Flight target joint tracking and identification method based on position and flight envelope information

Abstract

The invention discloses a flight target joint tracking and identification method based on position and flight envelope information, and the method comprises the steps: building a shareable and parsable motion model set based on the motion state of a flight target; receiving measurement data at an initial moment, and initializing flight target track and motion feature information; performing filtering estimation on the motion state of the flight target by using an optimal model expansion algorithm, and extracting motion state features of the flight target; calculating an acceleration-velocity flight envelope likelihood function; identifying a flight target by using a sequential hypothesis testing method; and judging a recognition result, and repeatedly carrying out flight target motion state filtering estimation, feature extraction, likelihood function calculation and flight target recognition until flight target track estimation, recognition and classification are completed. According to the invention, high-precision and stable tracking of the flight target based on the position data and the flight envelope information, and estimation, identification and classification of the position, the speed and the acceleration state are realized.

Description

Technical Field

[0001] The present invention relates to the fields of multi-model target tracking technology, target recognition technology, and joint tracking and recognition technology, in particular to a flying target joint tracking and recognition method based on position and flight envelope information. Background Art

[0002] With the advancement of sensor technology, target tracking technology, and computer technology, a large number of target tracking systems for advanced and complex applications have emerged. While the low-altitude economy is a key national development direction and needs to be managed in an orderly manner, current means for detecting and identifying low-altitude flying targets are relatively lacking. Furthermore, low-altitude threats such as drones and birds are small, diverse, low-altitude, slow, and highly maneuverable, resulting in significant challenges in detecting, tracking, and identifying these targets in complex scenarios. To address this challenge, telepathic base stations and radars are among the most effective means of monitoring aerial targets, and target tracking is a crucial supporting technology. This involves the comprehensive use of information processing techniques, integrating prior target information with online measurement information provided by various sensors, to estimate and track various targets in the environment in real time.

[0003] With the development and widespread use of drones, and the accompanying increase in air traffic, tracking and identifying low-altitude targets such as drones and birds using base stations and radar has become increasingly important. First, the measurement data obtained by radar and telemetry base stations provides location information, and its all-weather and all-day capabilities meet the requirements for low-altitude target detection. Second, the widespread use of drones presents corresponding challenges for air traffic management. Base station sensing technology can monitor drones' position, speed, and altitude in real time, effectively avoiding collisions with other aircraft, ensuring air traffic safety and maintaining public order and social tranquility. Furthermore, the presence of birds also poses a potential threat to aircraft. Base station sensing technology allows for timely detection and tracking of bird flight paths, enabling proactive measures to mitigate the risk of collisions. Furthermore, this technology is crucial for distinguishing drones from other flying objects, such as birds. The appearance and flight characteristics of drones can be similar to those of other flying objects, such as birds, necessitating efficient recognition systems to ensure accurate identification of these different targets. In summary, tracking and identifying low-altitude flying targets based on position information measurement data is of great significance and value to both national defense and civilian fields.

[0004] In order to solve the problem of tracking and identifying low-altitude flying targets, providing a method for achieving high-precision and high-stability joint tracking and identification of flying targets based on position data has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The present invention addresses the problems of difficulty in discovering, tracking, and identifying flying threat targets in complex scenarios, such as the complex and diverse motion patterns of point targets, the nonlinearity of target motion models, and the uncertainty of target system patterns. A method for joint tracking and identification of flying targets based on position and flight envelope information is provided. The present invention introduces the flight envelope into the type identification of flying targets, thereby achieving accurate identification of targets; introduces the velocity acceleration input model into the trajectory tracking of flying targets, thereby achieving stable tracking of flying targets; and realizes high-precision and stable joint tracking and identification classification of flying targets based on position data and flight envelope information.

[0006] The present invention is achieved through the following technical solutions.

