A motorized frequency adaptive cooperative positioning method based on multi-sensor fusion

Abstract

A kind of motorized frequency adaptive cooperative positioning method based on multi-sensor fusion, the present application is used to solve the problem that fusion model in prior art is prone to filter divergence and track distortion, the present application includes: using multiple sensors to continuously receive target position information, the received track of each type of information source is unified to CGCS2000 geocentric coordinate system, then the track of each information source converted to the unified coordinate system is adaptively filtered, finally, the track information of each information source is fused by using enhanced level fusion algorithm combined with motorized frequency adaptation, to obtain the integrated track of target.The present application is specially used for multi-sensor receiver to receive motorized target position information, and through track fusion network, integrated track with higher precision and stronger stability is obtained.The fusion process can fully utilize the advantage of each type of sensor and consider various motion states of motorized target.The present application belongs to the field of multi-source heterogeneous information fusion technology.

Description

TECHNICAL FIELD

[0001] The present application relates to a positioning method, in particular to a motorized frequency adaptive cooperative positioning method based on multi-sensor fusion, and belongs to the technical field of multi-source heterogeneous information fusion. BACKGROUND

[0002] Information fusion technology is a relatively hot research direction at present, and its related research content and subject range are relatively broad. The definition of information fusion technology widely recognized by relevant scholars in the field before is as follows: the data from each sensor in the fusion system is combined, associated and combined to obtain the identity recognition and position estimation of the target, so as to realize the comprehensive evaluation of the threat and importance of the target. Based on the breakthrough of sensor technology and the progress of information fusion in theory and technology, the more generalized expression of sensor information fusion technology is as follows: various types of sensor resources at different times and spaces are integrated in a system, the measurement data from each sensor in the system is analyzed and fused under the fusion rule in the fusion center, the consistent description of the observed target under different sensor nodes is obtained, and the target motion state estimation and comprehensive decision are completed.

[0003] The feature fusion of trajectory data is the fusion of multiple features of trajectory objects, which can comprehensively consider the multiple features of trajectory data for trajectory data similarity measurement. Rao Yuanqi's feature fusion trajectory similarity measurement method designs different similarity measurement methods for different trajectory features, solving the problem that one measurement method can only measure one trajectory feature. Second, the similarity matrix of the trajectory is constructed, and the trajectory data is classified by clustering method, and a measurement method based on trajectory feature fusion is given. Ding Feng proposes a target data association algorithm based on the fuzzy theory of Han Junfeng and Li Yuhui. The membership degree concept is introduced for comprehensive decision-making, and the size of the membership degree function value is used to judge whether the trajectory information from multiple sensors corresponds to the same trajectory, so as to realize the association decision between different sensor trajectory data. Buede assumes that the trajectories of multiple sensors are successfully associated in the fusion center, and the multi-target tracking problem can be converted into single-target tracking. Suppose that the two trajectories associated successfully are trajectory i and trajectory j, trajectory i comes from radar, and trajectory j is the trajectory information of the infrared sensor. In the following fusion, it is assumed that the trajectories have been associated, and one target in each sensor multi-target can be analyzed. The prediction covariance matrices of the two sensors are P1 and P2 respectively. The main work of trajectory fusion is to obtain better target state estimation and prediction covariance matrix through fusion algorithm. When performing trajectory fusion, the estimation errors of the two trajectories are independent. Zhang Xudong, Tang Jiaqiao, and Zou Yuan propose to fuse positioning data of multiple sensors such as wheeled odometry, inertial measurement unit, ultra-wideband, and laser radar, and use the extended Kalman filter (EKF) and adaptive Monte-Carlo positioning algorithm to improve the positioning stability of indoor mobile robots. Zhang Jianfan, Ding Sheng, and Chen Jun propose to use the extended Kalman filter to fuse depth vision information and laser radar information, and then use Bayesian estimation method to fuse the information of the two sensors again, thereby reducing the redundant information of the target and improving the accuracy of environment position construction. Liu Chang fuses multi-sensor data of cameras, radars, and automatic identification systems (AIS) to determine the position of a ship in the surrounding environment, and uses the IOU matching algorithm to fuse the position information of radars and AIS and visual perception information, thereby improving the perception accuracy.

