Search

CN-119147038-B - Maneuvering group target intelligent tracking method based on deep neural network

CN119147038BCN 119147038 BCN119147038 BCN 119147038BCN-119147038-B

Abstract

The invention discloses an intelligent tracking method of maneuvering group targets based on a deep neural network, which introduces deep learning into a Bayesian filtering framework, extracts group target motion characteristics and process noise statistical characteristics from measurement data based on a plurality of deep neural networks respectively, further estimates a group target motion state transition matrix and a process noise variance matrix on line, then estimates the motion state of mass centers of the group targets with high precision through Bayesian filtering, and models the group target contour as an ellipse with the center at the mass centers of the group targets, thereby improving the estimation precision of the group target contour by utilizing the high-precision mass center motion state estimation result of the group targets, and finally completing the tracking of the group targets. Compared with the existing group target tracking method, the method provided by the invention does not need prior information to pre-establish a target motion model, and can implement high-precision tracking on the non-cooperative maneuvering group target lacking prior information in a complex environment.

Inventors

  • LI YINYA
  • LIANG YUAN
  • CHEN YE
  • QI GUOQING
  • Sheng Andong

Assignees

  • 南京理工大学

Dates

Publication Date
20260505
Application Date
20240830

Claims (8)

  1. 1. The intelligent maneuvering group target tracking method based on the deep neural network is characterized by comprising the following steps of: The first step, obtaining group target measurement information at the current moment, wherein the method specifically comprises the following steps: 1) For group targets with densely distributed units in the group, the detection device can acquire a plurality of measurement data generated by the units in the group at each sampling moment, namely, a measurement data set is formed Wherein For the number of measurement data acquired at time k, the measurement data Generated by the first measuring source appearing at the moment k, the measuring equation is shown as follows ; Where h (. Cndot.) represents the measurement function, Representing the state vector of the first metrology source, Representing measurement noise; 2) Calculating mass center measurement information of the k-moment group target according to the measurement data set ; Secondly, estimating the mass center motion state of the group target at the current moment; the method comprises the following steps: based on the volume Kalman filtering method, the mass center motion state of the group target is estimated, and the method comprises the following specific steps of 1) Calculating a measured volume point ; In the formula, , Motion state vector for mass center of group target Dimension of (2); A square root factor of prior error covariance of mass center motion state of the group target; Is a collection The j-th element of (3) The definition is as follows ; 2) Computing group target centroid prior measurements ; 3) Computing a square root factor for a priori innovation covariance ; In the formula, R k is the covariance of the measured noise; Representing a full rank matrix Performing orthogonal decomposition to obtain a transposed matrix of the upper triangular matrix; 4) Calculating cross covariance ; 5) Using group target centroid measurements Estimating centroid motion state ; Thirdly, estimating the outline shape of the group target at the current moment; Fourth, predicting mass center motion state of the group target at the next moment; and fifthly, predicting the group target contour form at the next moment.
  2. 2. The intelligent tracking method for maneuvering group targets based on the deep neural network according to claim 1, wherein the third step is specifically as follows: in a two-dimensional scene, the group target outline morphology is simplified and modeled into a time-varying ellipse, which is shown in the following formula ; Wherein y is the coordinate vector of any point on the elliptic outline of the group target, E k is the symmetric positive definite random matrix, and the transfer probability distribution function is shown as follows ; In the formula, Is a degree of freedom; W (Y; a, C) represents the Wishare distribution of the symmetric positive definite random matrix C satisfying the degree of freedom a and the parameter matrix Y; the estimation of the target contour form of the current time group is completed through Bayes filtering, and the method is shown as the following formula ; Wherein, the A distribution covariance matrix is measured for the cluster centroid, For the intermediate calculation of the parameter matrix, Modeling an error matrix for the extended target contour morphology, Wherein D is the estimation result of the group target outline parameter matrix Is used for the number of dimensions of (c), Is a degree of freedom parameter in the inverse Wishart distribution; if the tracking task for the group target is finished, the group target tracking algorithm is exited, and otherwise, the fourth step is continuously executed.
  3. 3. The intelligent tracking method for maneuvering group targets based on the deep neural network according to claim 2, wherein the fourth step is specifically: predicting the mass center motion state of the group target at the moment k+1 by using the mass center motion state estimation result of the group target at the moment k, wherein the method comprises the following specific steps: 1) Estimating state transition matrix based on deep neural network The method comprises the steps of inputting mass center measurement data of a group target from k-w+1 time to current k time in a time sequence in a dynamic sliding window mode, wherein w is the length of the sliding window, extracting mass center motion characteristics of the group target from the mass center measurement data by using a deep neural network, and outputting key parameters in a state transition matrix; 2) Estimating process noise variance based on deep neural network The method comprises the steps of forming a time sequence by centroid measurement data from k-w+1 time to k time in a dynamic sliding window mode, converting the centroid measurement data into a coordinate system in which a centroid motion state is located, performing dispersion standardization, extracting features in the data through convolution filtering, judging whether a group target motion mode is changed or not through second-order difference, extracting group target motion characteristics from the time sequence through a multi-layer Bi-LSTM network, outputting the probability that the group target centroid motion mode is matched with a constant-speed motion and constant-turning rate model by utilizing a multi-layer perceptron and a normalized exponential function, selecting a structural form of a process noise variance matrix according to the maximum matching probability, extracting process noise characteristics from the time sequence through a multi-layer Bi-LSTM network, outputting a process noise intensity estimation result through a full-connection layer, and finally combining the results of the two aspects to obtain an estimation result of the process noise variance matrix; 3) Predicting mass center motion state of group target State transition matrix estimation result obtained by deep neural network And process noise variance matrix estimation results Based on a volume Kalman filtering method, predicting the mass center motion state of the group target at the moment k+1, and specifically comprises the following steps: (3.1) calculating the State volume Point ; (3.2) Predicting group target centroid motion states 。
  4. 4. The intelligent tracking method for maneuvering group targets based on the deep neural network according to claim 3, wherein the fifth step is specifically: Predicting the group target contour form at the time of k+1 by using the group target contour form estimation result at the time of k, as shown in the following formula ; In the formula, For oval outline parameter matrix Parameters for measuring the degree of influence of the noise variance; after the steps are finished, returning to the first step, waiting for the start of the next sampling moment, and then continuing to finish the tracking of the group targets according to the steps.
  5. 5. A deep neural network-based motor group target intelligent tracking system for implementing the method of any one of claims 1-4, the system comprising: the first module is used for acquiring group target measurement information at the current moment; the second module is used for estimating the mass center motion state of the group target at the current moment; a third module, configured to estimate a group target contour shape at a current time; A fourth module, configured to predict a mass center motion state of the group target at a next moment; and a fifth module for predicting the group target contour form at the next moment.
  6. 6. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method of any of claims 1-4 when the program is executed.
  7. 7. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 1-4.
  8. 8. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any of claims 1-4.

