Maneuvering group target intelligent tracking method based on deep neural network

Abstract

The invention discloses a deep neural network-based maneuvering group target intelligent tracking method, which comprises the following steps of: introducing deep learning into a Bayesian filtering framework, and respectively extracting group target motion characteristics and process noise statistical characteristics from measurement data based on a plurality of deep neural networks; a group target motion state transition matrix and a process noise variance matrix are estimated online, and then the motion state of the mass center of the group target is estimated with high precision through Bayesian filtering; a group target contour is modeled as an ellipse of which the center is located at the mass center of the group target, so that the estimation precision of the group target contour is improved by using a high-precision group target mass center motion state estimation result, and finally, the tracking of the group target is completed. Compared with the existing group target tracking method, the method provided by the invention does not need priori information to pre-establish a target motion model, and can implement high-precision tracking on the non-cooperative maneuvering group target lacking priori information in a complex environment.

Description

Technical Field

[0001] The present invention relates to the field of target tracking, and in particular to a method for intelligent tracking of maneuvering group targets based on a deep neural network. Background Art

[0002] In recent years, with the in-depth development of computer, communication, navigation and other technologies, cluster weapon systems such as swarm drones and multiple salvo missile groups have gradually been applied. On the one hand, the above-mentioned cluster weapon systems contain a large number of combat units, which quickly saturates the detection, tracking and interception capabilities of the air defense system, making it difficult to effectively intercept them; on the other hand, the combat units in the above-mentioned cluster weapon systems have a high degree of coordination. After a small number of combat units are damaged, the remaining units can quickly fill in, which will not significantly reduce the overall combat capability of the cluster, making it difficult to effectively damage them. Therefore, it is difficult to defend and intercept the above-mentioned cluster weapon systems, which has become the main challenge that modern air defense systems must deal with.

[0003] Accurately tracking the above-mentioned group targets and obtaining the change history and trend of the group target motion state and the group target contour shape over time are necessary prerequisites for subsequent weapon systems to formulate defense strategies, solve attack elements and implement effective interception. At present, for group targets with a large number of targets and dense distribution, in order to improve the real-time performance of the tracking system and avoid saturation of tracking capabilities, the group overall tracking method is mainly adopted, that is, the motion state of the group center of mass and the shape of the group contour are estimated in turn. In terms of the estimation of the group center of mass motion state, the existing methods and technologies need to establish a motion model of the tracked target in advance, which belongs to the model-driven target tracking method. However, when tracking non-cooperative targets in a complex battlefield environment, due to the lack of prior information, the variable target motion mode, the unknown statistical characteristics of the process noise and even the time-varying factors, the pre-established motion model may be mismatched, which leads to a significant decrease in target tracking accuracy and serious degradation of tracking performance. Therefore, the existing model-driven group target tracking methods and technologies cannot accurately estimate the state of the above-mentioned non-cooperative maneuvering group targets in a complex environment. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent tracking method for mobile group targets based on deep neural networks to solve the problem of high-precision tracking of mobile group targets in complex environments such as variable group target motion modes, lack of target prior information, and time-varying statistical characteristics of process noise.

[0005] The technical solution to achieve the purpose of the present invention is as follows: In the first aspect, the present invention provides a method for intelligent tracking of mobile group targets based on a deep neural network, comprising the following steps:

[0006] The first step is to obtain the group target measurement information at the current moment;

[0007] The second step is to estimate the motion state of the center of mass of the group targets at the current moment;

[0008] The third step is to estimate the contour shape of the group target at the current moment;

[0009] The fourth step is to predict the motion state of the center of mass of the group targets at the next moment;

[0010] The fifth step is to predict the contour shape of the group target at the next moment.

[0011] In a second aspect, the present invention provides a mobile group target intelligent tracking system based on a deep neural network, which is used to implement the method described in the first aspect, and the system includes:

[0012] The first module is used to obtain the group target measurement information at the current moment;

[0013] The second module is used to estimate the motion state of the center of mass of the group targets at the current moment;

[0014] The third module is used to estimate the contour shape of the group target at the current moment;

[0015] The fourth module is used to predict the motion state of the center of mass of the group targets at the next moment;

[0016] The fifth module is used to predict the contour shape of the group target at the next moment.

[0017] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the program.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0019] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which implements the steps of the method described in the first aspect when executed by a processor.