[0004] In summary, with the vigorous development of unmanned aerial vehicles, various black flight phenomena emerge in an endless stream, and only relying on a single sensor to track illegal unmanned aerial vehicles faces the defects of low reliability, easy to be disturbed, and difficult to cover all time periods, and a track information fusion model accessing multiple sensors is urgently needed. However, in the existing multi-source track information fusion algorithm, the target is tracked for a long time and the track is generated, and the experience parameters under ideal receiving conditions are relied on, but the tracked target often has multiple maneuvering styles in the actual scene, so that the fusion model based on experience parameters is prone to filter divergence and track distortion. SUMMARY

[0005] The present application is to solve the above-mentioned problems in the prior art, and further proposes a maneuvering frequency adaptive cooperative positioning method based on multi-sensor fusion.

[0006] The technical scheme adopted by the present application to solve the above-mentioned problems is:

[0007] The present application comprises the following steps: step 1, using multiple sensors to continuously receive target position information;

[0008] Step 2, unify the received track of each information source to the CGCS2000 geocentric coordinate system;

[0009] Step 3, performing adaptive frequency filtering on the track of each information source converted to the unified coordinate system;

[0010] Step 4, using an enhanced hierarchical fusion algorithm combined with maneuvering frequency adaptation to fuse the track information of each information source to obtain the comprehensive track of the target.

[0011] Further, in step 1, each sensor is at a different spatial position and has a cooperative relationship, each sensor acts as a node of the fusion network, and the reception frame rate, detection accuracy, tracking time, and coordinate system of the original data all have differences, and can receive position information from different targets.

[0012] Further, in step 2, the position information obtained by all sensor nodes is input into an error conversion model and is converted to the CGCS2000 geocentric coordinate system.

[0013] Further, in step 3, the improved maneuvering frequency adaptive Auto-α model proposes a method of setting multiple corresponding parameters α and σ according to the possible level of target maneuvering 2 In the filtering process, the maneuvering level of the target at the current time is determined according to the filtered value of the acceleration, and the parameters α and σ of the Singer model corresponding to the current time are set 2 In the subsequent tracking process, a new state equation is used for filtering.

[0014] Further, in step 4, the model is based on the hierarchical track fusion algorithm of multi-source filtering information, and the basic idea is to decouple the information from the perspective, identify prior information through the information graph, and then delete the prior information in the fusion algorithm to avoid the double influence of the prior information on the fusion estimation. And finally generate a comprehensive track through multiple information source tracks for a specific target. Specifically, it includes the following steps:

[0015] Step 4.1: sensor data acquisition;

[0016] Step 4.2: track association and state estimation;

[0017] Step 4.3: update state estimation and covariance matrix;

[0018] Step 4.4: repeat the process until the final convergence, and obtain the final target comprehensive track.

[0019] The beneficial effects of the present application are:

[0020] 1. The present application improves the problem that when the target motion information received by various sensors in the actual scene is greatly different from the preset target motion model, the real-time tracking of the target is lost or the filtering divergence is caused due to the failure of the model parameter to update in time, and the precision of the corresponding fusion algorithm is distorted.

[0021] 2. The hierarchical fusion realizes the simultaneous compression of process variables by the central fusion potential precision through the distributed structure, so that the overall system maximally transmits only the track and its covariance of each sensor, and the prediction estimation and its covariance are calculated by the process model, thereby reducing the system transmission and the calculation amount of the central processing.

[0022] 3. The high operation efficiency of the enhanced hierarchical fusion makes the time effectiveness of the adaptive frequency parameter update strong, which realizes the operation speed of the distributed fusion and approximates the fusion precision of the central fusion. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 It is a schematic diagram for multi-sensor cooperative detection scene;

[0024] Figure 2 It is a spatial schematic diagram of CGCS2000 geocentric coordinate system;

[0025] Figure 3 It is a model architecture diagram of the space coordinate system conversion model;

[0026] Figure 4 It is a flowchart of the enhanced hierarchical fusion algorithm based on adaptive frequency. DETAILED DESCRIPTION

[0027] DETAILED DESCRIPTION: The motorized frequency adaptive cooperative positioning method based on multi-sensor fusion in the embodiment includes the following steps.

