A nonlinear scene multi-target tracking method, system and storage medium

Abstract

The application discloses a kind of nonlinear scene multi-target tracking method, system and storage medium, method includes: by constructing nonlinear motion model and measurement model, specifically includes: by nonlinear model represents the nonlinear state process and measurement process;According to the multi-target tracking in two-dimensional plane, the motion equation of each target is determined to be set as cooperative turning model, the motion equation of target is determined;System state transition matrix, process noise distribution matrix, measurement equation are constructed;According to the nonlinear motion model and measurement model, the extraction of multi-target state estimation is carried out by FAUKF-PHD filter, obtains the state vector of each target, completes the tracking of each target;Based on multiple preset scene information, the tracking result of each target is verified.The application has high tracking precision and low cost, can improve the effectiveness of method, and can be widely applied in computer technology field.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, system, and storage medium for multi-target tracking in nonlinear scenes. Background Technology

[0002] Holographic staring radar is a multi-functional radar that simultaneously covers the entire airspace. Its key feature is the use of a wide beam for transmission and the generation of multiple narrow beams for reception using digital beamforming to cover the transmission area. The application of a low-gain, wide beam weakens the received target echo signal, necessitating long-term accumulation to improve target detection probability. This long-term accumulation allows holographic staring radar to achieve higher gain and Doppler resolution, enabling effective separation of moving targets from clutter. Like traditional radar, target tracking is a classic problem in holographic radar surveillance. How to track multiple targets accurately and in real-time in a cluttered environment while shielding them from external interference such as clutter remains a challenge and key issue in target tracking. Currently, the two main multi-target tracking (MTT) technologies are data association (DA) based methods and random finite set (RFS) based methods. Data association-based MTT methods transform the multi-target tracking problem into multiple single-target tracking problems, including classic algorithms such as Multiple Hypothesis Tracking (MHT) and Joint Probabilistic Data Association (JPDA). These algorithms achieve good tracking results in scenarios with low environmental clutter, a fixed number of targets, and high signal-to-noise ratios. However, when dealing with more complex multi-target problems, data association errors are prone to occur, leading to a decline in tracking performance. Furthermore, the computational load increases exponentially when the number of targets and false alarms is large. Mahler's MTT method based on random finite sets transforms the multi-target tracking problem into a set-valued estimation problem, using a probability hypothesis density function (PHD) to describe the target's state and observation values. Therefore, it can achieve multi-target state estimation within a Bayesian framework. Compared to data association algorithms, it avoids explicit association between states and measurements, eliminates complex data association calculations, and significantly improves efficiency. As an emerging multi-target tracking method, it is gradually becoming a research hotspot in the MTT field.

[0003] PHD filters typically fail to yield closed-form solutions due to the computationally difficult ensemble integrals involved in their iteration process. To achieve this, Vo et al. proposed two main approximation methods: the Sequential Monte Carlo PHD (SMC-PHD) filter and the Gaussian Mixture PHD (GM-PHD) filter. The former's nonlinear fitting characteristic makes it suitable for nonlinear, non-Gaussian environments. However, the SMC-PHD filter suffers from drawbacks such as particle degradation and difficulty in extracting the states of multiple targets. Furthermore, given the high real-time performance requirements in radar multi-target tracking, increasing the number of particles to improve tracking accuracy significantly reduces computational efficiency and is impractical for engineering applications. The latter represents the probability distribution of the target's existence as a Gaussian mixture model. The update and prediction of the probability density function can be accomplished by operating on the Gaussian mixture parameters, reducing the computational burden. However, the GM-PHD filter only considers the linear motion of the target and the linear measurement model. In practical applications, sensors such as radar obtain nonlinear distance and angle information in the measurement equation. Faced with nonlinear models, the filter may have problems with inaccurate estimation of the target state or the number of targets.