[0007] One aspect of the present invention provides a method for jointly tracking and identifying a flying target based on position and flight envelope information, comprising:

[0008] Based on the motion state of the flying target, a set of shared and analyzable motion models is constructed;

[0009] Receive the measurement data at the initial moment and initialize the flight target track and motion feature information based on the motion model set;

[0010] Based on the base station or radar measurement data obtained by the sensor at each moment, the optimal model expansion algorithm is used to filter and estimate the motion state of the flying target and extract the motion state characteristics of the flying target;

[0011] Calculate the acceleration-velocity flight envelope likelihood function based on the flight target motion state characteristics;

[0012] Based on the acceleration-velocity flight envelope likelihood function, the flying target is identified using the sequential hypothesis testing method.

[0013] If the likelihood ratio accumulation in the sequential hypothesis testing algorithm exceeds a certain decision boundary for the first time, the corresponding identification result is output; otherwise, no output is given, indicating an unidentified state.

[0014] Repeat the flight target motion state filtering estimation, feature extraction, likelihood function calculation and flight target identification until the flight target track estimation and identification classification are completed.

[0015] Preferably, a shared and analyzable set of motion models is constructed, including:

[0016] Add a uniform motion model to the model collection;

[0017] Add a uniform acceleration motion model to the model collection;

[0018] Add a planar uniform turning motion model to the model collection;

[0019] Add a three-dimensional uniform turning motion model to the motion model set;

[0020] Add velocity acceleration input model to the motion model collection.

[0021] Preferably, initializing the flight target track and motion characteristic information includes:

[0022] Convert the sensor measurements in the polar coordinate system at the initial moment to the rectangular coordinate system;

[0023] Initialize the target's state parameters, the probability values ​​of each motion model, the state parameters, and the model transition probability matrix based on the measurement data and prior information;

[0024] The state parameters of the target include the initial state of the target including position, velocity and acceleration information;

[0025] Initialize the probability value of each model to the average value.

[0026] Preferably, an optimal model expansion algorithm is used to filter and estimate the motion state of the flying target and extract the motion state characteristics of the flying target, including:

[0027] The state is predicted one step ahead using the variable structure multi-interaction model (VSIMM) algorithm based on the set of motion models activated at the previous moment.

[0028] The Kullback-Leiber criterion is used to measure the distance between the candidate model and the true pattern, the corresponding model is terminated, the model with the smallest distance is activated, and the basic model set is expanded and combined into the model set at that moment;

[0029] Obtain real measurements and use the variable structure multi-interaction model (VSIMM) algorithm to estimate and update the target state at the current moment.

[0030] Preferably, a variable structure multi-interaction model (VSIMM) algorithm is used to estimate and update the target state at the current moment, including:

[0031] Complete the status update and covariance matrix update of each model;

[0032] Update the probability of each model;

[0033] The fusion completes the overall state estimation.

[0034] Preferably, calculating the acceleration-velocity flight envelope likelihood function based on the flight target motion state characteristics includes:

[0035] Collect data on the speed and acceleration of a large number of real flying targets;

[0036] The statistical data is fitted with Gaussian distribution to obtain the corresponding six-dimensional joint probability density function of speed acceleration;

[0037] Based on the flight envelope information obtained by fitting and the speed and acceleration of the current flight target at that moment, the corresponding likelihood function value is calculated.

[0038] Preferably, a sequential hypothesis testing method is used to identify the flying target, including:

[0039] Take the logarithm of the calculated likelihood value of each model;

[0040] Construct multiple hypotheses, decompose the multiple hypothesis testing into multiple binary hypothesis testing, and use the sequential hypothesis testing algorithm to make step-by-step decisions, identify and distinguish flight targets.

[0041] Preferably, judging the recognition result includes:

[0042] Output the result of flight target tracking estimation;

[0043] If the cumulative likelihood ratio of the sequential hypothesis test exceeds a certain decision boundary for the first time, the corresponding identification result is output, otherwise no output is given.