[0028] Step 1, continuously receiving target position information using multiple sensors, see attached Figure 1 ;

[0029] The following is a description of the specific application scenarios of the application content:

[0030] The target position information is provided by four types of sensors simultaneously, i.e., unmanned reconnaissance positioning device SC-P5000, SC-P5000+, SC-P8000, and radar detection device SC-R3000. The detection accuracy, refresh rate, detection frequency, and detection distance of the sensors are different. The specific input characteristics are shown in the following table:

[0031]

[0032] Table 1: Input characteristics of information sources of multi-source information fusion model

[0033] The data sets used in the application are collected by the above-mentioned sensors.

[0034] The received track of each type of information source is unified to the CGCS2000 geocentric coordinate system, wherein the CGCS2000 geocentric coordinate system is shown in the attached Figure 2 , and the coordinate conversion model is shown in the attached Figure 3 .

[0035] Since the coordinate systems of the original data of various sensors include polar coordinate system and WGS-84 geographic coordinate system, the corresponding coordinate conversion model is needed to convert them to a unified coordinate system.

[0036] The aerodynamic target observed by the WGS-84 geographic coordinate system is represented by longitude, latitude, and height: [L, B, H]. It is converted to the CGCS2000 geocentric coordinate system and represented by XYZ axis values: [X c ,Y c ,Z c ]. The conversion formula is as follows:

[0037]

[0038] X c =(N+H)*cos(B)*cos(L) (3)

[0039] Y c =(N+H)*cos(B)*sin(L) (4)

[0040] Z c =(N*(1-e2)+H)*sin(B) (5)

[0041] Where a, b are the parameters of the approximate ellipsoid of the earth, the long semi-axis a ≈ 6378137.0 m, and the short semi-axis b ≈ 6356752.31414 m.

[0042] The aerodynamic target is described in polar coordinate system, and is expressed by distance, azimuth angle and elevation angle: [R, θ, ε]. First, it is converted to the North-East-Down coordinate system, and is expressed by [X nse ,Y nse ,Z nse ]; and then it is converted to the CGCS2000 geocentric coordinate system. The conversion formula from the polar coordinate system to the CGCS2000 geocentric coordinate system is as follows:

[0043] X nse = R*sin(ε)*cos(θ) (6)

[0044] Y nse = R*cos(ε) (7)

[0045] Z nse = R*sin(ε)*sin(θ) (8)

[0046]

[0047] Where X O , Y O , Z O are the coordinates of the observation site in the CGCS2000 geocentric coordinate system, L O , B O are the coordinates of the observation site in the WGS-84 geographic coordinate system.

[0048] (3) performing adaptive frequency filtering on the tracks of each information source converted to the unified coordinate system

[0049] The method of the application improves the traditional global statistical model depending on parameters into an Auto-α model which can describe multiple maneuvering states and is adaptive to parameters. The specific process of the improvement is as follows:

[0050] The parameter α in the Singer model is 1 / τ m , which is the inverse of the maneuvering time constant τ m , and the parameter σ 2 is the variance of the target acceleration. According to the suggestion of Singer, τ m = 60 s for slow turning of an airplane; τ m = 20 s for evading maneuver; and τ m = 1 s for atmospheric disturbance. Singer also makes the following assumption on the distribution of the maneuvering acceleration a(t): the probability of a(t) being 0 is P0, and the probability of a(t) being the maximum acceleration ±amax The probability of P max , and in the above it obeys uniform distribution. According to this assumption, we can get:

[0051]

[0052] The determination of parameters α and σ 2 in Singer model depends on prior knowledge, and once determined, they will not change in the filtering process. When the prior determined α and σ 2 have a large difference from the actual situation, the filtering accuracy will become lower. To make up for this shortcoming of Singer model, according to the possible level of target maneuver, we set multiple groups of corresponding parameters α and σ 2 , and in the filtering process, we determine the maneuver level of the target at the current time according to the filtered value of acceleration, and set the corresponding parameters α and σ 2 of Singer model at the current time, and then use the new state equation to filter in the subsequent tracking process. The specific process is as follows:

[0053] Since the state transition matrix φ and the noise matrix Q both contain the maneuver frequency α. Therefore, by changing α, we can adjust φ and Q to make them closer to the true state of the target.