[0004] Therefore, to balance real-time performance and tracking accuracy, improvements to the GM-PHD filter are considered to make it suitable for nonlinear environments. In single-target filtering, the analytical approximations of nonlinear Bayesian filters mainly include the Extended Kalman Filter (EKF) and the Insensitive Kalman Filter (UKF). Therefore, a similar approach can be used to extend GM-PHD to be compatible with nonlinear Gaussian motion models and measurement models. Similar to the single-target case, the EKF-PHD filter linearizes the nonlinear system through Taylor series expansion. However, the approximation algorithm results in large errors, limiting its application to differentiable nonlinear models. Furthermore, the calculation of the Jacobian matrix is ​​lengthy and error-prone. The UKF-PHD filter uses unscented transformations to handle the nonlinear propagation of the mean and covariance. It does not require differentiation of the Jacobian matrix and does not neglect higher-order terms, thus achieving higher computational accuracy for statistics of nonlinear distributions. The algorithm is simple and highly accurate, making it more suitable for nonlinear radar surveillance scenarios.

[0005] Holographic staring radar echoes contain a wealth of characteristic information, such as range, azimuth, amplitude, signal-to-noise ratio, and Doppler information. Since certain characteristic information differs between targets and clutter, and even among different targets themselves, one or more of these characteristic information can be used to distinguish between targets and clutter, as well as between different targets. This characteristic information can be used to correct the weights of Gaussian components, thereby significantly reducing the weight of clutter and preventing the merging of different targets. In scenarios with a large number of clutter targets and a low detection probability, this effectively improves the tracking accuracy of multi-target radar. Furthermore, during the pruning and merging steps, Doppler information can be used to further eliminate clutter, reduce iterative components, and improve real-time performance.

[0006] In the field of multi-target tracking algorithms based on random finite sets, the following main problems exist:

[0007] 1. Few studies have focused on multi-target tracking methods that utilize radar echo information, and features such as Doppler, amplitude, and signal-to-noise ratio have not been effectively utilized.

[0008] 2. Existing feature-assisted GM-PHD filters are only applicable to Gaussian linear models, and have large tracking errors when tracking nonlinear targets.

[0009] 3. Existing feature-assisted SMC-PHD filters require a large number of particles and additional clustering operations to extract multi-objective state estimates, resulting in high computational costs and difficulty in engineering implementation.

[0010] 4. Such algorithms are rarely used in practical engineering applications and lack verification with measured data. Summary of the Invention

[0011] The main objective of this invention is to propose a multi-target tracking method, system, and storage medium for nonlinear scenes with high tracking accuracy and low cost, which can improve the effectiveness of the method.

[0012] To achieve the above objectives, one aspect of this invention proposes a multi-target tracking method for nonlinear scenes, comprising the following steps:

[0013] The nonlinear motion model and measurement model are constructed, specifically including: representing the nonlinear state process and measurement process through the nonlinear model; determining the motion equation of each target based on multi-target tracking in a two-dimensional plane and setting it as a cooperative turning model to determine the motion equation of the target; and constructing the system state transition matrix, process noise distribution matrix, and measurement equation.

[0014] Based on the aforementioned nonlinear motion model and measurement model, the state estimation of multiple targets is extracted using the FAUKF-PHD filter to obtain the state vector of each target, thereby completing the tracking of each target.

[0015] The tracking results of each target are verified based on multiple preset scenario information.

[0016] In some embodiments, constructing the nonlinear motion model and measurement model includes the following steps:

[0017] Nonlinear state processes are represented by nonlinear models.

[0018] Considering multi-target tracking in a two-dimensional plane, the motion equations of each target are set as a cooperative turning model.

[0019] In some embodiments, the expression for the nonlinear state process is:

[0020] x k =f k|k-1 (x k-1 q k-1 )

[0021] z k =h k (x k r k )

[0022] Where, x k and z k Let f represent the target state and measurement set at time k, respectively; k|k-1 and h k These are the nonlinear state transition function and the measurement function, respectively; Gaussian noise q k-1 and r k The process noise and measurement noise are zero-mean, respectively, and their corresponding covariance matrices are Q. k-1 and R k ;

[0023] The equation of motion for the target is expressed as follows:

[0024] x k =F(ω) k-1 )x k-1 +Gω k-1

[0025] Where F represents the system state transition matrix; G represents the process noise distribution matrix; ω k-1 This represents process noise.

