load.

Description

TECHNICAL FIELD

[0001] The present application relates to an information fusion method, in particular to a distributed information fusion method based on sample greedy gossip communication, and belongs to the field of information management. BACKGROUND

[0002] In many cases, the detection of a single unmanned aerial vehicle on a target position will have a large error, and the detection information of multiple unmanned aerial vehicles needs to be fused to realize collaborative task execution.

[0003] The existing fusion strategy needs a large amount of communication between unmanned aerial vehicles, and when the communication is disturbed, a large amount of communication will bring a large delay, resulting in poor target tracking effect.

[0004] At the same time, the large amount of data processing process brought by a large amount of communication requires the unmanned aerial vehicle to have a strong on-board computer, which increases the communication load, increases the cost, and has many limitations.

[0005] In addition, the existing fusion method, such as random gossip distributed Kalman filter (RG-DKF), randomly selects a locally connected neighbor node for each sensor in each round of gossip communication iteration, and averages the two nodes, which has a low calculation amount, but due to its randomness, its convergence speed is relatively slow.

[0006] The greedy gossip distributed Kalman filter (GG-DKF) requires each local sensor to communicate with all adjacent sensor nodes.

[0007] Therefore, it is necessary to study an information fusion method capable of solving the above problems. SUMMARY

[0008] In order to overcome the above problems, the present application has been designed based on the in-depth research of the present application, and a distributed information fusion method based on sample greedy gossip communication is designed, which comprises the following steps:

[0009] S1, a plurality of unmanned aerial vehicles form a communication network;

[0010] S2, each unmanned aerial vehicle obtains target position information;

[0011] S3, select an unmanned aerial vehicle, and obtain the information matrix of the unmanned aerial vehicle;

[0012] S4, fuse the information matrix of the unmanned aerial vehicle and other unmanned aerial vehicles;

[0013] S5, update the target position information obtained by the unmanned aerial vehicle, and repeat steps S2-S5 to continuously monitor the target.

[0014] In a preferred embodiment, in S1, the plurality of unmanned aerial vehicles form a communication network, each unmanned aerial vehicle being a node in the network, and different unmanned aerial vehicles being capable of communicating with each other.

[0015] In a preferred embodiment, in S2, the target position information comprises a target estimation matrix and an error covariance matrix.

[0016] In a preferred embodiment, in S2, the unmanned aerial vehicle obtains the target position estimation matrix and the covariance matrix using an extended Kalman filter method according to its own unmanned aerial vehicle coordinates, unmanned aerial vehicle attitude angle and photoelectric sphere attitude angle.

[0017] In a preferred embodiment, in S3, a node is randomly selected from the unmanned aerial vehicles, and an information matrix of the node is obtained,

[0018] The information matrix can be expressed as:

[0019]

[0020]

[0021] wherein the subscript i represents the selected unmanned aerial vehicle, the superscript T represents transposition, u k,i represents an information form of the Kalman filter estimation value of the unmanned aerial vehicle i at the kth moment, U k,i represents an information form of the Kalman filter covariance of the unmanned aerial vehicle i at the kth moment, and N represents the total number of all unmanned aerial vehicles participating in information fusion.

[0022] In a preferred embodiment, in S4, an unmanned aerial vehicle j is selected from the other unmanned aerial vehicles to perform information matrix fusion with the unmanned aerial vehicle i, and the fused information matrix is taken as the information matrix of the unmanned aerial vehicle i.

[0023] In a preferred embodiment, the information matrix fusion is expressed as:

[0024]

[0025]

[0026] wherein u k,i , U k,i is the information matrix of the unmanned aerial vehicle i at the kth moment, u k,j , U k,j is the information matrix of the unmanned aerial vehicle j at the kth moment

[0027] In a preferred embodiment, the unmanned aerial vehicle j is selected by the following method:

[0028] selecting one or more unmanned aerial vehicles from the other unmanned aerial vehicles, so that the proportion of the selected unmanned aerial vehicles in the total number of all unmanned aerial vehicles is q;

[0029] comparing the Mahalanobis distances of all the selected unmanned aerial vehicles with the unmanned aerial vehicle i, and taking the unmanned aerial vehicle corresponding to the maximum Mahalanobis distance as the unmanned aerial vehicle j to be fused with the unmanned aerial vehicle i.