[0044] Another aspect of the present invention provides a flying target joint tracking and identification device based on position and flight envelope information of the method, comprising:

[0045] A building module, based on the motion state of the flying target, is used to construct a set of sharable and analytic motion models;

[0046] The initialization module is used to receive the measurement data at the initial moment and initialize the flight target track and motion feature information based on the motion model set;

[0047] The extraction module is used to filter and estimate the motion state of the flying target using the optimal model expansion algorithm based on the measurement data of the base station or radar at each moment acquired by the sensor, and extract the motion state characteristics of the flying target;

[0048] A calculation module, used for calculating the acceleration-velocity flight envelope likelihood function based on the motion state characteristics of the flight target;

[0049] Target recognition module, used to identify flying targets using sequential hypothesis testing method;

[0050] The judgment module is used to judge whether the likelihood ratio accumulation in the sequential hypothesis testing algorithm exceeds a certain decision boundary for the first time and to judge the recognition state; it repeats the flight target motion state filtering estimation, feature extraction, calculation of likelihood function and flight target recognition to complete the flight target track estimation and recognition classification.

[0051] The present invention adopts the above technical solution, which has the following beneficial effects:

[0052] 1. Compared with traditional uniform velocity (CV) motion and uniform acceleration (CA) motion tracking models, the present invention introduces the flight envelope into the type identification of flying targets, analyzes and utilizes the multi-dimensional motion characteristics of the target, and realizes accurate target identification.

[0053] 2. This invention incorporates a velocity-acceleration input model into target trajectory tracking. It integrates analytical target modeling based on motion characteristics, flight envelope prior information, an optimal model expansion algorithm, and sequential hypothesis testing to establish a unified theoretical framework. Its tracking model and estimator are analytically formulated, computationally simple, and highly efficient. Leveraging the motion characterization capabilities of the velocity-acceleration input motion model, this invention achieves stable tracking of targets.

[0054] 3. The proposed method integrates analytical modeling of flight targets based on motion characteristics, prior information about the flight envelope, an optimal model expansion algorithm, and sequential hypothesis testing to establish a unified theoretical framework. Its tracking model and estimator are analytically formulated, computationally simple, and efficient. Ultimately, it achieves highly accurate and stable tracking, state estimation (position, velocity, acceleration), and identification and classification of flight targets based on position data and flight envelope information.

[0055] 4. The present invention can be applied to low-altitude economy, synaesthesia and other fields. Through the position measurement data of the sensor, it can achieve high-precision and high-stability joint tracking and identification of flying targets with complex and diverse motion patterns, solving the current problems faced by low-altitude supervision in complex scenarios where it is difficult to discover, track and identify flying threat targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute an improper limitation of the present invention. In the drawings:

[0057] Figure 1 This is a flow chart of a method for joint tracking and identification of flying targets based on position data and flight envelope information;

[0058] Figure 2 A framework for joint tracking and identification of flying targets based on position data and flight envelope information;

[0059] Figure 3 Expanding the methodological framework for optimal models;

[0060] Figures 4(a)-4(d) Accelerated flight envelope information for statistical drone and bird speeds;

[0061] Figures 5(a)-5(b)To measure the actual flight paths of birds and drones;

[0062] Figures 6(a)-6(d) Tracking and velocity estimation results using the method for flying birds and drones;

[0063] Figures 7(a)-7(b) Identify the results of sequential hypothesis testing for the corresponding bird and drone scenarios;

[0064] Figure 8 Schematic diagram of the flying target joint tracking and identification device based on position data and flight envelope information of the present invention. DETAILED DESCRIPTION

[0065] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0066] like Figure 1 As shown, an embodiment of the present invention provides a method for jointly tracking and identifying a flying target based on position data and flight envelope information, comprising the following steps:

[0067] S101: Based on the motion state of the flying target, a set of shared and analyzable motion models is constructed.

[0068] Specifically include:

[0069] 11) If the flight target has a uniform speed flight mode, add a uniform speed motion model to the model set;

[0070] 12) Add a uniform acceleration motion model to the model set;

[0071] 13) The flying target can hover in the air, and a plane uniform turning motion model is added to the model set;

[0072] 14) Considering the complex motion of the flying target, a three-dimensional uniform turning motion model is added to the motion model set;

[0073] 15) Add velocity acceleration input model to the motion model set.