[0026] In some embodiments, the system state transition matrix F is expressed as:

[0027]

[0028] Where T represents the sampling interval;

[0029] The expression for the process noise distribution matrix G is:

[0030]

[0031] The expression for the process noise is:

[0032]

[0033] Among them, v k x is Gaussian white noise with zero mean. x,k and x y，k This represents the position and state of the target at time k.

[0034] In some embodiments, the step of extracting the state vector of each target by using a FAUKF-PHD filter for multi-target state estimation based on the nonlinear motion model and measurement model includes the following steps:

[0035] Construct a Gaussian mixture of the posterior intensities at time k-1 to obtain a new measurement set;

[0036] Based on the new measurement set, the mean, covariance, and weight of the posterior intensity are calculated using the mean, covariance, and weight of the predicted intensity.

[0037] Based on the feature-assisted UKF-PHD filter, the Gaussian component weights are corrected using high-precision Doppler information in the holographic staring radar echo.

[0038] The minor components are pruned according to the Gaussian mixed pruning technique, and similar components are merged.

[0039] Based on the final Gaussian mixture representation, extract the state vectors of all targets.

[0040] In some embodiments, in the step of constructing the Gaussian mixture form of the posterior intensity at time k-1 to obtain a new measurement set, the expression of the Gaussian mixture form is:

[0041]

[0042] Among them, v k-1 (x) represents the posterior strength at time k-1; J k-1 Represents the total number of given new target components; These represent the weights, mean, and covariance of the i-th mixture component representing the intensity of new formation, respectively. This represents a Gaussian density with mean m and covariance P;

[0043] The expression for the process of calculating the mean, covariance, and weight of the posterior intensity based on the new measurement set and the predicted intensity's mean, covariance, and weight is as follows:

[0044]

[0045] Among them, v k (x) represents the posterior strength at time k; p D,k Z represents the probability of object detection; k This represents information about a measurement point at time k;

[0046] The step of correcting the Gaussian component weights using the high-precision Doppler information in the holographic staring radar echo based on the feature-assisted UKF-PHD filter includes the following steps:

[0047] Using the location information in the measurement set, a Gaussian probability density function is used to describe the degree of matching between the location information in the measurement set and the location information contained in the predicted Gaussian component;

[0048] Update the weights of the Gaussian components with the corrected likelihood values;

[0049] Substitute the new weights into the posterior strength to complete the feature matching step;

[0050] The step of extracting the state vectors of all targets based on the final Gaussian mixture representation includes the following steps:

[0051] The number of targets is estimated by obtaining the sum of weights based on the Gaussian mixture expression of the posterior intensity.

[0052] Select the component with the largest corresponding weight as the state estimate, or extract the mean of the Gaussian components with weights greater than a preset threshold as the state vector of all targets existing at time k.

[0053] In some embodiments, the verification of the tracking results of each target based on multiple preset scene information includes the following steps:

[0054] Based on the preset first detection probability and first clutter number, the tracking performance of the filter is compared with the average OSPA distance and the number of targets, and the real-time performance of the filter is compared with the running time.

[0055] Based on the preset detection probability, the number of clutter is gradually increased to determine how the average OSPA distance of different filters changes with the increase of the number of clutter.

[0056] Based on a preset clutter number, the detection probability is gradually reduced to determine how the average OSPA distance of different filters changes as the detection probability decreases.

[0057] Based on the preset second detection probability and second clutter number, the number of targets for different filters is compared, and the tracking results of each target are verified.

[0058] Another aspect of this invention provides a multi-target tracking system for nonlinear scenes, comprising:

[0059] The first module is used to construct nonlinear motion and measurement models, specifically including: representing nonlinear state and measurement processes through nonlinear models; determining the motion equations of each target based on multi-target tracking in a two-dimensional plane, setting up a cooperative turning model, and determining the motion equations of the targets; constructing the system state transition matrix, process noise distribution matrix, and measurement equations.

[0060] The second module is used to extract the state estimation of multiple targets through the FAUKF-PHD filter based on the nonlinear motion model and measurement model, so as to obtain the state vector of each target and complete the tracking of each target.

[0061] The third module is used to verify the tracking results of each target based on multiple preset scenario information.

