A three-dimensional spatial configuration tracking method, device and medium for formation group targets

Abstract

The embodiments of this specification disclose a three-dimensional spatial configuration tracking method, device and medium for formation group targets, which relate to the field of radar target tracking technology and are used to solve the problem of low accuracy of structure estimation that is difficult to adapt to the dynamic changes of formation target structures. The method includes: modeling the three-dimensional spatial structure of the current formation group target according to a preset number of substructures with three-dimensional orientation angles to obtain a three-dimensional structure model; predicting the posterior distribution of each variable at the previous moment according to the dynamic model to obtain a predicted distribution; determining the observation equation corresponding to the current formation group target, performing multi-layer linearization processing on the observation equation in a linearization processing method based on multi-level Taylor expansion to obtain a parameter update formula for the three-dimensional structure model; updating and correcting the parameters of the current formation group target in an iterative manner based on the parameter update formula and the predicted distribution to achieve tracking and three-dimensional configuration estimation of the current formation group target.

Description

Technical Field

[0001] The present invention relates to the field of radar target tracking technology, and in particular to a method, device and medium for tracking the three-dimensional spatial configuration of a formation group of targets. Background Art

[0002] The spatial position and three-dimensional configuration of formation group targets are important components of cluster situation. Accurately estimating the spatial position and three-dimensional configuration of formation group targets is the key to building cluster situation and interpreting cluster intention. The main goal of formation group target positioning and three-dimensional structure reconstruction is to use multiple measurements containing three-dimensional position information to jointly estimate the motion state and spatial structure of the group targets. Through the effective integration and analysis of these data, the real-time state and distribution form of the formation group targets in three-dimensional space can be more accurately depicted to achieve real-time tracking.

[0003] However, there are many problems that need to be solved in the current tracking process of formation group targets. On the one hand, from the perspective of substructure modeling, the traditional multi-substructure method only uses a single random matrix to model each substructure, so that the substructure size, that is, the extension axis length and orientation are coupled together, and cannot adapt to changes in the substructure direction. On the other hand, most of the existing group tracking models and algorithms are aimed at two-dimensional scenes and cannot be directly upgraded to three-dimensional scenes, which makes it impossible to fully reconstruct the three-dimensional spatial structure of cluster targets. Therefore, there is an urgent need for a robust estimation method for the three-dimensional configuration of formation group targets suitable for three-dimensional scenes, which can adapt to the dynamic changes of the formation group target structure and accurately estimate the three-dimensional structure of the formation group target. Summary of the invention

[0004] In order to solve the above technical problems, one or more embodiments of the present specification provide a three-dimensional spatial configuration tracking method, device and medium for formation group targets, which are used to accurately estimate the three-dimensional orientation, spatial position and size of the complex structure of dynamic formation group targets.

[0005] One or more embodiments of this specification adopt the following technical solutions:

[0006] One or more embodiments of this specification provide a method for tracking the three-dimensional spatial configuration of a formation group target. The method includes:

[0007] Modeling the three-dimensional spatial structure of the current formation group target according to a preset number of substructures with three-dimensional orientation angles to obtain a three-dimensional structural model of the current formation group target; wherein the three-dimensional structural model includes: the three-dimensional structural state, prior distribution, dynamic model and measurement model of the current formation group target;

[0008] Predicting the posterior distribution of each variable in the three-dimensional structural model at a previous moment according to the dynamic model to obtain a predicted distribution;

[0009] Determine the observation equation corresponding to the current formation group target according to the relationship between the three-dimensional structure state and the measurement model, perform multi-layer linearization processing on the observation equation in a linearization processing method based on multi-level Taylor expansion, and obtain the parameter update formula of the three-dimensional structure model;

[0010] Based on the parameter updating formula and the predicted distribution, the parameters of the current formation group target are updated and corrected in an iterative manner to achieve tracking and three-dimensional configuration estimation of the current formation group target.

[0011] Optionally, in one or more embodiments of the present specification, modeling the three-dimensional spatial structure of the current formation group target according to a preset number of substructures with three-dimensional orientation angles to obtain the three-dimensional structural model of the current formation group target specifically includes:

[0012] Dividing the current formation group target into substructures to obtain a preset number of substructures with three-dimensional orientation angles, determining scalar angle random variables of the substructures to process the substructures based on a random matrix to obtain a three-dimensional structural state of the current formation group target;

[0013] Obtaining the prior distribution form of each variable in the process of determining the three-dimensional structure state;

[0014] Evolving the motion state of the current formation group target and the scalar angle random variable based on a linear Gaussian model to obtain a dynamic model;

[0015] Acquire a measurement data set of the current formation group target collected by a preset radar, and describe the measurement data set based on a preset conditional Gaussian mixture model to obtain a measurement model;

[0016] The three-dimensional structure model of the current formation group target is determined according to the three-dimensional structure state, prior distribution form, dynamic model and the measurement model of the current formation group target.