[0030] In a preferred embodiment, step S5 is performed after repeating step S4 multiple times.

[0031] In a preferred embodiment, in step S5, the obtained target position information of the unmanned aerial vehicle i in the extended Kalman filtering process is updated by k|k and P k|k :

[0032] X k|k,i =[U k,i ] -1 u k,i

[0033]

[0034] wherein X k|k,i represents the estimation value of the unmanned aerial vehicle i obtained at the k time, U k,i represents the information form of the Kalman filtering covariance of the unmanned aerial vehicle i at the k time, u k,i represents the information form of the Kalman filtering estimation value of the unmanned aerial vehicle i at the k time, P k|k,i represents the covariance propagation matrix of the estimation value X k|k,i of the unmanned aerial vehicle i at the k time, and N represents the total number of all unmanned aerial vehicles participating in the information fusion.

[0035] The present application has the beneficial effects including:

[0036] (1) good stability and high positioning accuracy;

[0037] (2) fast positioning convergence speed;

[0038] (3) low communication load, low calculation amount, and low requirement for unmanned aerial vehicles. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 showing a schematic diagram of a distributed information fusion method based on sample greedy small channel communication according to a preferred embodiment of the present application;

[0040] Figure 2 showing a target position estimation error diagram obtained by individual measurement of the four unmanned aerial vehicles in Example 1;

[0041] Figure 3A target position estimation error graph obtained after fusion in Embodiment 1 is shown;

[0042] Figure 4 A comparison graph of the target position obtained by the unmanned aerial vehicle pod 4 alone, the target position obtained after fusion and the real target position in Embodiment 1 is shown.

[0043] Figure 5 A comparison graph of the target position obtained by the unmanned aerial vehicle pod 4 alone, the target position obtained after fusion and the real target position in Embodiment 1 is shown. DETAILED DESCRIPTION

[0044] The application will be further described in details by the accompanying drawings and embodiments. The features and advantages of the application will become more apparent through these descriptions.

[0045] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations. Unless specifically stated otherwise, the drawings are not drawn to scale and the appended description is not intended to be limiting.

[0046] According to the application, a distributed information fusion method based on sample greedy small channel communication is provided, comprising the following steps:

[0047] S1, a plurality of unmanned aerial vehicles form a communication network;

[0048] S2, each unmanned aerial vehicle obtains target position information;

[0049] S3, one unmanned aerial vehicle is selected, and an information matrix of the unmanned aerial vehicle is obtained.

[0050] S4, the information matrix of the unmanned aerial vehicle and the information matrices of other unmanned aerial vehicles are fused;

[0051] S5, the target position information obtained by the unmanned aerial vehicle is updated, and steps S2-S5 are repeated to continuously monitor the target.

[0052] In S1, a plurality of unmanned aerial vehicles form a communication network, each unmanned aerial vehicle is a node on the network, and different unmanned aerial vehicles can communicate with each other.

[0053] Further, an optoelectronic head is arranged on the unmanned aerial vehicle, and the optoelectronic head is used to identify and lock a target and output angle information of the target relative to the unmanned aerial vehicle.

[0054] According to the application, the unmanned aerial vehicle further has a navigation device and a communication device,

[0055] The navigation device is used to output a position coordinate and an attitude angle of the unmanned aerial vehicle itself;

[0056] The communication device is used for target estimation information transmission with other unmanned aerial vehicles.

[0057] In the present application, the specific structure model of the navigation device and the communication device is not particularly limited, as long as the above functions can be realized.

[0058] In S2, the target position information includes a target estimation matrix and an error covariance matrix.

[0059] Each unmanned aerial vehicle simultaneously detects the target, and uses an extended Kalman filtering method to obtain a target position estimation matrix and a covariance matrix according to the unmanned aerial vehicle coordinates, the unmanned aerial vehicle attitude angle and the optoelectronic ball attitude angle.

[0060] According to the present application, the unmanned aerial vehicle can obtain the position of the target through detection, the position of the target can be represented by a target position state equation and an observation equation, and the target position estimation matrix and the covariance matrix can be obtained by performing extended Kalman filtering on the target position state equation and the observation equation.