[0062] Another aspect of the present invention provides an electronic device, including a processor and a memory;

[0063] The memory is used to store programs;

[0064] The processor executes the program to implement the method described above.

[0065] Another aspect of this invention provides a computer-readable storage medium storing a program that is executed by a processor to implement the methods described above.

[0066] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method.

[0067] The embodiments of this invention include at least the following beneficial effects: This invention provides a method, system, and storage medium for multi-target tracking in nonlinear scenes. This scheme constructs a nonlinear motion model and a measurement model, specifically including: representing the nonlinear state process and measurement process using a nonlinear model; determining the motion equation of each target as a cooperative turning model based on multi-target tracking in a two-dimensional plane; constructing a system state transition matrix, a process noise distribution matrix, and measurement equations; extracting multi-target state estimates using a FAUKF-PHD filter based on the nonlinear motion model and measurement model to obtain the state vector of each target, thus completing the tracking of each target; and verifying the tracking results of each target based on multiple preset scene information. This invention offers high tracking accuracy and low cost, improving the effectiveness of the method. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present invention;

[0069] Figure 2 This is a flowchart of the overall steps provided in the embodiments of the present invention;

[0070] Figure 3 This is a target real trajectory map provided in the embodiments of the present invention;

[0071] Figure 4 This is a graph showing the change of potential estimates over time for different filters provided in this embodiment of the invention;

[0072] Figure 5 This is a graph showing the variation of the average OSPA distance over time for different filters provided in this embodiment of the invention;

[0073] Figure 6 This is a comparison chart of the running times of different filters provided in the embodiments of the present invention;

[0074] Figure 7 This is a graph showing how the average OSPA distance of different filters varies with the number of clutter particles, as provided in the embodiments of the present invention.

[0075] Figure 8 This is a graph showing the change in average OSPA distance of different filters as the detection probability decreases, provided in the embodiments of the present invention.

[0076] Figure 9 This is a comparison chart of potential estimation for different filters in a low detection probability and high clutter wavenumber scenario provided by an embodiment of the present invention;

[0077] Figure 10 This is a diagram showing the processing results of the FAUKF-PHD filter provided in this embodiment of the invention on the measured data. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0079] It is understood that the terms “first,” “second,” etc., used in this invention may be used herein to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are used only to distinguish one concept from another. For example, first information may also be referred to as second information without departing from the scope of embodiments of the invention, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to determination” as used herein may be interpreted as “when…” or “when…” or “in response to determination.”

[0080] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0081] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0082] Before providing a detailed description of the embodiments of the present invention, some related technologies involved in the embodiments of the present invention will be described first, as follows:

[0083] Holographic staring radar: Holographic staring radar is an advanced radar system. Unlike the mechanical scanning mode of traditional radar systems, it can monitor multiple directions in real time without mechanical movement. Its key feature is the use of a wide beam for transmission and a staring array for reception, simultaneously forming multiple narrow beams to cover the transmission area, thus improving the radar system's monitoring range and efficiency.

[0084] Multi-target tracking: Multi-target tracking technology is used to simultaneously detect, identify, and track multiple targets in dynamic environments. Its main task is to accurately identify the trajectories of multiple targets from continuous observation data acquired by sensors, addressing the practical need to simultaneously obtain the number of targets and information about multiple targets from a set of measurements from unknown target sources.

[0085] UKF (Unscented Kalman Filter) is a filtering algorithm for state estimation of nonlinear systems. UKF approximates the nonlinear function through an unscented transformation, that is, by selecting a set of deterministic sampling points to approximate the propagation of the distribution, thus avoiding first-order linearization of the nonlinear function and improving filtering accuracy and robustness.

[0086] PHD filtering: PHD (Probability Hypothesis Density) filtering is a Bayesian filtering method for multi-target tracking, designed to estimate and track the number and state distribution of targets without explicit data association. PHD filtering describes the distribution of targets in space using a probability hypothesis density function, and predicts and updates the function using a recursive Bayesian update formula.

[0087] GM-PHD Filtering: The GM-PHD (Gaussian Mixture PHD) filter is an advanced filtering algorithm for multi-target tracking. It combines the advantages of PHD filtering and Gaussian mixture models, describing the dynamic distribution of the target through a probability hypothesis density function and using a Gaussian mixture model to approximate and update this density function, which greatly simplifies computational complexity.