[0017] Optionally, in one or more embodiments of the present specification, the current formation group target is divided into substructures to obtain a preset number of substructures with three-dimensional orientation angles, and a scalar angle random variable of the substructure is determined to process the substructure based on a random matrix to obtain a three-dimensional structural state of the current formation group target, specifically including:

[0018] Dividing the three-dimensional spatial structure of the current formation group target into a preset number of substructures with three-dimensional orientation angles, and describing the three-dimensional spatial position of each substructure according to the three-dimensional spatial position of the preset first substructure;

[0019] Determining the motion state variables of the current formation group targets according to the three-dimensional spatial position of each substructure and the high-order motion state of the current formation group targets;

[0020] Simplifying each of the substructures based on a preset random matrix to obtain an extended matrix corresponding to the substructure, and determining a scalar angle random variable according to the extended matrix;

[0021] The extended matrix is ​​decomposed to obtain a three-dimensional rotation matrix corresponding to the substructure, so as to determine the three-dimensional structural state of the current formation group target based on the three-dimensional rotation matrix.

[0022] Optionally, in one or more embodiments of the present specification, predicting the posterior distribution of each variable in the three-dimensional structure model at a previous moment according to the dynamic model to obtain the predicted distribution specifically includes:

[0023] Classifying the variables corresponding to the three-dimensional structural model to obtain a first variable with a dynamic model and a second variable without a dynamic model;

[0024] According to the standard Kalman prediction method, the first variable is predicted to obtain a first prediction distribution; wherein the first variable includes the motion state and the scalar angle random variable of each of the substructures;

[0025] Predicting the second variables based on the preset forgetting factor to obtain a second prediction distribution; wherein the second variables include the weight random variables of the simplified substructure and the parameters of the expansion matrix;

[0026] The first predicted distribution and the second predicted distribution are used as predicted distributions obtained after the dynamic model predicts the prior distribution at the previous moment.

[0027] Optionally, in one or more embodiments of the present specification, multi-layer linearization processing is performed on the observation equation based on a linearization processing method of a multi-level Taylor expansion to obtain a parameter update formula of the three-dimensional structure model, specifically including:

[0028] Performing a first-order Taylor expansion of the observation equation to realize the first-level linearization of the observation equation, obtaining a first-level linearization result, processing the first-level linearization result based on the mean field theory, and obtaining a closed-form solution for parameter update of the azimuth angle in the scalar angle random variable by taking the average of the pitch angle in the scalar angle random variable;

[0029] Performing a second-level Taylor expansion on the observation equation to realize the second-level linearization of the observation equation, obtaining a second-level linearization result, processing the second-level linearization result based on mean field theory, and obtaining a closed-form solution for parameter update of the pitch angle in the scalar angle random variable;

[0030] According to the closed-form solution for updating the parameters of the azimuth angle and the closed-form solution for updating the parameters of the elevation angle, the parameters of the three-dimensional structure model are updated in sequence to obtain a parameter updating formula for the three-dimensional structure model.

[0031] Optionally, in one or more embodiments of the present specification, a first-order Taylor expansion is performed on the observation equation to implement the first-level linearization of the observation equation, and a first-level linearization result is obtained, and the first-level linearization result is processed based on the mean field theory, and the parameter update closed-form solution of the azimuth angle in the scalar angle random variable is obtained by taking the average of the pitch angle in the scalar angle random variable, specifically including:

[0032] Find the observation equation with respect to The first-order Taylor expansion of the unary variable realizes the first-level linearization of the observation equation, and the first-level linearization result is obtained as follows:

[0033] ;in, Represents rotation along the z-axis in three-dimensional space The rotation matrix of the angle, symbol Indicates the transpose of the matrix. , Represents rotation along the y-axis in three-dimensional space The rotation matrix of the angle, is a single measurement acquired by the sensor, Indicates The observation matrix, represents the three-dimensional group target motion state vector, Represents a function Ask about The partial derivative of , and let the result , is the average estimated value of the azimuth angle of the last iteration, , , The The pitch angle of each substructure is a random variable; based on the mean field theory, the angle Taking the mean value, we can complete the Gaussian azimuth angle random variable The closed-form solution for parameter update is as follows:

[0034] ;

[0035] ;

[0036] in, and Respectively The predicted mean and predicted variance of the Gaussian azimuth angle random variable of the substructure, and is the Gaussian mean and variance after the i+1th iteration, and for:

[0037] , ;

[0038] in, Represents a random variable Seek hope, ; Represents the matrix Take the derivative of the elements in ; Indicates the i-th iteration The estimated results of the mean of the random angles of the substructures; Indicates that it belongs to The number of measurements of each substructure, specifically:

[0039] ,

[0040] ,in, Indicates The weight variables of the substructures, is the number of substructures, Indicates The expanded axis length matrix of the substructure is a diagonal matrix, and s is a scalar value of 1 / 4.