[0054] According to experience, the value range of α is: turning maneuver α = 1 / 60, escape maneuver α = 1 / 20, atmospheric disturbance α = 1. Usually when performing state filtering, we determine the value of α in advance according to experience.

[0055] The information vector of the target is:

[0056] d(k) = Z(k) - H(k)X(k|k-1) (12)

[0057] d(k) is a mean Gaussian white noise process, and its covariance matrix is:

[0058] S(k) = H(k)P(k|k-1)H T (k) + R(k) (13)

[0059] Define the distance function as:

[0060] D(k) = d T (k)S -1 (k)d(k) (14)

[0061] From the statistical properties of information sequence, we know that D(k) obeys χ 2 distribution with degree of freedom m. If the target maneuvers, the new information d(k) will not be zero mean Gaussian white noise, and D(k) will become larger, so we can use the following method to detect the occurrence and elimination of maneuver. Take the probability of D(k) being greater than a certain threshold M as Pf ,Right now:

[0062] P{D(k)>M}=P f (15)

[0063] In the formula, P f Let α be the allowable false alarm probability. Then, the adaptive choice of α is:

[0064] When D(k) > M, it indicates that the target maneuver has occurred, and the value of α is increased; when D(k) < M, it indicates that the target maneuver has been eliminated, and the value of α is decreased; in engineering, to simplify the value of α, we perform a simplified calculation:

[0065]

[0066] During target tracking, an empirical value of 1 / 20 can be predetermined for α. When adjusting α, the distance function of the state is not estimated at every step. Generally, the distance function of the state is estimated and α is adjusted only after 3 to 5 steps.

[0067] (4) Using an enhanced hierarchical fusion algorithm that combines adaptive maneuvering frequency, the trajectory information from various information sources is fused to obtain the target's comprehensive trajectory, as shown in the appendix. Figure 4

[0068] Filter fusion is an information fusion technique that combines information from multiple sensors or data sources to optimize the estimated state of a target. Its core principle is to use filtering algorithms, such as extended Kalman filtering, unscented Kalman filtering, or particle filtering, to combine the measurement values ​​from different sensors or data with their error characteristics. Through dynamic models and state transition equations, prediction and correction are performed to obtain a more accurate and stable estimate of the target state. The steps are as follows:

[0069] Step 4.1: Sensor data acquisition: Multiple sensors are typically used to collect environmental information, including various UAV detection and positioning devices and radar detection devices, etc.

[0070] Step 4.2: Track Association and State Estimation: Sensor i and sensor j are associated and track the same target. At time k, the state estimate of sensor i is obtained through filtering. Its error covariance is P i (k), the state estimate of sensor j is Its error covariance is P j (k), and send these tracks to the fusion center.

[0071] Step 4.3: Update the state estimate and covariance matrix:

[0072] Suppose there are N sensors tracking the same target and the information Kalman filter is used, the state estimation and the corresponding error covariance of sensor i at time k can be expressed as:

[0073]

[0074] The so-called information Kalman filter is to recursively calculate the inverse of the covariance matrix in both prediction and update stages.

[0075] Suppose the sensor track is the same as above, and the hierarchical fusion algorithm is:

[0076]

[0077] Wherein, And The local priori estimation information of the system and sensor i respectively.

[0078] Step 4.4: Repeat the iteration until the final convergence is obtained, and the final target comprehensive track is obtained.