[0088] SMC-PHD Filter: The SMC-PHD (Sequential Monte Carlo PHD) filter is a multi-target tracking algorithm based on particle filtering. It combines the theoretical framework of PHD filtering with the numerical computation capabilities of the sequential Monte Carlo method. It uses a set of weighted random particles to approximate and update the probability hypothesis density function, aiming to solve the problem of target quantity and state estimation in dynamic environments.

[0089] The nonlinear scene multi-target tracking method, system, and storage medium provided in this invention relate to the field of computer technology. The nonlinear scene multi-target tracking method provided in this invention can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited thereto; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network; the software can be an application implementing the nonlinear scene multi-target tracking method, but is not limited to the above forms.

[0090] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0091] like Figure 1 The diagram shown is a schematic representation of an implementation environment provided by an embodiment of the present invention. (Refer to...) Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0092] Server 101 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0093] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0094] Terminal 102 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc. It can also be a vehicle-mounted terminal of the various device types described above, but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the invention does not impose any limitations.

[0095] Exemplary based on Figure 1 The implementation environment shown in this embodiment of the invention provides a nonlinear scene multi-target tracking method. The following description uses the application of this nonlinear scene multi-target tracking method in server 101 as an example. It can be understood that this method can also be applied to terminal 102.

[0096] Reference Figure 2 , Figure 2 The flowchart illustrates a multi-target tracking method for nonlinear scenes applied to a server, as provided in this embodiment of the invention. The execution subject of this method can be any of the aforementioned computer devices (including a server or a terminal). (Refer to...) Figure 2 The method may include the following steps:

[0097] The nonlinear motion model and measurement model are constructed, specifically including: representing the nonlinear state process and measurement process through the nonlinear model; determining the motion equation of each target based on multi-target tracking in a two-dimensional plane and setting it as a cooperative turning model to determine the motion equation of the target; and constructing the system state transition matrix, process noise distribution matrix, and measurement equation.

[0098] Based on the aforementioned nonlinear motion model and measurement model, the state estimation of multiple targets is extracted using the FAUKF-PHD filter to obtain the state vector of each target, thereby completing the tracking of each target.

[0099] The tracking results of each target are verified based on multiple preset scenario information.

[0100] In some embodiments, constructing the nonlinear motion model and measurement model includes the following steps:

[0101] Nonlinear state processes are represented by nonlinear models.

[0102] Considering multi-target tracking in a two-dimensional plane, the motion equations of each target are set as a cooperative turning model.

[0103] In some embodiments, the expression for the nonlinear state process is:

[0104] x k =f k|k-1 (x k-1 q k-1 )

[0105] z k =h k (x k r k )

[0106] Where, x k and z k Let f represent the target state and measurement set at time k, respectively; k|k-1 and h k These are the nonlinear state transition function and the measurement function, respectively; Gaussian noise q k-1 and r k The process noise and measurement noise are zero-mean, respectively, and their corresponding covariance matrices are Q. k-1 and R k ;

[0107] The equation of motion for the target is expressed as follows:

[0108] x k =F(ω) k-1 )x k-1 +Gω k-1

[0109] Where F represents the system state transition matrix; G represents the process noise distribution matrix; ω k-1 This represents process noise.

[0110] In some embodiments, the system state transition matrix F is expressed as:

[0111]

[0112] Where T represents the sampling interval;

[0113] The expression for the process noise distribution matrix G is:

[0114]

[0115] The expression for the process noise is:

[0116]

[0117] Among them, v k x is Gaussian white noise with zero mean. x,k and x y，k This represents the position and state of the target at time k.

[0118] In some embodiments, the step of extracting the state vector of each target by using a FAUKF-PHD filter for multi-target state estimation based on the nonlinear motion model and measurement model includes the following steps:

[0119] Construct a Gaussian mixture of the posterior intensities at time k-1 to obtain a new measurement set;

[0120] Based on the new measurement set, the mean, covariance, and weight of the posterior intensity are calculated using the mean, covariance, and weight of the predicted intensity.