[0074] Among them, the velocity acceleration input model is as follows:

[0075] The velocity acceleration input model state equation can be expressed in the following way:

[0076]

[0077] Here, the state vector is defined as x k =[x k ,y k ,z k ]T , x k ,y k ,z k represents the position of the target at time k on each axis, F is the state transfer matrix, u k =[v x ,v y ,v z ,a x ,a y ,a z ] T Including the velocity and acceleration of each dimension, which is the input of the system state, G is the control input matrix, w k is the process noise, Q k represents the covariance of the process noise, Represents 0 as the mean and Q k is the Gaussian distribution of variance, T is the time interval period, and I is the identity matrix.

[0078] In order to better approximate the real target in the optimal model extension BMA algorithm, the state is used as a common variable when optimizing the model selection. The present invention improves the acceleration input model with speed, expands the state variables into position and speed, and the improved model state equation is:

[0079]

[0080] in, represents the Kronecker product.

[0081] S102: Receive measurement data at the initial moment and initialize target track and motion feature information;

[0082] The specific steps include:

[0083] 21) Convert the initial polar coordinate sensor measurements to the rectangular coordinate system;

[0084] 22) Initialize the target's state parameters, the probability values ​​of each motion model, the state parameters, and the model transition probability matrix based on the measurement data and prior information.

[0085] The initialization method is as follows:

[0086] Initialize the state of the target based on the measurement data and prior information: the initial state of the target The covariance matrix P0 of the initial state of the target, the initial state of the target Including position, velocity and acceleration information, specifically For sensor measurement; In addition, the state parameters of different models have specific forms and need to be initialized. The probability value of each model is initialized to the average value. If there are 4 models, then the probability of each model is initialized to 0.25; In addition, the initialization model transition probability matrix is:

[0087]

[0088] S103: receiving the measurement data at each moment, using the optimal model expansion algorithm to filter and estimate the motion state of the flying target, and then extracting relevant flight target motion state features;

[0089] The specific steps include:

[0090] 31) Based on the set of motion models activated at the previous moment, the variable structure multi-interaction model (VSIMM) algorithm is used to perform one-step state prediction;

[0091] First, complete the reinitialization of the model:

[0092] Model probability prediction:

[0093]

[0094] Interaction weight:

[0095]

[0096] Interaction estimation:

[0097]

[0098] Interaction variance:

[0099]

[0100] Among them, M k-1 Represents the model set at the previous moment, represents the model that works at time k, z k-1 Represents the measurement at the previous moment, π j|i represents the prior probability from model i to model j, Represents the probability of each model at the previous moment, Represents the estimated result of the model at the previous moment, Represents the results of state estimation of each model after reinitialization, represents the interaction variance after reinitialization, E represents the expectation, and (·) represents that the content of this formula is the same as before.

[0101] Then complete the one-step state prediction of each model:

[0102] Prediction status:

[0103]

[0104] Forecast state variance:

[0105]

[0106] in, is the state transfer matrix of each model, is the external output at time k-1, is the control input matrix, is the one-step prediction state of each model, is the one-step prediction state covariance of each model, is the process noise of model j at time k-1.

[0107] 32) Use the Kullback-Leiber criterion to measure the distance between the candidate model and the true pattern, terminate the corresponding model, activate the model with the smallest distance, and expand the basic model set to form the model set at that moment;

[0108] Given the candidate model set at k moments The optimal model in can be selected as the corresponding model with the minimum KL criterion, that is:

[0109]

[0110] in, That is, the optimal model that can be activated by the model set adaptive algorithm, s k is the actual motion mode at the current moment, Defined as:

[0111]

[0112] Where: p[y|s k ]and s k and is the probability density function of the conditional y, ln is the logarithm, and ∫ represents the integral.

[0113] When the comparison needs to be performed online in the variable structure multi-model algorithm at time k, it is necessary to consider the k-1 and adaptive process information Provides real-time information, and assumes that y has a distribution Gaussian vector of It can be specifically given as:

[0114]

[0115] Where: n is the dimension of y, are the common variables of the true model, is the corresponding covariance matrix, is the common variable corresponding to the optimization model, is the corresponding covariance matrix, and tr represents the trace.