[0041] Optionally, in one or more embodiments of the present specification, the observation equation is subjected to a second-level Taylor expansion to implement a second-level linearization of the observation equation, and a second-level linearization result is obtained, and the second-level linearization result is processed based on the mean field theory to obtain a closed-form solution for parameter update of the pitch angle in the scalar angle random variable, specifically including:

[0042] right In Perform the second-level Taylor expansion to realize the second-level linearization of the observation equation, and obtain the second-level linearization result as follows:

[0043] ;

[0044] in, To find the function about The derivative of , and let the result is the average estimated value of the pitch angle in the last iteration, and is defined as follows:

[0045] ;

[0046] ;

[0047] Based on mean field theory, complete the Gaussian pitch angle random variable The closed-form solution for parameter update is as follows:

[0048] ;

[0049] ;

[0050] in, and Respectively The predicted mean and predicted variance of the Gaussian pitch angle random variable of the substructure, and is the Gaussian mean and variance after the i+1th iteration, and The calculation method is as follows:

[0051] ;

[0052] ;

[0053] in, Indicates the i-th iteration The mean estimation result of the random pitch angle of each substructure is: Represents the matrix Take the element-wise derivative of .

[0054] Optionally, in one or more embodiments of the present specification, according to the closed-form solution for updating the parameters of the azimuth angle and the closed-form solution for updating the parameters of the elevation angle, the parameters of the three-dimensional structure model are updated in sequence to obtain a parameter updating formula of the three-dimensional structure model, which specifically includes:

[0055] Through the mean field theory, the parameter update formulas of the substructure motion state, substructure size and weight in the three-dimensional structure model are given in turn:

[0056] ;

[0057] ;

[0058] ;

[0059] in, Indicates the number of three-dimensional measurements of the formation group targets obtained by the sensor, Indicates The predicted values ​​of the distribution parameters of the substructure weight random variables, and are the predicted mean vector and the inverse of the predicted covariance matrix of the motion state random variables of the formation group targets, and the symbols are and is defined as follows:

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] in, The update formula for the distribution parameter of the weight variable at the i+1th iteration; and The update formula for the i+1 iteration of the distribution parameters representing the motion state variables; and The update formula for the i+1 iterations of the distribution parameter representing the substructure expansion size; and are the predicted values ​​of the two distribution parameters of the substructure size.

[0067] One or more embodiments of the present specification provide a three-dimensional spatial configuration tracking device for a formation group target, the device comprising:

[0068] at least one processor; and,

[0069] a memory communicatively connected to the at least one processor; wherein,

[0070] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform any of the above methods.

[0071] One or more embodiments of the present specification provide a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to execute any of the above-described methods.

[0072] At least one of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects:

[0073] By using a substructure with a three-dimensional orientation angle to model the three-dimensional spatial structure of the formation group targets, the layout and form of the formation group targets in three-dimensional space can be accurately portrayed, reflecting the complex structural characteristics of the formation group targets. Compared with simple two-dimensional or low-dimensional modeling methods, it is more in line with the actual scene. According to the dynamic model, the prior distribution at the current moment, that is, the posterior distribution at the previous moment, is predicted to obtain the predicted distribution. This enables the model to estimate the state distribution at the next moment according to the dynamic change law of the target, taking into account the movement trend and possible state changes of the target, and enhancing the foresight and adaptability of the target state estimation. By determining the observation equation and performing multi-layer linearization to obtain the parameter update formula, the current variational Bayesian technology for two-dimensional scenes solves the problem that the model cannot be directly extended to three dimensions by simply expanding the dimension, resulting in the inability to estimate the three-dimensional orientation angle, and realizes the precise positioning and tracking of the three-dimensional spatial configuration. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0075] Figure 1 A schematic flow chart of a three-dimensional spatial configuration tracking method for a formation group target provided in an embodiment of this specification;

[0076] Figure 2 A top view of a tracking result of a formation group target provided in an embodiment of this specification;

[0077] Figure 3 A schematic diagram of a three-dimensional spatial configuration tracking and estimation result of a formation group target provided in an embodiment of this specification;

[0078] Figure 4 A schematic diagram of the comparison results of tracking estimation between a traditional tracking method provided in an embodiment of this specification and the method of this specification;

[0079] Figure 5A schematic diagram of a three-dimensional spatial configuration tracking device for a formation group target provided in an embodiment of this specification;

[0080] Figure 6 A schematic diagram of the structure of a non-volatile storage medium provided in an embodiment of this specification. DETAILED DESCRIPTION

[0081] The embodiments of this specification provide a method, device and medium for tracking the three-dimensional spatial configuration of a formation group of targets.

[0082] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0083] like Figure 1 As shown, the present specification embodiment provides a method flow chart of a three-dimensional spatial configuration tracking method for a formation group target. Figure 1 It can be seen that in one or more embodiments of this specification, a three-dimensional spatial configuration tracking method of a formation group target includes the following steps:

[0084] S101: Modeling the three-dimensional spatial structure of the current formation group target according to a preset number of substructures with three-dimensional orientation angles to obtain a three-dimensional structural model of the current formation group target; wherein the three-dimensional structural model includes: the three-dimensional structural state, prior distribution, dynamic model and measurement model of the current formation group target.

[0085] In the embodiments of the present specification, a new three-dimensional spatial structure model of a complex formation is first constructed, that is, the three-dimensional spatial structure of the current formation group target is modeled according to a preset number of substructures with three-dimensional orientation angles to obtain a three-dimensional structure model of the current formation group target; it should be noted that the three-dimensional structure model includes: the three-dimensional structure state, prior distribution, dynamic model and measurement model of the current formation group target.