[0121] Based on the feature-assisted UKF-PHD filter, the Gaussian component weights are corrected using high-precision Doppler information in the holographic staring radar echo.

[0122] The minor components are pruned according to the Gaussian mixed pruning technique, and similar components are merged.

[0123] Based on the final Gaussian mixture representation, extract the state vectors of all targets.

[0124] In some embodiments, in the step of constructing the Gaussian mixture form of the posterior intensity at time k-1 to obtain a new measurement set, the expression of the Gaussian mixture form is:

[0125]

[0126] Among them, v k-1 (x) represents the posterior strength at time k-1; J k-1 Represents the total number of given new target components; These represent the weights, mean, and covariance of the i-th mixture component representing the intensity of new formation, respectively. This represents a Gaussian density with mean m and covariance P;

[0127] The expression for the process of calculating the mean, covariance, and weight of the posterior intensity based on the new measurement set and the predicted intensity's mean, covariance, and weight is as follows:

[0128]

[0129] Among them, v k (x) represents the posterior strength at time k; p D,k Z represents the probability of object detection; k This represents information about a measurement point at time k;

[0130] The step of correcting the Gaussian component weights using the high-precision Doppler information in the holographic staring radar echo based on the feature-assisted UKF-PHD filter includes the following steps:

[0131] Using the location information in the measurement set, a Gaussian probability density function is used to describe the degree of matching between the location information in the measurement set and the location information contained in the predicted Gaussian component;

[0132] Update the weights of the Gaussian components with the corrected likelihood values;

[0133] Substitute the new weights into the posterior strength to complete the feature matching step;

[0134] The step of extracting the state vectors of all targets based on the final Gaussian mixture representation includes the following steps:

[0135] The number of targets is estimated by obtaining the sum of weights based on the Gaussian mixture expression of the posterior intensity.

[0136] Select the component with the largest corresponding weight as the state estimate, or extract the mean of the Gaussian components with weights greater than a preset threshold as the state vector of all targets existing at time k.

[0137] In some embodiments, the verification of the tracking results of each target based on multiple preset scene information includes the following steps:

[0138] Based on the preset first detection probability and first clutter number, the tracking performance of the filter is compared with the average OSPA distance and the number of targets, and the real-time performance of the filter is compared with the running time.

[0139] Based on the preset detection probability, the number of clutter is gradually increased to determine how the average OSPA distance of different filters changes with the increase of the number of clutter.

[0140] Based on a preset clutter number, the detection probability is gradually reduced to determine how the average OSPA distance of different filters changes as the detection probability decreases.

[0141] Based on the preset second detection probability and second clutter number, the number of targets for different filters is compared, and the tracking results of each target are verified.

[0142] The implementation process of the present invention in a specific scenario will be described in detail below with reference to the accompanying drawings:

[0143] To address the problems existing in the prior art, the purpose of this invention is to provide a radar multi-target tracking method based on a feature-assisted UKF-PHD filter. This invention has the following characteristics:

[0144] 1. By leveraging the advantage of high Doppler resolution achieved through long-term accumulation of data by holographic staring radar, and combining this with Doppler information in radar echoes, the target can be distinguished from clutter and from other targets, thus improving tracking accuracy.

[0145] 2. The UKF-PHD filter was selected as the basic filter for improvement, which not only extended the GM-PHD filter to nonlinear scenarios, but also took into account the low computational cost of the GM-PHD filter.

[0146] 3. The effectiveness of the algorithm was verified by the measured data collected by holographic staring radar.

[0147] The specific operation steps of this invention are as follows:

[0148] Step 1: Modeling

[0149] Nonlinear state processes and measurement processes can be represented by the following nonlinear model:

[0150] x k =f k|k-1 (x k-1 q k-1 (1)

[0151] z k =h k (x k r k (2)

[0152] Where, x k and z k Let f represent the target state and measurement set at time k, respectively. k|k-1 and h k These are the nonlinear state transition function and the measurement function, respectively; Gaussian noise q k-1 and r k The process noise and measurement noise are zero-mean, respectively, and their corresponding covariance matrices are Q. k-1 and R k Since the state function and measurement function are nonlinear functions, the posterior intensity of the multi-objective state predicted and updated by the PHD filter is a weighted sum of different non-Gaussian functions. Using the UKF, each non-Gaussian function can be approximated using a single Gaussian function; specifically, this involves using the insensitive transform to calculate the Gaussian mixture approximation components of the posterior intensity.