[0116] Take the state as the common variable used by the KL criterion, and Represents the accumulated measurement before k-1, based on the prediction model set The following results are obtained:

[0117]

[0118] in:

[0119]

[0120] Model

[0121]

[0122] The conditions for activation and termination of the KL criterion model set can be given as:

[0123] 1) Activated model set:

[0124]

[0125] 2) Terminated model set:

[0126]

[0127] 3) The retained model set:

[0128]

[0129] Among them, ε a and ε t They are the threshold values ​​for model set activation and model set termination, respectively, and are generally given a priori.

[0130] 33) Obtain real measurements and use the variable structure multi-interaction model (VSIMM) algorithm to estimate and update the target state at the current moment.

[0131] The state estimation update steps are as follows:

[0132] 331) Complete the state update and covariance matrix update of each model;

[0133] Status update of each model:

[0134]

[0135] Update the covariance matrix of each model state estimate:

[0136]

[0137] in, is the Kalman gain of each model at time k, is the measurement matrix of each model, z k is the real measurement at time k, is the measurement matrix, and I is the identity matrix.

[0138] 332) Update the probability of each model;

[0139] Likelihood functions of each model:

[0140]

[0141] Probability of each model:

[0142]

[0143] in, is the innovation covariance of each model, is a Gaussian distribution.

[0144] 333) The fusion completes the overall state estimation.

[0145] Overall state estimation:

[0146]

[0147] Population estimated covariance matrix:

[0148]

[0149] S104: Calculating a corresponding likelihood function based on the acceleration-velocity flight envelope information of the flight target;

[0150] The specific steps include:

[0151] 41) Collect statistics on the speed and acceleration data of multiple real flight targets [1, 2, 3...k];

[0152] 42) Fit the statistical data using Gaussian distribution to obtain the corresponding six-dimensional joint probability density function of speed acceleration;

[0153] 43) By fitting the statistical data with Gaussian distribution, the mean μ of the Gaussian distribution corresponding to the target is obtained. i , covariance matrix ∑ i . The Gaussian distribution is defined as:

[0154]

[0155] 44) Based on the flight envelope information obtained by fitting and the speed and acceleration of the current flight target at that moment, calculate the corresponding likelihood function value N i [v,a|μ,∑].

[0156] S105: Identify the flying target using a sequential hypothesis testing method based on the calculated likelihood function to obtain an identification result;

[0157] The specific steps include:

[0158] 51) Taking the logarithm of the calculated likelihood values ​​of each model;

[0159] 52) Construct multiple hypotheses, decompose the multiple hypothesis testing into multiple binary hypothesis testing, and use the sequential hypothesis testing algorithm to make step-by-step decisions, identify and distinguish flight targets.

[0160] In one stage, the sequential likelihood ratio of hypothesis H1: the flight target is type 1 compared to hypothesis H2: the flight target is type 2 is

[0161]

[0162] The corresponding decision rule can be given by the following formula

[0163]

[0164] Where logA and logB are two decision boundaries, is N[v,a|μ1,∑1], is N[v,a|μ2,∑2].

[0165] The hypothesis accepted in the previous stage is then tested against the remaining hypotheses step by step until all hypothesis tests are completed.

[0166] S106: Output the flight target's track and identification and classification results.

[0167] The specific steps include:

[0168] 61) Output the result of flight target tracking estimation;

[0169] 62) If the cumulative likelihood ratio of the sequential hypothesis test exceeds a certain decision boundary for the first time, the corresponding identification result is output; otherwise, no output is given and the state is unidentified.

[0170] Repeat steps 103-105 to perform flight target motion state filtering estimation, feature extraction, likelihood function calculation, and flight target identification until flight target track estimation and identification classification are completed.

[0171] The present invention is further illustrated below by using a specific embodiment to track and identify drones and flying birds.