[0086] Specifically, in one or more embodiments of the present specification, the three-dimensional spatial structure of the current formation group target is modeled according to a preset number of substructures with three-dimensional orientation angles to obtain a three-dimensional structural model of the current formation group target, which specifically includes the following process:

[0087] First, the current formation group target is divided into substructures to obtain a preset number of substructures with three-dimensional orientation angles, and then the scalar angle random variables of the substructures are determined to process the substructures based on a random matrix to obtain the three-dimensional structural state of the current formation group target. Further, the current formation group target is divided into substructures to obtain a preset number of substructures with three-dimensional orientation angles, and the scalar angle random variables of the substructures are determined to process the substructures based on a random matrix to obtain the three-dimensional structural state of the current formation group target, specifically including:

[0088] The three-dimensional spatial structure of the current formation group target is divided into a preset number of substructures with three-dimensional orientation angles, and the three-dimensional spatial position of each substructure is described according to the three-dimensional spatial position of the preset first substructure. For example, the complex spatial structure of the formation group target is divided into L substructures, and it is assumed that the three-dimensional spatial position of the first substructure, that is, the preset first substructure, is , are the coordinates of the center of the first substructure on the x-axis, y-axis, and z-axis in the rectangular coordinate system, then you can use the variable Describes the spatial position of the remaining substructures relative to the first substructure, where Indicates The displacement vector of the center coordinates of the substructure relative to the center coordinates of the first substructure. Since they belong to the same group target, each substructure shares high-order motion states such as velocity and acceleration. Therefore, according to the three-dimensional spatial position of each substructure and the high-order motion state of the current formation group target, the motion state variable of the current formation group target is determined. for:

[0089] ;

[0090] Then, each substructure is simplified according to the preset random matrix to obtain the expansion matrix corresponding to the substructure, and the scalar angle random variable is determined according to the expansion matrix. In a certain scenario, a new ellipsoid expansion representation model is proposed based on the random matrix theory. At this time, the shape of each substructure is simplified to an ellipsoid. Taking the lth substructure as an example, the size of its ellipsoidal expansion is determined by a diagonally positive random matrix Description, where the elements , represents the square of the semi-major axis of the ellipsoid. In addition, two scalar angle random variables and It is used to describe the azimuth and elevation of the ellipsoid in three-dimensional space, which together determine the orientation of the ellipsoid. is defined as the angle between the major axis of the ellipsoid and the x-axis, Defined as the angle between the ellipsoid and the xoy plane.

[0091] Then the expansion matrix is ​​decomposed to obtain the three-dimensional rotation matrix corresponding to the substructure, so as to determine the three-dimensional structural state of the current formation group target based on the three-dimensional rotation matrix, that is, in the random matrix framework, the expansion matrix of the lth substructure is Can be decomposed into:

[0092] ;

[0093] in and Represents the three-dimensional rotation matrix along the z-axis and y-axis respectively, and the specific form is

[0094] ;

[0095] Furthermore, the weight of the ellipsoidal substructure is defined as ,in For the The weight random variable of each substructure.

[0096] Obtaining the prior distribution form of each variable in the process of determining the three-dimensional structure state: wherein the prior distribution form of the above variables is as follows:

[0097] ;

[0098] in, represents the spatial dimension, is the initial value of the target state vector set of the formation group, represents a Gaussian distribution, represents the inverse gamma distribution, represents the gamma distribution, represents the Dirichlet distribution. For simplicity, we denote the model parameters of the above joint distribution as .

[0099] The motion state of the current formation group target and the scalar angle random variable are evolved based on the linear Gaussian model to obtain a dynamic model, that is, for the motion state 、Azimuth angle random vector , random vector of pitch angle For , its evolution model is linear Gaussian:

[0100] ;

[0101] ;

[0102] ;

[0103] in, is the motion state transfer matrix, The covariance is Gaussian white process noise. The covariance is Gaussian white process noise, The covariance is Gaussian white process noise. is the unit matrix of L rows and L columns.

[0104] for and , due to the lack of necessary prior information, it is assumed that the mean of the above variable distribution is approximately constant in adjacent time intervals:

[0105] .

[0106] The measurement data set of the current formation group target collected by the preset radar is obtained, and the measurement data set is described based on the preset conditional Gaussian mixture model to obtain the measurement model, that is, the following conditional Gaussian mixture model is used to describe the complex three-dimensional spatial measurement distribution of the formation group target:

[0107] ;

[0108] in, It is a set of group target measurements collected by the radar containing three-dimensional position information. Structural variables of formation group targets And group motion state are integrated into a unified measurement model. It is a 3*3 matrix, which represents the measurement error of the radar sensor to the three-dimensional space position. Indicates The observation matrix, The calculated result is Then, according to the three-dimensional structural state, prior distribution form, dynamic model and measurement model of the current formation group target, the three-dimensional structural model of the current formation group target can be determined.

[0109] S102: Predicting the posterior distribution of each variable in the three-dimensional structural model at a previous moment according to the dynamic model to obtain a predicted distribution.