[0020] Compared with the existing group target tracking methods, the present invention has the following significant advantages:

[0021] (1) When the motion mode of the center of mass of a group target is changeable and the target prior information is unknown, the present invention uses a deep neural network to extract the target motion characteristics from the group target measurement data online, and then estimates the motion state transfer matrix in real time, which solves the problem that the pre-established motion model is difficult to always match the actual motion mode of the maneuvering target without prior information, and effectively improves the estimation accuracy of the motion state of the center of mass of the group target.

[0022] (2) In a complex environment where the statistical characteristics of process noise are unknown or even time-varying, the present invention uses a deep neural network to extract the statistical characteristics of process noise from the group target measurement data, and then estimates the process noise variance matrix in real time, which solves the problem that the statistical characteristics of process noise are difficult to accurately obtain in a complex environment, and further effectively improves the estimation accuracy of the motion state of the center of mass of the group target.

[0023] (3) The present invention simplifies the modeling of the group target contour as a time-varying ellipse centered at the group target's center of mass. On the one hand, it can improve the real-time performance of the group target tracking algorithm. On the other hand, it can use the high-precision estimation result of the motion state of the group target's center of mass to improve the estimation accuracy of the group target contour morphology. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Flowchart of the group target tracking method based on deep neural network.

[0025] Figure 2 Schematic diagram of the group target contour shape, centroid and measurement data.

[0026] Figure 3 This is the overall structure diagram of the motion state transfer matrix estimation model.

[0027] Figure 4 This is the overall structure diagram of the process noise variance matrix estimation model.

[0028] Figure 5 Schematic diagram of group target motion trajectory and measurement data in a typical scenario.

[0029] Figure 6 Schematic diagram of group target tracking results in a typical scenario. DETAILED DESCRIPTION

[0030] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] The present invention proposes an intelligent tracking method for mobile group targets based on deep neural networks, which can solve the problem of non-cooperative mobile group target tracking in complex environments such as variable target motion modes, unknown target prior information, and unknown process noise variance. The estimation accuracy of the non-cooperative mobile group target state is significantly better than the existing group target tracking method that requires prior information to pre-establish the target motion model.

[0032] The group target intelligent tracking method based on deep neural network proposed in the present invention uses deep neural network to extract target motion characteristics and process noise statistical characteristics from group target measurement data in real time, and then estimates the motion state transfer matrix and process noise variance matrix online, and then estimates the group target centroid motion state and group target contour morphology in turn through Bayesian filtering, and finally completes group target tracking. The implementation process of the mobile group target intelligent tracking method based on deep neural network proposed in the present invention is as follows: Figure 1 As shown, the specific implementation is as follows:

[0033] Step 1: Get the current group target measurement information

[0034] (1.1) Figure 2 As shown in the figure, for a group target with densely distributed units in the group, the detection device can obtain multiple measurement data generated by the units in the group at each sampling time, thereby forming a measurement data set Where N k is the number of measurement data obtained at time k. Due to factors such as mutual occlusion among units in the group, N k Time-varying, that is, the number of measurement sources varies with time. Therefore, the measurement data z in the measurement data set k,l It can be regarded as being generated by the lth measurement source appearing at time k, and the measurement equation is as follows:

[0035] z k,l =h(x k,l )+v k,l (1) where h(·) represents the measurement function; x k,l represents the state vector of the lth measurement source; v k,l To measure noise.

[0036] (1.2) Based on the above measurement data set, the center of mass measurement information of the group target at time k is calculated as follows:

[0037]

[0038] Step 2: Estimate the motion state of the center of mass of the group target at the current moment

[0039] The cubature Kalman filter method is used to estimate the motion state of the center of mass of the group target. The specific steps are as follows:

[0040] (2.1) Calculation of measurement volume points

[0041]

[0042] Where j = 1, 2, ..., 2n x , n x is the motion state vector of the group target center of mass The dimension of is the square root factor of the prior error covariance of the group target center of mass motion state; Φ j is the jth element in the set Φ, which is defined as follows

[0043]

[0044] (2.2) Calculate the prior measurement of the mass center of the group target

[0045]

[0046] (2.3) Calculate the square root factor of the prior innovation covariance

[0047]

[0048] Where Chol(·) represents Cholesky decomposition; R k is the covariance of the measurement noise; Tria(A) represents the transposed matrix of the upper triangular matrix obtained by orthogonal decomposition of the column full rank matrix A.

[0049] (2.4) Calculate the cross-covariance

[0050]

[0051] (2.5) Using the centroid measurement information of group targets Estimating the center of mass motion state

[0052]

[0053] G k is the filter gain matrix.