[0061] Specifically, the target position state equation can be represented as:

[0062] X k =F k X k-1 +W k

[0063] The observation equation can be represented as:

[0064]

[0065] Wherein, k represents the time; X represents a state vector, which is an estimated coordinate of the target, X k represents the state vector at k time; Z represents an observation variable, which is a line-of-sight angle measured by the optoelectronic pod, Z k represents the observation variable at k time; F k represents a state transition matrix at k time, which is a unit matrix, x(k), y(k), z(k) represent the position coordinates of the target relative to the unmanned aerial vehicle at k time, x0, y0, z0 represent the position coordinates of the target relative to the unmanned aerial vehicle at the initial time, W k is noise in the state equation and the observation equation, V k is noise in the observation equation.

[0066] The extended Kalman filtering is a commonly used filtering method, which can be represented as:

[0067] X k|k-1 =F k X k-1|k-1 +W k

[0068] Pk|k-1 =FP k-1|k-1 F T +Q k

[0069] K k =P k|k-1 H k T (H k P k|k-1 H k T +R k ) -1

[0070] X k|k =X k|k-1 +K k (Z k -Z k|k-1 )

[0071] P k|k =(IK k H k )P k|k-1

[0072]

[0073] Among them, X k|k P represents the target position estimate at time k. k|k X represents k|k The corresponding error covariance matrix, X k|k-1 P represents the estimated matrix predicted based on the state vector at time k-1. k|k-1 X represents k|k-1 The corresponding covariance matrix, Z k|k-1 This represents the measurement prediction value K calculated from the information at time k-1. k R represents the Kalman filter gain coefficient at time k. k Let I represent the measurement noise covariance at time k, and let Q represent the identity matrix. k =0.

[0074] In S3, a drone is randomly selected as a node, and the information matrix of that drone is obtained.

[0075] The information matrix is ​​used to characterize the similarity between two nodes, serving as the basis for fusion.

[0076] The information matrix can be represented as:

[0077]

[0078]

[0079] wherein, the subscript i represents the selected unmanned aerial vehicle, the superscript T represents the transpose, u k,i represents the information form of the Kalman filter estimation value of the unmanned aerial vehicle i at the k moment, U k,i represents the information form of the Kalman filter covariance of the unmanned aerial vehicle i at the k moment, and N represents the total number of all unmanned aerial vehicles participating in information fusion.

[0080] In S4, one unmanned aerial vehicle j is selected from other unmanned aerial vehicles to perform information matrix fusion with the unmanned aerial vehicle i, and the fused information matrix is taken as the information matrix of the unmanned aerial vehicle i.

[0081] The information matrix fusion is represented as:

[0082]

[0083]

[0084] wherein, the subscript i represents the selected unmanned aerial vehicle, the superscript T represents the transpose, u k,i , U k,i is the information matrix of the unmanned aerial vehicle i at the k moment, u k,j , U k,j is the information matrix of the unmanned aerial vehicle j at the k moment.

[0085] In a preferred embodiment, the unmanned aerial vehicle j is selected by the following manner:

[0086] One or more unmanned aerial vehicles are selected from other unmanned aerial vehicles, so that the proportion of the selected unmanned aerial vehicles in the total number of all unmanned aerial vehicles is q, and preferably, the value of q is 0.3-0.7;

[0087] Further, the Mahalanobis distance of all selected unmanned aerial vehicles from the unmanned aerial vehicle i is compared, and the unmanned aerial vehicle corresponding to the maximum Mahalanobis distance is taken as the unmanned aerial vehicle j to perform fusion with the unmanned aerial vehicle i.

[0088] Further, when the unmanned aerial vehicle j cannot be selected from other unmanned aerial vehicles according to the proportion q, one unmanned aerial vehicle is randomly selected from other unmanned aerial vehicles as the unmanned aerial vehicle j.

[0089] Since the accuracy of information collected by a single unmanned aerial vehicle is limited, it means that the single unmanned aerial vehicle can observe part of the region, but cannot observe part of the region. In the present application, the information of multiple unmanned aerial vehicles is fused to realize collaborative observation.