[0110] Based on the above step S101, the three-dimensional spatial structure modeling of the complex formation group targets is realized, that is, the complex motion state of the formation group targets is precisely described by defining multiple ellipsoid motion state variables, and the axis length of each ellipsoid is modeled as a diagonally positive random matrix, and the ellipsoid orientation is decomposed into two independent random variables, azimuth angle and pitch angle, and a reasonable prior distribution is assigned to them; secondly, the group state and structure information are integrated into the modified conditional Gaussian mixture observation model to achieve an effective description of the three-dimensional spatial distribution of the group measurement. In the embodiment of this specification, the posterior distribution of each variable in the three-dimensional structure model at the previous moment is predicted according to the dynamic model to obtain the predicted distribution.

[0111] Specifically, in one or more embodiments of the present specification, the posterior distribution of each variable in the three-dimensional structure model at the previous moment is predicted according to the dynamic model to obtain the predicted distribution, which specifically includes the following process:

[0112] First, the variables corresponding to the three-dimensional structural model are classified to obtain the first variable with a dynamic model and the second variable without a dynamic model. Then, the first variable is predicted according to the standard Kalman prediction method to obtain the first prediction distribution; it should be noted that the first variable includes the motion state and the scalar angle random variables of each substructure. In a certain application scenario, the following standard Kalman prediction equation is used to predict the motion state and the orientation of each substructure:

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] ;

[0118] ;

[0119] in," ” represents the estimated value of each distribution parameter at time k-1.

[0120] In addition, the second variable is predicted based on the preset forgetting factor to obtain the second prediction distribution. The second variable includes the weight random variable of the simplified substructure and the parameters of the expansion matrix, that is, the forgetting factor is used in a certain application scenario. and For the random matrix of the expanded size and weight random variables One-step prediction of the parameters in the prior distribution of :

[0121] ;

[0122] ;

[0123] ;

[0124] After using the first predicted distribution and the second predicted distribution as a dynamic model to predict the prior distribution of the previous moment, the predicted distribution is obtained. ,in is the set of measurements accumulated to time k-1.

[0125] S103: Determine the observation equation corresponding to the current formation group target according to the relationship between the three-dimensional structure state and the measurement model, perform multi-layer linearization processing on the observation equation in a linearization processing method based on multi-level Taylor expansion, and obtain the parameter update formula of the three-dimensional structure model.

[0126] Based on the above steps S101-S102, it can be known that due to the introduction of two angle variables in the model building process, the observation equation is nonlinear relative to the formation configuration state, resulting in the inability of the traditional Bayesian formula to directly estimate the model parameter update. The existing multi-ellipsoid method, although it can increase the extension matrix dimension, such as changing the matrix from a 2*2 matrix to a 3*3 matrix, realizes modeling of a three-dimensional ellipsoid, but the essence of the model is to model the three-dimensional orientation angle of the ellipsoid with the axis length coupling. However, in such a modeling method, it is believed that the rotation of the substructure in three-dimensional space is due to the change of its extended form, which also causes it to process the scene of the substructure rotation maneuver, and the processing effect is not good. In addition, although the current method based on the variational Bayesian technique approximate solution can model the overall orientation and substructure orientation of the formation group separately, these models are all proposed for two-dimensional scenes, and the model cannot be directly extended to three dimensions by simply expanding the dimension, and the three-dimensional orientation angle of the substructure cannot be estimated. In addition, the three-dimensional orientation angle is split into two independent scalar angles at the modeling level, which can accurately describe the three-dimensional orientation. Since the rotation equations corresponding to the two orientation angles are different, the model has a high degree of nonlinearity, and a simple linearization method cannot be used directly. Therefore, in the embodiment of this specification, after the observation equation corresponding to the current formation group target is determined according to the relationship between the three-dimensional structural state and the measurement model, the observation equation will be subjected to multi-layer linearization processing through a multi-level Taylor expansion linearization processing method to obtain the parameter update formula of the three-dimensional structural model, and realize the closed-form solution of the three-dimensional orientation angle.

[0127] Specifically, in one or more embodiments of the present specification, a multi-level linearization processing method based on a multi-level Taylor expansion is used to perform multi-level linearization processing on the observation equation to obtain a parameter update formula for the three-dimensional structure model, which specifically includes the following process:

[0128] First, the observation equation is expanded by a first-order Taylor expansion to realize the first-level linearization of the observation equation, and the first-level linearization result is obtained. The first-level linearization result is processed based on the mean field theory, and the closed-form solution of the parameter update of the azimuth angle in the scalar angle random variable is obtained by taking the mean of the pitch angle in the scalar angle random variable. Specifically, the following definition is first made:

[0129] ;

[0130] Since the binary Taylor expansion will lead to coupling between the two angle variables, it is impossible to derive a closed-form solution. and All of them can realize closed-form analytical solutions. In the embodiments of this specification, the observation equation will be solved for its The first-order Taylor expansion of the unary variable is used to realize the first-level linearization of the observation equation, and the first-level linearization result is obtained as follows:

[0131] ;in, Represents rotation along the z-axis in three-dimensional space The rotation matrix of the angle, the symbol T means to transpose the matrix, , Represents rotation along the y-axis in three-dimensional space The rotation matrix of the angle, is a single measurement acquired by the sensor, Indicates The observation matrix, represents the three-dimensional group target motion state vector, Represents a function Ask about The partial derivative of , and let the result , is the average estimated value of the azimuth angle of the last iteration, The The pitch angle random variable of the substructure is and is defined as follows:

[0132] ,

[0133] ;

[0134] At this point, the first level of linearization is completed, and then based on the mean field theory, the angle Taking the mean value, we can complete the Gaussian azimuth angle random variable The closed-form solution for parameter update is as follows:

[0135] ;

[0136] ;

[0137] in, and Respectively The predicted mean and predicted variance of the Gaussian azimuth angle random variable of the substructure, and is the Gaussian mean and variance after the i+1th iteration, and for:

[0138] , ;

[0139] in, Represents a random variable Seek hope, . Represents the matrix Take the derivative of the elements in ; Indicates the i-th iteration The estimated results of the mean of the random angles of the substructures; Indicates that it belongs to The number of measurements of a substructure can be calculated according to the following formula:

[0140] ,

[0141] ,in, Indicates The weight variables of the substructures, is the number of substructures, Indicates The expanded axis length matrix of the substructure is a diagonal matrix, and s is a scalar value of 1 / 4.

[0142] Now, in order to solve The observation equation will be expanded at the second level by Taylor to realize the second level linearization of the observation equation and obtain the second level linearization result. The second level linearization result will be processed based on the mean field theory to obtain the closed-form solution for the parameter update of the pitch angle in the scalar angle random variable. In other words, since the first level linearization processing has removed So now it will have an impact on Perform the same linearization operation, the specific processing process is as follows:

[0143] right Perform the second-level Taylor expansion to realize the second-level linearization of the observation equation, and obtain the second-level linearization result as follows:

[0144] ;

[0145] in,

[0146] ;

[0147] ;

[0148] At this point, the second layer of linearization is completed, and then based on the mean field theory, the Gaussian pitch angle random variable can be completed The closed-form solution for parameter update is as follows:

[0149] ;

[0150] ;

[0151] in, and is the Gaussian mean and variance after the i+1th iteration, and The calculation method is as follows:

[0152] ;

[0153] ;

[0154] in Represents the matrix So far, the linearization of the model has been completed through a two-level Taylor expansion method, and the closed-form solution of the parameter update formula of the three-dimensional orientation angle, the core parameter of the three-dimensional configuration, has been derived.

[0155] Then, the parameters of the three-dimensional structure model can be updated in sequence according to the closed-form solution of the azimuth angle parameter update and the pitch angle parameter update, and the parameter update formula of the three-dimensional structure model can be obtained. That is to say, through the mean field theory, the parameter update formula of the substructure motion state in the three-dimensional structure model, i.e., the substructure position and the formation group target speed, the substructure size, i.e., the extended axis length and the weight are given in sequence:

[0156] ;

[0157] ;

[0158] ;

[0159] in, Indicates the number of three-dimensional measurements of the formation group targets obtained by the sensor, Indicates The predicted values ​​of the distribution parameters of the substructure weight random variables, and are the predicted mean vector and the inverse of the predicted covariance matrix of the motion state random variables of the formation group targets, and the symbols are and is defined as follows:

[0160] ;

[0161] ;

[0162] ;

[0163] ;

[0164] ;

[0165] ;

[0166] in, The update formula for the distribution parameter of the weight variable at the i+1th iteration; and The update formula for the i+1 iteration of the distribution parameters representing the motion state variables; and The update formula for the i+1 iterations of the distribution parameter representing the substructure expansion size; and are the predicted values ​​of the two distribution parameters of the substructure size.

[0167] S104: Based on the parameter updating formula and the predicted distribution, the parameters of the current formation group targets are updated and corrected in an iterative manner to achieve tracking and three-dimensional configuration estimation of the current formation group targets.

[0168] After obtaining the parameter update formula of the three-dimensional structure model according to the above steps S101-S103, the parameters of the variable distribution can be iteratively updated in sequence according to the above update formula, and the optimal estimate of each approximate distribution can be determined based on the mean field theory. Then, the three-dimensional spatial position, three-dimensional orientation, estimated value of the extended axis length and other parameters of the substructure can be calculated based on the approximate distribution, thereby realizing the positioning of the formation group targets and the three-dimensional configuration estimation.

[0169] In order to illustrate and analyze the above method, a scenario based on a single radar detecting a group of targets in a three-dimensional space formation is used as an example. In this example scenario, a radar continuously observes a group of targets in the detection area at a sampling interval of 3s, and collects a total of 25 screens of measurement data. The k-time measurement set is recorded as , measurement noise covariance for The formation configuration of the group target is V-shaped, and the formation as a whole flies at a constant speed along the entire direction of the x-axis. During the movement, the angle between the two sides of the V-shaped formation gradually decreases to 120°, and the changed configuration is maintained until the end of the simulation. The present invention is applied to the group target robust tracking experiment, in which the algorithm parameters are set as follows: the number of substructures is set to 2, the covariance of the motion state process noise is set to Set to , average the process noise variance of the azimuth and elevation random angles is 0.01, each forgetting factor is set to 0.9, and the number of variational iterations is set to 60. The top view of the group target tracking result is shown in Figure 2 As shown in FIG. 1 , it can be seen that the method proposed in the present invention can stably track the dynamic orientation angle of the internal structure of the formation group target. The three-dimensional tracking result of the formation group target is shown in FIG. Figure 3 As shown in FIG. 1 , it can be seen that the present invention can accurately track and estimate the three-dimensional spatial configuration of a group target. In addition, the tracking results of the traditional multi-ellipsoid method after three-dimensional expansion are also shown as follows: Figure 4 As shown, through Figure 4 It can be seen that the traditional tracking method has poor tracking effect for the formation structure, especially for the scene with dynamic changes in the three-dimensional orientation angle.