[0079] Embodiment:

[0080] The application provides a kind of motor frequency self-adapting cooperative positioning method based on multi-sensor fusion, further illustrated by combining example and drawing, specific process is as follows:

[0081] 1. Multiple sensors form a distributed fusion network, receive the unmanned aerial vehicle track information in a certain airspace, see the attached Figure 1 ;

[0082] 2. Each sensor has a specific receiving frame rate, tracking time and detection accuracy;

[0083] 3. The target information received by each node of the fusion network is transmitted back to the fusion center according to frame data, and after one node sends a frame of data, it waits for the destination node to return the frame data before sending the next frame of data;

[0084] 4. When the measurement coordinate system of a node is WGS-84 geographic coordinate system, the input data format is [L, B, H], and when the measurement coordinate system is polar coordinate system, the input data format is [R, θ, ε];

[0085] 5. All data is unified to CGCS2000 geocentric coordinate system through coordinate system conversion model. See the attached Figure 3 .

[0086] 6. Set the initial value of the maneuvering frequency α, and select a suitable filter model;

[0087] 7. Obtain the information vector of the target, denoted as d(k) and its covariance S(k), define the distance function D(k) and the maneuvering threshold M;

[0088] 8. Updating the parameter a according to the real-time relationship between D(k) and M;

[0089] 9. Dividing the track information of each type according to targets through a track association model, and performing subsequent fusion operation on the multi-source track cluster of a single target;

[0090] 10. Assuming that the initial state estimation of the i-th sensor and the j-th sensor in the n sensor nodes is represented as X i = [x 1i , x 2i , …, x ni ] T and X j = [x 1j , x 2j , …, x nj ] T , and the covariance is P ii and P jj , respectively;

[0091] 11. Selecting the information Kalman filter which recursively calculates the inverse of the covariance matrix in both prediction and update stages as the filter;

[0092] 12. Calculating the state error cross-covariance matrix P ij (k) in the maximum likelihood sense at the current time;

[0093] 13. At time k, updating the state estimation of the i-th sensor through filtering as and the error covariance as P i (k), updating the state estimation of the j-th sensor as and the error covariance as P j (k), and sending these tracks to the fusion center;

[0094] 14. Continuously updating the state estimation of the i-th sensor at the next time based on the Kalman filter as and the corresponding error covariance as

[0095]

[0096] 15. Repeating steps 12 to 14 multiple times to obtain the final state estimation of the i-th sensor, denoted as see the attached Figure 4 ;

[0097] 16. Superimposing the state updating processes of multiple sensors on the comprehensive track estimation through the enhanced hierarchical fusion model, denoted as the final comprehensive track of the current target.

[0098] The present application is especially used for multi-sensor receiver to integrate the position information of the maneuvering target into a more precise and stable comprehensive track through a track fusion network. The fusion process can make full use of the advantages of various sensors and consider various motion states of the maneuvering target.

[0099] The above is only the preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with the preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, without departing from the technical solution of the present application. Any simple modification, equivalent replacement and improvement of the above embodiments, which are within the scope of the technical solution of the present application and the spirit and principles of the present application, are still within the protection scope of the present application.