[0054] Step 3: Estimate the current group target contour shape

[0055] To ensure that the group target tracking algorithm has high real-time performance, such as Figure 1 As shown, in a two-dimensional scene, the contour of the group target is simplified and modeled as a time-varying ellipse as shown in the following formula, and the measurement sources are all located inside the ellipse

[0056]

[0057] Where y is the coordinate vector of any point on the elliptical contour of the group target; E k is the parameter matrix of the ellipse, which is a symmetric positive definite random matrix, and its transition probability distribution function is shown as follows

[0058]

[0059] In the formula, δ is the degree of freedom; A k-1is the transfer matrix; W(Y; a, C) represents the symmetric positive definite random matrix C that satisfies the Wishart distribution with a degree of freedom and a parameter matrix Y.

[0060] Based on the above modeling method, the group target centroid motion state information obtained in the second step is used to estimate the contour of the group target at the current moment through Bayesian filtering, as shown in the following formula:

[0061]

[0062] is the group centroid measurement distribution covariance matrix, D k|k-1 It is an intermediate calculation parameter matrix with no clear physical meaning. k The error matrix for modeling the extended target contour morphology, in is the degree of freedom parameter in the inverse Wishart distribution;

[0063] If the tracking task for the group target has been completed, the group target tracking algorithm is exited; otherwise, the fourth step is continued.

[0064] Step 4: Predict the motion state of the group target center of mass at the next moment

[0065] Using the estimated results of the group target center of mass motion state at the current time k, the group target center of mass motion state at the time k+1 is predicted. The specific steps are as follows:

[0066] (4.1) Estimation of motion state transfer matrix based on deep neural network

[0067] like Figure 3 As shown in the figure, the group target centroid measurement data from time k-w+1 (w is the sliding window length) to time k is used as a time series input through a dynamic sliding window method, and the motion characteristics of the group target centroid are extracted from it using a deep neural network, thereby outputting the key parameters in the state transfer matrix. First, the centroid measurement data is converted to the coordinate system where the centroid motion state is located; then the deviation is normalized, and then convolution filtering is performed to extract the features in the data; then the bidirectional long short-term memory (Bi-LSTM) network is used to extract the relevant motion characteristic information of the group target centroid from the time series; finally, the fully connected layer is used to output the key parameters in the motion state transfer matrix.

[0068] Without loss of generality, if the group target center of mass motion state vector is When (p x 、p y are the coordinates of the mass center of the group target on the x-axis and y-axis of the Cartesian coordinate system, v x 、v yare the speeds of the center of mass of the group target in the x-axis and y-axis directions respectively), and the estimation result of the motion state transfer matrix obtained based on the deep neural network is shown as follows

[0069]

[0070] In the formula, f 1,k and f 2,k The two parameters determine the displacement increment of the mass center of the group target in the x-axis and y-axis directions in the two-dimensional plane; f 3,k and f 4,k The two parameters determine the speed of the center of mass of the group targets in the x-axis and y-axis directions in the two-dimensional plane.

[0071] (4.2) Estimation of process noise variance matrix based on deep neural network

[0072] like Figure 4 As shown in the figure, the measurement data of the mass center of the group target from time k-w+1 (w is the sliding window length) to time k is input through a dynamic sliding window method, and then converted to the coordinate system where the mass center motion state is located; then the deviation is normalized, and then the features in the data are extracted through convolution filtering; next, on the one hand, the second-order difference is used to determine whether the motion mode of the mass center of the group target has changed, and then the motion characteristics of the mass center of the group target are extracted from the time series through a multi-layer Bi-LSTM network; then the multi-layer perceptron and the normalized exponential function are used to output the probability of its motion mode matching models such as constant velocity motion (CV) and constant turning rate (CT), and the structural form of the process noise variance matrix is ​​selected according to the maximum matching probability; on the other hand, the statistical characteristics of the process noise are extracted from the time series through a multi-layer Bi-LSTM network, and then the process noise intensity estimation result is output through the fully connected layer; the above results are combined to obtain the estimation result of the process noise variance matrix.

[0073] For example, when Figure 4 When the deep neural network in the above example judges that the motion mode of the group target center of mass conforms to the CV model, the estimated result of the process noise variance matrix is ​​as follows:

[0074]

[0075] Where T is the sampling period of the detection equipment; is the process noise intensity estimation result.