[0170] like Figure 5 In the embodiment of this specification, a schematic diagram of the structure of a three-dimensional spatial configuration tracking device for a formation group target is provided. Figure 5 It can be seen that a three-dimensional spatial configuration tracking device for a formation group target includes:

[0171] at least one processor; and,

[0172] a memory communicatively connected to the at least one processor; wherein,

[0173] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any of the above-mentioned methods.

[0174] like Figure 6 As shown, the embodiment of this specification provides a non-volatile storage medium storing computer executable instructions. Figure 6 It can be seen that in one or more embodiments of the present specification, a non-volatile storage medium stores computer-executable instructions, and the computer-executable instructions can execute any of the methods described above.

[0175] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0176] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0177] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A three-dimensional spatial configuration tracking method for a formation group target, characterized in that: The method comprises: Modeling the three-dimensional spatial structure of the current formation group target according to a preset number of substructures with three-dimensional orientation angles to obtain a three-dimensional structural model of the current formation group target; wherein the three-dimensional structural model includes: the three-dimensional structural state, prior distribution, dynamic model and measurement model of the current formation group target; Predicting the posterior distribution of each variable in the three-dimensional structural model at a previous moment according to the dynamic model to obtain a predicted distribution; Determine the observation equation corresponding to the current formation group target according to the relationship between the three-dimensional structure state and the measurement model, perform multi-layer linearization processing on the observation equation in a linearization processing method based on multi-level Taylor expansion, and obtain the parameter update formula of the three-dimensional structure model; Based on the parameter update formula and the predicted distribution, the parameters of the current formation group target are updated and corrected in an iterative manner to achieve tracking and three-dimensional configuration estimation of the current formation group target; The observation equation is subjected to multi-layer linearization processing based on a multi-layer Taylor expansion linearization processing method to obtain a parameter update formula for the three-dimensional structural model, which specifically includes: Performing a first-order Taylor expansion of the observation equation to realize the first-level linearization of the observation equation, obtaining a first-level linearization result, processing the first-level linearization result based on the mean field theory, and obtaining a closed-form solution for parameter update of the azimuth angle in the scalar angle random variable by taking the average of the pitch angle in the scalar angle random variable; Performing a second-level Taylor expansion on the observation equation to realize the second-level linearization of the observation equation, obtaining a second-level linearization result, processing the second-level linearization result based on mean field theory, and obtaining a closed-form solution for parameter update of the pitch angle in the scalar angle random variable; According to the closed-form solution for updating the parameters of the azimuth angle and the closed-form solution for updating the parameters of the elevation angle, the parameters of the three-dimensional structure model are updated in sequence to obtain a parameter updating formula for the three-dimensional structure model.

2. A three-dimensional spatial configuration tracking method for a formation group target according to claim 1, characterized in that: Modeling the three-dimensional spatial structure of the current formation group target according to a preset number of substructures with three-dimensional orientation angles to obtain the three-dimensional structural model of the current formation group target specifically includes: Dividing the current formation group target into substructures to obtain a preset number of substructures with three-dimensional orientation angles, determining scalar angle random variables of the substructures to process the substructures based on a random matrix to obtain a three-dimensional structural state of the current formation group target; Obtaining the prior distribution form of each variable in the process of determining the three-dimensional structure state; Evolving the motion state of the current formation group target and the scalar angle random variable based on a linear Gaussian model to obtain a dynamic model; Acquire a measurement data set of the current formation group target collected by a preset radar, and describe the measurement data set based on a preset conditional Gaussian mixture model to obtain a measurement model; The three-dimensional structure model of the current formation group target is determined according to the three-dimensional structure state, prior distribution form, dynamic model and the measurement model of the current formation group target.

3. A three-dimensional spatial configuration tracking method for a formation group target according to claim 2, characterized in that: The current formation group target is divided into substructures to obtain a preset number of substructures with three-dimensional orientation angles, and a scalar angle random variable of the substructure is determined to process the substructure based on a random matrix to obtain a three-dimensional structural state of the current formation group target, specifically including: Dividing the three-dimensional spatial structure of the current formation group target into a preset number of substructures with three-dimensional orientation angles, and describing the three-dimensional spatial position of each substructure according to the three-dimensional spatial position of the preset first substructure; Determining the motion state variables of the current formation group targets according to the three-dimensional spatial position of each substructure and the high-order motion state of the current formation group targets; Simplifying each of the substructures based on a preset random matrix to obtain an extended matrix corresponding to the substructure, and determining a scalar angle random variable according to the extended matrix; The extended matrix is ​​decomposed to obtain a three-dimensional rotation matrix corresponding to the substructure, so as to determine the three-dimensional structural state of the current formation group target based on the three-dimensional rotation matrix.