[0153] Considering multi-target tracking in a two-dimensional plane, the motion equations of each target are set as a Coordinate Turn (CT) model. The motion equations of the targets are then:

[0154] x k =F(ω) k-1 )x k-1 +Gω k-1 (3)

[0155] Wherein, process noise ω k-1 The system consists of independent zero-mean Gaussian white noise, and the system state transition matrix is:

[0156]

[0157] The corresponding process noise distribution matrix is:

[0158]

[0159] The measurement equation for the target is:

[0160]

[0161] Among them, v k x is Gaussian white noise with zero mean. x，k and x y，k This represents the position and state of the target at time k.

[0162] Step 2: FAUKF-PHD Filter

[0163] Step 1: Assume the posterior strength v at time k-1 k-1 It takes the following Gaussian mixture form:

[0164]

[0165] The prediction step is based on v k-1 v is calculated using the mean, covariance, and weights. k|k-1 The process of expressing the mean, covariance, and weights. Let v be the predicted intensity at time k, which is also in Gaussian mixture form. k|k-1 As shown in the following formula:

[0166]

[0167] Among them, v γ,k (·) represents the intensity of the newly formed target at time k, v S,k|k-1 (·) represents the surviving target strength at time k. Based on the given Gaussian component set... Record the mean covariance Using insensitive transformation, a series of sigma points and their weights are generated based on the mean and covariance, denoted as . Here, N is twice the covariance dimension. In addition, there are several parameters that control the performance of the UKF filter. The 'a' parameter determines the confidence level of the state estimate in the UKF, and the 'β' parameter measures the uncertainty between the Gaussian distribution of the state and the Gaussian distribution of the measurement. The parameters affect the choice of sigma points. For n = 0, 1, ..., N, the partitioning... Then the predicted intensity v at time k k|k-1 The relevant variable expressions are shown below:

[0168]

[0169] Step 2: Update

[0170] Once a new measurement set Z is obtained k The update step provides v k|k-1 v is calculated using the mean, covariance, and weights. k The steps for expressing the mean, covariance, and weights. Let v be the posterior intensity at time k, which is also in Gaussian mixture form. k As shown in the following formula:

[0171]

[0172] Then the posterior strength v at time k k The relevant variable expressions are shown below:

[0173]

[0174] Among them, κ k Let be the intensity function of the clutter RFS of the Poisson distribution at time k.

[0175] Step 3: Feature Assistance

[0176] The feature-assisted UKF-PHD filter proposed in this invention utilizes high-precision Doppler information from holographic staring radar echoes to correct the weights of Gaussian components, increasing the weight of target components and decreasing the weight of clutter components. This achieves the distinction between targets and clutter, and between targets themselves, thereby improving the accuracy of radar multi-target tracking. Furthermore, after weight correction, more irrelevant Gaussian components can be filtered out during the pruning and merging steps, improving the efficiency of radar multi-target tracking.

[0177] Based on the expression for the Gaussian component weights in the update step, the variables affecting the weight magnitude are: This formula uses only the location information from the measurements, employing a Gaussian probability density function to describe the degree of matching between the location information in the measurement set and the location information contained in the predicted Gaussian components. To update the Gaussian component weights using feature information, a Gaussian probability density function is added to describe the degree of matching between the Doppler information in the measurement set and the Doppler information contained in the predicted Gaussian components. The expression representing the Doppler frequency feature likelihood is then shown below:

[0178]

[0179] Among them, z D This represents the measured Doppler frequency. In this invention, f... d，k Let k represent the Doppler frequency of the target at time k. This represents the overall error between the Doppler information in the measurement set and the Doppler information of the predicted Gaussian components. The greater the difference in Doppler frequencies between different targets, the better the tracking performance; the greater the difference in Doppler frequencies between the target and clutter, the better the tracking performance; and the smaller the change in the Doppler frequency of the same target between adjacent frames, the better the tracking performance. The expression after adding feature information is:

[0180]

[0181] Among them, z all This represents the measurement set containing both location and Doppler information. The weights of the Gaussian components are then updated using the corrected likelihood values; the corrected weight expression is as follows:

[0182]

[0183] Substitute the new weights into the posterior strength v k This completes the feature matching step. At this point, in addition to the positional information of all Gaussian components, the Gaussian components also incorporate the Doppler information contained in all Gaussian components. Therefore, the feature-aided step can be placed before the update step.