[0090] Further, in actual scenarios, due to the limited field of view of the UAV and the non-uniform detection probability, not all UAVs can obtain the target information, and the target state vector deviation obtained by the UAV without target information will be very large, far from the true state. Fusing these relatively poor information with other relatively good information may reduce the performance of the fusion result. In the present application, the Mahalanobis distance solves this problem, so that the target information obtained by the UAV with poor information is not fused with each other, thereby helping to improve the stability of the fusion process.

[0091] In a more preferred embodiment, step S5 is performed after repeating step S4 multiple times, and since the UAV j is randomly generated each time, the information transmission fusion of multiple UAVs can be achieved.

[0092] In the present application, the Mahalanobis distance is used as the similarity measure between two local estimates, and multiple fusions are performed, which can realize sample greedy gossip communication, thereby finding a suboptimal communication path for each UAV node to perform fusion, so that the information can be quickly fused, and the convergence speed of the fused information is extremely fast.

[0093] Further, this kind of fusion method has extremely low information communication load, that is, the target positioning estimation information of each UAV is more accurate under limited communication load.

[0094] In S5, the target position information obtained by the UAV is updated, and steps S2-S5 are repeated to continuously track the target.

[0095] In step S5, the target position information obtained by the UAV i in the extended Kalman filtering process is updated by the following formula k|k and P k|k :

[0096] X k|k,i = [U k,i ] -1 u k,i

[0097]

[0098] so that the target position information obtained by the UAV i is more accurate.

[0099] wherein X k|k,i represents the target position estimation value of the UAV i at time k, U k,i represents the information form of the Kalman filter estimation value of the UAV i, u k,i represents the information form of the Kalman filter covariance of the UAV i, P k|k,i represents the estimation covariance of the UAV i, and N represents the total number of all UAVs participating in information fusion.

[0100] Example

[0101] Example 1

[0102] A swarm of drones approaches a target and continuously monitors it, including the following steps:

[0103] S1, Multiple drone groups form a communication network;

[0104] S2. Each UAV obtains target location information;

[0105] S3. Select a drone and obtain its information matrix;

[0106] S4. Integrate the information matrix of this drone with that of other drones;

[0107] S5. Update the target location information obtained by the drone, and repeat steps S2 to S5 to continuously monitor the target.

[0108] In S1, there are four drone groups forming a communication network. The drones are pod1, pod2, pod3, and pod4. Each drone is equipped with navigation and communication devices and has an electro-optical ball.

[0109] In S2, each UAV simultaneously detects the target and, based on its own UAV coordinates, UAV attitude angles, and electro-optical ball attitude angles, uses the extended Kalman filter method to obtain the target position estimation matrix and covariance matrix. Figure 2 The target position estimation errors obtained from four separate UAV measurements are shown.

[0110] In S3, a drone is randomly selected as a node; in this embodiment, pod4 is selected. The information matrix of this drone is obtained, and the information matrix can be represented as:

[0111]

[0112]

[0113] In S4, select one drone j from the other drones and fuse it with drone i to form an information matrix. The fused information matrix is ​​then used as the information matrix of drone i.

[0114] The information matrix fusion is represented as follows:

[0115]

[0116]

[0117] The drone j is selected in the following manner:

[0118] Select one or more unmanned aerial vehicles from the other unmanned aerial vehicles, so that the proportion of the selected unmanned aerial vehicles in the total number of unmanned aerial vehicles is q, q is 0.5;

[0119] Compare the Mahalanobis distance of all unmanned aerial vehicles with the Mahalanobis distance of the unmanned aerial vehicle i, and take the unmanned aerial vehicle corresponding to the maximum Mahalanobis distance as the unmanned aerial vehicle j to be fused with the unmanned aerial vehicle i.

[0120] When the unmanned aerial vehicle j cannot be selected from the other unmanned aerial vehicles according to the proportion q, an unmanned aerial vehicle is randomly selected from the other unmanned aerial vehicles as the unmanned aerial vehicle j.

[0121] The step S4 is repeated four times to realize the information transmission fusion of the four unmanned aerial vehicles.

[0122] In step S5, the obtained target position information of the unmanned aerial vehicle i in the extended Kalman filtering process is updated by the following formula: k|k and P k|k :

[0123] X k|k,i =[U k,i ] -1 u k,i

[0124]

[0125] Figure 3 The target position estimation error diagram obtained after fusion is shown, Figures 4-5 The target position error detected by pod4 after fusion by the above method and the target position error detected by unmanned aerial vehicle pod4 alone before fusion are shown, and it can be seen from the diagram that the detection of the target position is more accurate through the fusion of the above method.