Description

Maneuvering group target intelligent tracking method based on deep neural network Technical Field The invention relates to the field of target tracking, in particular to a maneuvering group target intelligent tracking method based on a deep neural network. Background In recent years, with the advanced development of computer, communication, navigation, etc., a clustered weapon system such as a swarm unmanned plane, a multi-shot director series, etc. has been gradually applied. On the one hand, the cluster weapon system comprises a large number of fight units, so that the detection, tracking and interception capabilities of the air defense system are saturated rapidly, and effective interception of the air defense system is difficult to realize, and on the other hand, each fight unit in the cluster weapon system has high synergism, and after a small number of fight units are damaged, the rest units can be quickly repaired, so that the overall fight capability of the cluster is not reduced obviously, and effective damage to the cluster is difficult to realize. Thus, the difficulty of defending against the clustered weapon systems described above has become a major challenge that modern air defense systems must address. The group targets are precisely tracked, so that the change process and trend of the movement states of the group targets and the outline forms of the group targets along with time are obtained, and the method is a necessary premise for a subsequent weapon system to formulate a defense strategy, solve attack elements and implement effective interception. At present, aiming at group targets with a large number of targets and dense distribution in a group, in order to improve the real-time performance of a tracking system and avoid the saturation of tracking capacity, a group integral tracking method is mainly adopted, namely, the motion state of a group centroid and the shape of a group outline are estimated in sequence. In the aspect of group centroid motion state estimation, the existing method and technology need to build a motion model of a tracked target in advance, and belong to a model-driven target tracking method. However, when a non-cooperative target is tracked in a complex battlefield environment, due to factors such as lack of prior information, variable target motion modes, unknown process noise statistics characteristics, time variation and the like, the pre-established motion model is possibly mismatched, so that the target tracking precision is obviously reduced, and the tracking performance is seriously degraded. Therefore, existing model-driven group target tracking methods and techniques cannot accurately estimate the state of the non-cooperative maneuver group target in a complex environment. Disclosure of Invention The invention aims to provide an intelligent maneuvering group target tracking method based on a deep neural network, which solves the problem of maneuvering group target high-precision tracking in complex environments such as changeable group target motion modes, lack of target priori information, time-varying process noise statistics characteristics and the like. The technical scheme for achieving the purpose of the invention is that in a first aspect, the invention provides a maneuvering group target intelligent tracking method based on a deep neural network, which comprises the following steps: the method comprises the steps of firstly, obtaining group target measurement information at the current moment; Secondly, estimating the mass center motion state of the group target at the current moment; thirdly, estimating the outline shape of the group target at the current moment; Fourth, predicting mass center motion state of the group target at the next moment; and fifthly, predicting the group target contour form at the next moment. In a second aspect, the present invention provides a motor group target intelligent tracking system based on a deep neural network, for implementing the method of the first aspect, where the system includes: the first module is used for acquiring group target measurement information at the current moment; the second module is used for estimating the mass center motion state of the group target at the current moment; a third module, configured to estimate a group target contour shape at a current time; A fourth module, configured to predict a mass center motion state of the group target at a next moment; and a fifth module for predicting the group target contour form at the next moment. In a third aspect, the present invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method of the first aspect when the program is executed. In a fourth aspect, the present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the s