4. A three-dimensional spatial configuration tracking method for a formation group target according to claim 3, characterized in that: Predicting the posterior distribution of each variable in the three-dimensional structural model at the last moment according to the dynamic model to obtain the predicted distribution specifically includes: Classifying the variables corresponding to the three-dimensional structural model to obtain a first variable with a dynamic model and a second variable without a dynamic model; According to the standard Kalman prediction method, the first variable is predicted to obtain a first prediction distribution; wherein the first variable includes the motion state and the scalar angle random variable of each of the substructures; Predicting the second variables based on the preset forgetting factor to obtain a second prediction distribution; wherein the second variables include the weight random variables of the simplified substructure and the parameters of the expansion matrix; The first predicted distribution and the second predicted distribution are used as predicted distributions obtained after the dynamic model predicts the posterior distribution at the previous moment.

5. The three-dimensional spatial configuration tracking method of a formation group target according to claim 1, characterized in that: The observation equation is subjected to a first-order Taylor expansion to realize the first-level linearization of the observation equation, and the first-level linearization result is obtained. The first-level linearization result is processed based on the mean field theory, and the parameter update closed-form solution of the azimuth angle in the scalar angle random variable is obtained by taking the average of the pitch angle in the scalar angle random variable, which specifically includes: Find the observation equation with respect to The first-order Taylor expansion of the unary variable realizes the first-level linearization of the observation equation, and the first-level linearization result is obtained as follows: ;in, Represents rotation along the z-axis in three-dimensional space The rotation matrix of the angle, symbol It means to transpose the matrix. , Represents rotation along the y-axis in three-dimensional space The rotation matrix of the angle, is a single measurement acquired by the sensor, Indicates The observation matrix, represents the three-dimensional group target motion state vector, Represents the function Ask about The partial derivative of , and let the result , is the average estimated value of the azimuth angle of the last iteration, , , The The pitch angle random variable of each substructure; Based on mean field theory, the angle Taking the mean value, we can complete the Gaussian azimuth angle random variable The closed-form solution for parameter update is as follows: ； ； in, and Respectively The predicted mean and predicted variance of the Gaussian azimuth angle random variable of the substructure, and is the Gaussian mean and variance after the i+1th iteration, and for: ， ； in, Represents a random variable Seek hope, ; Represents the matrix Take the derivative of the elements in ; Indicates the i-th iteration The estimated results of the mean of the random angles of the substructures; Indicates that it belongs to The number of measurements of each substructure, specifically: ， ,in, Indicates The weight variables of the substructures, is the number of substructures, Indicates The expanded axis length matrix of the substructure is a diagonal matrix, and s is a scalar value of 1 / 4.

6. A three-dimensional spatial configuration tracking method for a formation group target according to claim 5, characterized in that: Performing a second-level Taylor expansion on the observation equation to realize the second-level linearization of the observation equation, obtaining a second-level linearization result, and processing the second-level linearization result based on the mean field theory to obtain a closed-form solution for parameter update of the pitch angle in the scalar angle random variable, specifically including: right In Perform the second-level Taylor expansion to realize the second-level linearization of the observation equation, and obtain the second-level linearization result as follows: ； in, To find the function about The derivative of , and let the result is the average estimated value of the pitch angle in the last iteration, and is defined as follows: ； ； Based on mean field theory, complete the Gaussian pitch angle random variable The closed-form solution for parameter update is as follows: ； ； in, and Respectively The predicted mean and predicted variance of the Gaussian pitch angle random variable of the substructure, and is the Gaussian mean and variance after the i+1th iteration, and The calculation method is as follows: ； ； in, Indicates the i-th iteration The mean estimation result of the random pitch angle of each substructure is: Represents the matrix Take the derivative of the elements in .

7. The three-dimensional spatial configuration tracking method of a formation group target according to claim 1, characterized in that: According to the closed-form solution for updating the parameters of the azimuth angle and the closed-form solution for updating the parameters of the elevation angle, the parameters of the three-dimensional structure model are updated in sequence to obtain a parameter updating formula for the three-dimensional structure model, which specifically includes: Through the mean field theory, the parameter update formulas of the substructure motion state, substructure size and weight in the three-dimensional structure model are given in turn: ； ； ； in, Indicates the number of three-dimensional measurements of the formation group targets obtained by the sensor, Indicates The predicted values ​​of the distribution parameters of the substructure weight random variables, and are the predicted mean vector and the inverse of the predicted covariance matrix of the motion state random variables of the formation group targets, and the symbols are and is defined as follows: ； ； ； ； ； ； in, The update formula for the distribution parameter of the weight variable at the i+1th iteration; and The update formula for the i+1 iteration of the distribution parameters representing the motion state variables; and The update formula for the i+1 iterations of the distribution parameter representing the substructure expansion size; and are the predicted values ​​of the two distribution parameters of the substructure size.

8. A three-dimensional spatial configuration tracking device for a formation group target, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: execute any of the methods described in claims 1-7.

9. A non-volatile storage medium storing computer executable instructions, characterized in that: The computer executable instructions can: execute the method described in any one of claims 1 to 7.

Images