[0126] In the description of the present application, it should be noted that the terms "upper", "lower", "inner", "outer", "front", "back" and the like indicate the orientation or positional relationship in the working state of the present application, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third", "fourth" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0127] In the description of the application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting" should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium; it can be internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood according to the specific circumstances.

[0128] The above describes the application in combination with the preferred embodiments, but these embodiments are only exemplary and serve only to illustrate. On this basis, various substitutions and improvements can be made to the application, which all fall within the protection scope of the application.

Claims

1. A distributed information fusion method based on sample greedy gossip communication, characterized in that, The method comprises the following steps: S1, a plurality of unmanned aerial vehicles form a communication network; S2, each unmanned aerial vehicle obtains target position information; S3, a random unmanned aerial vehicle is selected as a node, and an information matrix of the unmanned aerial vehicle is obtained; S4, an unmanned aerial vehicle j is selected from other unmanned aerial vehicles to fuse the information matrix with the unmanned aerial vehicle i, and the fused information matrix is taken as the information matrix of the unmanned aerial vehicle i; S5, the unmanned aerial vehicle updates the target position information, and steps S2 to S5 are repeated to continuously monitor the target; wherein, step S5 is performed after step S4 is repeated for multiple times; In S4, the unmanned aerial vehicle j is selected by the following method: One or more unmanned aerial vehicles are selected from other unmanned aerial vehicles, so that the proportion of the selected unmanned aerial vehicles in the total number of all unmanned aerial vehicles is q; The Mahalanobis distance of all selected unmanned aerial vehicles with the unmanned aerial vehicle i is compared, and the unmanned aerial vehicle corresponding to the maximum Mahalanobis distance is taken as the unmanned aerial vehicle j to fuse with the unmanned aerial vehicle i.

2. The distributed information fusion method based on sample greedy whisper communication according to claim 1, wherein in S1, a plurality of unmanned aerial vehicles form a communication network, and each unmanned aerial vehicle is a node on the network, and different unmanned aerial vehicles can communicate with each other.

3. The distributed information fusion method based on sample greedy whisper communication according to claim 1, wherein in S2, the target position information comprises a target estimation matrix and an error covariance matrix.

4. The distributed information fusion method based on sample greedy whisper communication according to claim 1, wherein in S2, the unmanned aerial vehicle obtains the target position estimation matrix and the covariance matrix by using an extended Kalman filtering method according to the unmanned aerial vehicle coordinates, the unmanned aerial vehicle attitude angle and the photoelectric sphere attitude angle of the unmanned aerial vehicle.

5. The distributed information fusion method based on sample greedy whisper communication according to claim 1, wherein the information matrix can be expressed as:

6. The distributed information fusion method based on sample greedy whisper communication according to claim 1, wherein the information matrix fusion is expressed as:

7. The distributed information fusion method based on sample greedy whisper communication according to claim 1, wherein ​ ​ wherein, the subscript i represents the selected unmanned aerial vehicle, the superscript T represents the transpose, u k,i represents the information form of the Kalman filter estimation value of the unmanned aerial vehicle i at the k moment, U k,i represents the information form of the Kalman filter covariance of the unmanned aerial vehicle i at the k moment, and N represents the total number of all unmanned aerial vehicles participating in information fusion. ​ ​ Wherein, u k,i , U k,i is the information matrix of the unmanned aerial vehicle i at the k moment, u k,j , U k,j is the information matrix of the unmanned aerial vehicle j at the k moment. ​ In step S5, the obtained target position information of the UAV i is updated in the extended Kalman filtering process by the following formula k|k and P k|k : wherein X k|k,i represents the estimation value obtained by the UAV i at the k moment, U k,i represents the information form of the Kalman filtering covariance of the UAV i at the k moment, u k,i represents the information form of the Kalman filtering estimation value of the UAV i at the k moment, P k|k,i represents the covariance propagation matrix of the estimation value X k|k,i of the UAV i at the k moment, and N represents the total number of all the UAVs participating in the information fusion.

Images