[0076] For example, when Figure 4 When the deep neural network in the above example judges that the motion mode of the group target center of mass conforms to the CT model, the estimated process noise variance matrix is ​​as follows:

[0077]

[0078] (4.3) Predicting the motion state of the center of mass of group targets

[0079] State transfer matrix estimation results obtained using deep neural network Estimation results of process noise variance matrix Based on the cubature Kalman filter method, the motion state of the center of mass of the group targets at time k+1 is predicted. The specific steps are as follows:

[0080] (4.3.1) Calculation of state volume points

[0081]

[0082] (4.3.2) Predicting the motion state of the center of mass of a group of targets

[0083]

[0084] Step 5: Predict the contour of the group target at the next moment

[0085] Using the estimation result of the group target contour at the current time k, the group target contour at the time k+1 is predicted as shown in the following formula:

[0086]

[0087] Where d is the estimated result of the group target contour parameter matrix The dimension of k is the ellipse contour parameter matrix E k Parameter that affects the variance of measurement noise.

[0088] After completing the above steps, return to the first step, wait for the next sampling moment to start, and then continue to track the group targets according to the above steps.

[0089] Example

[0090] First, according to the characteristics of the group targets being tracked, the training set and the validation set are generated. The target motion modes include constant speed motion, constant rate turning, etc. Figure 3 In the motion state transfer matrix estimation model shown in the figure, the convolution layer consists of two-dimensional convolution and one-dimensional convolution, and the Bi-LSTM network has three layers. During the training process, the mean square error loss function is selected as the loss function, and the Adam optimizer is used. Figure 4 In the process noise variance matrix estimation model shown in the figure, the convolution layer consists of two-dimensional convolution and one-dimensional convolution, and the Bi-LSTM network has 2 layers. The cross entropy loss function is used for training the Bi-LSTM network for judging the motion mode, and the mean square error loss function is used for training the Bi-LSTM network for estimating the process noise intensity, and both use the Adam optimizer.

[0091] The initial motion state of the group target center of mass is The mass center of the group target makes uniform linear motion from 0 to 60 seconds, makes constant rate turning motion from 61 to 120 seconds, makes uniform linear motion from 121 to 180 seconds, makes constant rate turning motion from 181 to 240 seconds, and makes uniform linear motion from 241 to 300 seconds; the sampling period of the detection equipment is 1 second.

[0092] The actual motion trajectory of the group target and the measurement data at each sampling time are as follows: Figure 5 As shown in the figure, the tracking results of the group target by the intelligent tracking method of mobile group target based on deep neural network proposed in the present invention are as follows: Figure 6 As shown; according to the above experimental results, the method proposed in the present invention can track mobile group targets in real time with high precision.

[0093] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent tracking of mobile group targets based on deep neural network, characterized in that: The following steps are involved: The first step is to obtain the group target measurement information at the current moment; The second step is to estimate the motion state of the center of mass of the group targets at the current moment; The third step is to estimate the contour shape of the group target at the current moment; The fourth step is to predict the motion state of the center of mass of the group targets at the next moment; The fifth step is to predict the contour shape of the group target at the next moment.

2. The method for intelligent tracking of mobile group targets based on deep neural network according to claim 1 is characterized in that: The first step is: 1) For a group target with densely distributed units within the group, the detection device can obtain multiple measurement data generated by the units within the group at each sampling time, that is, to form a measurement data set Where N k is the number of measurement data obtained at time k, and the measurement data z k,l Generated by the lth measurement source appearing at time k, the measurement equation is as follows z k,l =h(x k,l )+v k,l In the formula, h(·) represents the measurement function, x k,l represents the state vector of the lth measurement source, v k,l represents the measurement noise; 2) According to the above measurement data set, the center of mass measurement information of the group target at time k is calculated as follows:

3. The method for intelligent tracking of mobile group targets based on deep neural network according to claim 2 is characterized in that: The second step is as follows: The motion state of the center of mass of a group of targets is estimated based on the cubature Kalman filter method. The specific steps are as follows: 1) Calculate and measure volume points Where j = 1, 2, ..., 2n x , n x is the group target center of mass motion state vector The dimension of is the square root factor of the prior error covariance of the group target center of mass motion state; Φ j is the jth element in the set Φ, which is defined as follows 2) Calculate the prior measurement of the center of mass of the group target 3) Calculate the square root factor of the prior innovation covariance Where Chol(·) represents Cholesky decomposition; R k is the covariance of the measurement noise; Tria(A) represents the transposed matrix of the upper triangular matrix obtained by orthogonal decomposition of the column full rank matrix A; 4) Calculate the cross-covariance 5) Using the centroid of group targets for measurement Estimating the center of mass motion state 4. The method for intelligent tracking of mobile group targets based on deep neural network according to claim 3 is characterized in that: The third step is as follows: In a two-dimensional scene, the group target contour shape is simplified and modeled as a time-varying ellipse, as shown in the following formula: Where y is the coordinate vector of any point on the elliptical contour of the group target; E k The parameter matrix of the ellipse is a symmetric positive definite random matrix, and its transition probability distribution function is shown as follows In the formula, δ is the degree of freedom; A k-1 is the transfer matrix; W(Y; a, C) represents the symmetric positive definite random matrix C that satisfies the Wishart distribution with a degree of freedom and a parameter matrix Y; The current moment group target contour morphology is estimated through Bayesian filtering, as shown in the following formula: in, is the group centroid measurement distribution covariance matrix, D k|k-1 is the intermediate calculation parameter matrix, B k The error matrix for modeling the extended target contour morphology, in d is the estimation result of the group target shape profile parameter matrix The dimension of is the degree of freedom parameter in the inverse Wishart distribution; If the tracking task for the group target has been completed, the group target tracking algorithm is exited; otherwise, the fourth step is continued.

5. The method for intelligent tracking of mobile group targets based on deep neural network according to claim 4 is characterized in that: The fourth step is as follows: Using the estimated results of the group target center of mass motion state at the current time k, the group target center of mass motion state at the time k+1 is predicted. The specific steps are as follows: 1) Estimating the state transfer matrix based on deep neural network The group target mass center measurement data from time k-w+1 to the current time k is used as a time series input through a dynamic sliding window method, w is the sliding window length, and the deep neural network is used to extract the group target mass center motion characteristics from it, and output the key parameters in the state transfer matrix; first, the mass center measurement data is converted to the coordinate system where the mass center motion state is located; then the deviation is normalized, and then the features in the data are extracted through convolution filtering; next, the Bi-LSTM network is used to extract the relevant motion characteristics of the group target mass center from the time series; finally, the fully connected layer is used to output the key parameters in the state transfer matrix; 2) Estimating process noise variance based on deep neural network The centroid measurement data from time k-w+1 to time k are transformed into a time series input by a dynamic sliding window, and then converted to the coordinate system where the centroid motion state is located, and then the deviation is normalized, and then the features in the data are extracted by convolution filtering; next, the second-order difference is used to determine whether the group target motion mode has changed, and then the group target motion characteristics are extracted from the time series through a multi-layer Bi-LSTM network, and then the multi-layer perceptron and normalized exponential function are used to output the probability of the group target centroid motion mode matching the constant speed motion and constant turning rate model, and the structural form of the process noise variance matrix is ​​selected according to the maximum matching probability; the process noise characteristics are extracted from the time series through a multi-layer Bi-LSTM network, and then the process noise intensity estimation result is output through a fully connected layer; finally, the results of the two aspects are combined to obtain the estimation result of the process noise variance matrix; 3) Predict the motion state of the center of mass of group targets State transfer matrix estimation results obtained using deep neural network Estimation results of process noise variance matrix Based on the cubature Kalman filter method, the motion state of the center of mass of the group targets at time k+1 is predicted. The specific steps are as follows: (3.1) Calculation of state volume points (3.2) Predicting the motion state of the center of mass of group targets 6. The method for intelligent tracking of mobile group targets based on deep neural network according to claim 5 is characterized in that: The fifth step is as follows: Using the estimation result of the group target contour at time k, the group target contour at time k+1 is predicted as shown in the following formula: In the formula, λ k is the ellipse contour parameter matrix E k Parameters that affect the variance of measurement noise; After completing the above steps, return to the first step, wait for the next sampling moment to start, and then continue to follow the above steps to complete the tracking of the group target.

7. A mobile group target intelligent tracking system based on deep neural network, characterized in that: For implementing the method described in any one of claims 1 to 6, the system comprises: The first module is used to obtain the group target measurement information at the current moment; The second module is used to estimate the motion state of the center of mass of the group targets at the current moment; The third module is used to estimate the contour shape of the group target at the current moment; The fourth module is used to predict the motion state of the center of mass of the group targets at the next moment; The fifth module is used to predict the contour shape of the group target at the next moment.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of any method described in claims 1-6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

Images