[0172] S101: Drones and birds are both flying targets in the airspace and have the same motion modes, such as linear motion, circling, and accelerated motion. Based on this characteristic, a common motion model set for drones and birds is established, and a uniform motion model, a uniform acceleration motion model, a plane uniform turning model, a three-dimensional turning model, and a velocity-acceleration input model are added to the model set. Specifically, in order to better approximate the real target in the optimal model expansion BMA algorithm and to use the state as a common variable when optimizing the model selection, the present invention improves the velocity-acceleration input model and expands the state variables to position and velocity. This constructs a model set that can describe the common motion of drones and birds.

[0173] S102: Receive measurement data from sensors such as base stations and radars at the start time, convert it to a sensor rectangular coordinate system with the corresponding sensor as the origin, and then convert it to an absolute rectangular coordinate system based on the position and posture of the sensor as appropriate; then initialize the target state, the initial state of each model, and the probability of each model.

[0174] S103: After initialization, the variable structure multi-interaction model algorithm is used to predict the model set at the current moment and obtain the state prediction parameters of each model at time k. and Then, based on the KL criterion, the distance between the candidate model and the true model is measured, the corresponding model is terminated, the model with the smallest distance is activated, and the basic model set is expanded and combined into the model set at that moment Then, the sensor measurement at time k is received and converted into the sensor rectangular coordinate system with the corresponding sensor, such as a base station or radar, as the origin. Then, depending on the situation, it is converted into an absolute rectangular coordinate system based on the position and posture of the sensor. The variable structure multi-interaction model (VSIMM) algorithm update part is then used to estimate and update the target state of each model at the current moment, and finally the probability of each model and the overall estimation result are obtained.

[0175] S104: First, collect a large amount of speed and acceleration data of real birds and drones, and then use six-dimensional Gaussian fitting to obtain the corresponding flight envelope information, that is, the joint probability density function of speed and acceleration N[v,a|μ 飞鸟 ,∑ 飞鸟 ] and N[v,a|μ 无人机 ,∑ 无人机 ], this function only needs to be statistically fitted once; then, based on the target speed and acceleration estimated at the current moment, the corresponding likelihood value is calculated.

[0176] S105: First, take the logarithm of the calculated likelihood values ​​of the corresponding drone and bird models, logN[v,a|μ 飞鸟 ,∑ 飞鸟 ] and logN[v,a|μ 无人机 ,∑ 无人机 ]; then use the classification threshold boundaries of the two hypotheses to identify and classify drones and birds by feeding them into the sequential hypothesis testing algorithm.

[0177] S106: Output the target tracking estimation result. If the cumulative likelihood ratio in the sequential hypothesis testing algorithm exceeds a certain decision boundary for the first time, the corresponding recognition result is output; otherwise, no output is given, indicating an unrecognized state. Then, the process proceeds to step 3 and repeats the subsequent process until the termination condition is met, such as the system terminating.

[0178] Figure 2 This framework is a joint tracking and recognition method for flying targets based on position data and flight envelope information. It consists of two main components: target tracking and target recognition. The target tracking component uses the optimal extended model algorithm (BMA) to track and filter the corresponding target motion features. The target recognition component uses the sequential hypothesis testing method that incorporates the flight envelope to calculate the target correspondence likelihood based on the target features and perform sequential hypothesis testing (SPRT) to complete the target recognition.

[0179] Figure 3 This is a framework for optimal model expansion methods. It includes a fixed model set and a candidate model set. This method reinitializes each model based on its previous state, performs multi-model tracking through model adaptation, and then updates the probabilities of each model and the final fused estimate of the target state.

[0180] Figures 4(a)-4(d) The joint distribution of velocity and acceleration based on a large amount of measured acceleration and speed data of drones and birds is shown. Figure 4(a) shows the flight envelopes of drones and birds at a confidence level of 99%. Figures 4(b)-4(d) is the distribution of velocity and acceleration.

[0181] Figure 5(a) shows the flight trajectory of a real bird, and Figure 5(b) shows the flight trajectory of a real drone, with a sampling interval of T = 0.64s. The real trajectories of the bird and the drone show a high degree of similarity, with both exhibiting flight states such as straight flight, circling, and turning. Therefore, it is difficult to distinguish the two based solely on their trajectories.