Claims

1. A cooperative localization method based on multi-sensor fusion and adaptive maneuver frequency, characterized in that, Includes the following steps: Step 1: Continuously receive target location information using multiple sensors; Step 2: Unify the received track data from various information sources into the CGCS2000 geocentric coordinate system; Step 3: Perform adaptive frequency filtering on the tracks of each information source transformed to a unified coordinate system, specifically including: The traditional global statistical model is improved into an Auto-α model with adaptive parameters capable of describing various maneuver states. The specific process of the improvement is as follows: Parameters in the Singer model It is the maneuver time constant The reciprocal of the parameter It is the variance of the target acceleration; for slow turns of the aircraft =60s; for evasive maneuvers =20s, for atmospheric disturbances, =1s; for acceleration during maneuvering The distribution of is made as follows: The probability of being 0 is , Maximum acceleration The probability is Furthermore, it follows a uniform distribution on ; based on this assumption, we can obtain: Parameters in the Singer model and The determination of depends on prior knowledge, and once determined, it does not change during the filtering process; when the prior knowledge is determined... and When there is a significant difference from the actual situation, multiple sets of corresponding parameters are set according to the probability level of the target maneuver. and During the filtering process, the target's maneuverability level at the current moment is determined based on the filtered acceleration value, and the parameters of the Singer model at the current moment are set accordingly. and In the subsequent tracking process, a new state equation filtering method is used; the specific process is as follows: Due to the state transition matrix and noise matrix All contain maneuver frequencies Therefore, by changing To achieve adjustment and The purpose is to make it closer to the true state of the target; Based on experience, The value range is: turning maneuver =1 / 60, evasive maneuver =1 / 20, Atmospheric disturbance =1, which is usually determined in advance based on experience when performing state filtering. The value; Target information vector for: It is a mean Gaussian white noise process, and its covariance matrix is: Define the distance function as: According to the statistical properties of information sequences, Obeying the degree of freedom of Distribution; if the target maneuvers, the information vector It will not be zero-mean Gaussian white noise. It will become larger, therefore the following methods can be used to detect and eliminate the occurrence of maneuvers; take Greater than a certain threshold The probability is ,Right now: In the formula, The allowable false alarm probability; at this point, The adaptive selection is: when At this time, it indicates that the target maneuver has occurred, increasing the risk. The value of ; when At this time, it indicates that the target's maneuver has been eliminated, reducing the risk of injury. The value of ; Calculation: During target tracking, pre-defined... The experience value is 1 / 20, proceed During the adjustment, the distance function of the state is estimated in 3-5 steps, and then... Adjustments; Step 4: Utilizing an enhanced hierarchical fusion algorithm that incorporates adaptive maneuvering frequency, the track information from various information sources is fused. Prior information is identified through the information map, and then removed from the fusion algorithm. Finally, a comprehensive track is generated from multiple information source tracks targeting a specific target. This process includes: Step 4.1: Sensor Data Acquisition: Use multiple sensors to collect environmental information, including various UAV detection and positioning devices and radar detection equipment, and record the sensor data. i exist k The target observation value obtained at time 1 is ; Step 4.2: Track Association and State Estimation: Sensors i and sensors j Related, tracking the same target, in k At any given time, the sensor value is obtained through filtering. i The state estimate is Its error covariance is ,sensor j The state estimate is Its error covariance is And send these flight paths to the integration center; Step 4.3: Update the state estimate and covariance matrix: Assume there is N When multiple sensors track the same target, and an information-based Kalman filter is used, the result is obtained... k Time sensor i The state estimate and the corresponding error covariance are expressed as: Based on sensors i and sensors j If we assume that they are related and track the same target, then the hierarchical fusion algorithm is as follows: In the formula, , and , Systems and sensors i Local prior estimation information; Step 4.4: Iterate through steps 4.1 to 4.3 until convergence is achieved, and obtain the final target composite trajectory.

2. The cooperative localization method based on multi-sensor fusion and adaptive maneuver frequency as described in claim 1, characterized in that, In step 1, the sensors are located in different spatial positions and cooperate with each other. Each sensor, as a node of the fusion network, has different receiving frame rate, detection accuracy, tracking time, and coordinate system of the original data, and can receive position information from different targets.

3. The cooperative localization method based on multi-sensor fusion and adaptive maneuver frequency as described in claim 1, characterized in that, In step 2, the location information acquired by all sensor nodes is input into the error transformation model and uniformly transformed to the CGCS2000 geocentric coordinate system.

4. The cooperative localization method based on multi-sensor fusion and adaptive maneuver frequency as described in claim 1, characterized in that, In step 3, the improved maneuver frequency adaptive Auto-α model proposes a method that sets multiple sets of corresponding parameters based on the possible levels of the target maneuver. and During the filtering process, the target's maneuverability level at the current moment is determined based on the filtered acceleration value, and the parameters of the Singer model at the current moment are set accordingly. and A new state equation filtering method is used in subsequent tracking processes.

Images