[0182] Figure 6(a) shows the estimated result of the flight trajectory of a real bird, Figure 6(b) shows the estimated result of the flight speed of a bird, Figure 6(c) shows the estimated result of the flight trajectory of a real drone, and Figure 6(d) shows the estimated result of the flight speed of a drone. It can be seen that the method of the present invention can achieve stable tracking of the trajectories of birds and drones, and at the same time achieve high-precision estimation of their flight status.

[0183] Figure 7(a) shows the sequential hypothesis test likelihood ratio recognition results for a scene where the object is actually a bird, and Figure 7(b) shows the sequential hypothesis test likelihood ratio recognition results for a scene where the object is actually a drone. In this step, H0 assumes a bird and H1 assumes a drone. The upper threshold is the threshold for identifying it as a bird, and the lower threshold is the threshold for identifying it as a drone. Figures 7(a) and (b) show that the method of the present invention can achieve accurate and stable recognition of drones and birds, demonstrating the feasibility and effectiveness of the method of the present invention.

[0184] The above experimental results verify that the present invention can achieve high-precision and high-stability tracking and identification and classification of drones and flying birds based on position measurement data, and has great theoretical and practical value.

[0185] According to an exemplary embodiment of the present invention, Figure 8 As shown, a flying target joint tracking and identification device 100 based on position and flight envelope information is used to implement the method, including:

[0186] A construction module 110 is used to construct a set of sharable and analyzable motion models based on the motion state of the flying target;

[0187] Initialization module 120, for receiving measurement data at an initial moment, and initializing flight target track and motion feature information based on a set of motion models;

[0188] The extraction module 130 is used to filter and estimate the motion state of the flying target using the optimal model expansion algorithm based on the measurement data of the base station or radar at each moment acquired by the sensor, and extract the motion state characteristics of the flying target;

[0189] A calculation module 140 is used to calculate an acceleration-velocity flight envelope likelihood function based on the flight target motion state characteristics;

[0190] A target identification module 150 is used to identify flying targets using a sequential hypothesis testing method;

[0191] The judgment module 160 is used to determine whether the likelihood ratio accumulation in the sequential hypothesis testing algorithm exceeds a certain decision boundary for the first time, and to determine the recognition state; repeatedly perform the flight target motion state filtering estimation, feature extraction, likelihood function calculation and flight target recognition to complete the flight target track estimation and recognition classification.

[0192] The present invention is not limited to the above-mentioned embodiments. On the basis of the technical solutions disclosed in the present invention, those skilled in the art can make some substitutions and modifications to some of the technical features therein according to the disclosed technical content without creative labor, and these substitutions and modifications are all within the protection scope of the present invention.

Claims

1. A method for joint tracking and identification of flying targets based on position and flight envelope information, characterized in that: include: Based on the motion state of the flying target, a set of shared and analyzable motion models is constructed; Receive the measurement data at the initial moment and initialize the flight target track and motion characteristic information based on the motion model set; Based on the base station or radar measurement data obtained by the sensor at each moment, the optimal model expansion algorithm is used to filter and estimate the motion state of the flying target and extract the motion state characteristics of the flying target; Calculate the acceleration-velocity flight envelope likelihood function based on the flight target motion state characteristics; Based on the acceleration-velocity flight envelope likelihood function, the flying target is identified using the sequential hypothesis testing method. Determine the recognition result. If the cumulative likelihood ratio in the sequential hypothesis testing algorithm exceeds a certain decision boundary for the first time, the corresponding recognition result is output; Otherwise, no output is given and the status is unrecognized; Repeat the flight target motion state filtering estimation, feature extraction, likelihood function calculation and flight target identification until the flight target track estimation and identification classification are completed.

2. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 1, characterized in that: Build a collection of interoperable and analytic motion models, including: Add a uniform motion model to the model collection; Add a uniform acceleration motion model to the model collection; Add a planar uniform turning motion model to the model collection; Add a three-dimensional uniform turning motion model to the motion model set; Add velocity acceleration input model to the motion model collection.

3. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 2, characterized in that: In the velocity acceleration input model, the state variables are expanded to position and velocity. The improved model state equation is: in, represents the Kronecker product, and the state vector is defined as x k ,y k ,z k represents the position of the target at time k on each axis, F is the state transfer matrix, u k =[v x ,v y ,v z ,a x ,a y ,a z ] T , including the velocity and acceleration of each dimension, is the input of the system state, G is the control input matrix, w k is the process noise, Q k represents the covariance of the process noise, Represents 0 as the mean and Q k is the Gaussian distribution of variance, T is the time interval period, and I is the identity matrix.

4. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 1, characterized in that: Initialize the flight target track and motion characteristic information, including: Convert the sensor measurements in the polar coordinate system at the initial moment to the rectangular coordinate system; Initialize the target's state parameters, the probability values ​​of each motion model, the state parameters, and the model transition probability matrix based on the measurement data and prior information; The state parameters of the target include the initial state of the target including position, velocity and acceleration information; Initialize the probability value of each model to the average value.

5. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 1, characterized in that: Use the optimal model expansion algorithm to filter and estimate the flight target motion state and extract the flight target motion state characteristics, including: The state is predicted one step ahead using the variable structure multi-interaction model (VSIMM) algorithm based on the set of motion models activated at the previous moment. The Kullback-Leiber criterion is used to measure the distance between the candidate model and the true pattern, the corresponding model is terminated, the model with the smallest distance is activated, and the basic model set is expanded and combined into the model set at that moment; Obtain real measurements and use the variable structure multi-interaction model (VSIMM) algorithm to estimate and update the target state at the current moment.

6. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 5, characterized in that: Use the variable structure multi-interaction model (VSIMM) algorithm to estimate and update the current target state, including: Complete the status update and covariance matrix update of each model; Update the probability of each model; The fusion completes the overall state estimation.

7. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 1, characterized in that: Calculate the acceleration-velocity flight envelope likelihood function based on the flight target motion state characteristics, including: Collect data on the speed and acceleration of a large number of real flying targets; The statistical data is fitted with Gaussian distribution to obtain the corresponding six-dimensional joint probability density function of speed acceleration; Based on the flight envelope information obtained by fitting and the speed and acceleration of the current flight target at that moment, the corresponding likelihood function value is calculated.

8. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 1, characterized in that: Use sequential hypothesis testing method to identify flying targets, including: Take the logarithm of the calculated likelihood value of each model; Construct multiple hypotheses, decompose the multiple hypothesis testing into multiple binary hypothesis testing, and use the sequential hypothesis testing algorithm to make step-by-step decisions, identify and distinguish flight targets.

9. The method for joint tracking and identification of flying targets based on position and flight envelope information according to claim 1, characterized in that: Determine the recognition results, including: Output the result of flight target tracking estimation; If the cumulative likelihood ratio of the sequential hypothesis test exceeds a certain decision boundary for the first time, the corresponding identification result is output, otherwise no output is given.

10. A flying target joint tracking and identification device based on position and flight envelope information according to the method of any one of claims 1 to 9, characterized in that: include: A building module, based on the motion state of the flying target, is used to construct a set of sharable and analytic motion models; The initialization module is used to receive the measurement data at the initial moment and initialize the flight target track and motion feature information based on the motion model set; The extraction module is used to filter and estimate the motion state of the flying target using the optimal model expansion algorithm based on the measurement data of the base station or radar at each moment acquired by the sensor, and extract the motion state characteristics of the flying target; A calculation module, used for calculating the acceleration-velocity flight envelope likelihood function based on the motion state characteristics of the flight target; Target recognition module, used to identify flying targets using sequential hypothesis testing method; The judgment module is used to determine whether the likelihood ratio accumulation in the sequential hypothesis testing algorithm exceeds a certain decision boundary for the first time and to determine the recognition status; it repeats the flight target motion state filtering estimation, feature extraction, likelihood function calculation and flight target recognition to complete the flight target track estimation and recognition classification.

Images