[0184] Step 4: Pruning and Merging

[0185] As processing time increases, the number of Gaussian mixture components in the FAUKF-PHD filter continuously increases. To effectively control the ever-increasing number of Gaussian components, Gaussian mixture pruning techniques such as pruning minor components and merging similar components are needed. A pruning threshold T is set to discard components with weights below a preset threshold; a maximum allowable number of Gaussian terms J is also set. max When the number of components exceeds this value, only the components with the largest weight are retained; set a merging threshold U to merge some sufficiently close Gaussian components.

[0186] Step 5: State Extraction

[0187] The FAUKF-PHD filter is easy to extract multi-objective state estimates based on the posterior intensity v. k The Gaussian mixture representation can first obtain the sum of weights to estimate the number of targets, and then select the component with the largest corresponding weight as the state estimate, or extract the mean of the Gaussian components with weights greater than a certain threshold as the state vector of all targets existing at time k.

[0188] Step 3: Presentation of the results of this invention

[0189] The simulation experiment considers multi-target tracking in a two-dimensional plane, with the monitoring area set as [-2000 2000] × [0 2000] (unit: m), and the number of targets is unknown and time-varying. To verify the effectiveness of the proposed FAUKF-PHD filter, different scenarios were designed, and the results were compared with those of standard GM-PHD, SMC-PHD, and UKF-PHD. Figure 3 It represents the target's actual trajectory, including common scenarios such as target creation, target disappearance, and trajectory intersection. Under the same target trajectory, 100 Monte Carlo experiments are run independently, comparing the performance of each filter. Figure 3 In this context, True track represents the actual trajectory; Track start represents the starting point; and Track end represents the ending point.

[0190] Scenario 1: When the detection probability p D,k =0.98, clutter number λ=30, the tracking performance of the filters is compared based on the average OSPA distance and the number of targets, and the real-time performance of the filters is compared based on the runtime. The comparisons of the number of targets, average OSPA distance, and runtime for different filters are as follows: Figure 4 , Figure 5 , Figure 6 As shown.

[0191] in, Figure 4 The horizontal and vertical axes represent time and the number of targets, respectively. Figure 5 The horizontal and vertical axes represent time and OSPA distance, respectively. Figure 6 The horizontal and vertical axes represent the number of Monte Carlo experiments and the running time, respectively.

[0192] Depend on Figure 4 , Figure 5 It is evident that under nonlinear motion and measurement models, the GM-PHD filter exhibits very poor tracking performance and cannot accurately estimate the number of targets. The proposed FAUKF-PHD filter, however, estimates the number of targets closest to the actual number of targets and demonstrates the greatest stability during tracking estimation, while also maintaining the smallest average OSPA distance. Figure 6It is known that the SMC-PHD filter has high computational cost, while the proposed algorithm's running time is close to that of the GM-PHD filter, demonstrating high computational efficiency. Therefore, in scenarios with low clutter rates and high detection probabilities, the FAUKF-PHD filter improves the tracking and potential estimation performance of radar multi-target detection without significantly increasing time costs.

[0193] Scenario 2: When the detection probability p D,k =0.98, when the clutter number λ continuously increases, the change of the average OSPA distance of different filters with the increase of the clutter number is as follows: Figure 7 As shown. Figure 7 The x and y axes represent clutter number and OSPA distance, respectively; Figure 7 It is known that the average OSPA distance of each filter increases with the increase of clutter number, while the average OSPA distance of the FAUKF-PHD filter remains almost unchanged, showing less impact from the increase in clutter number, and exhibiting the best tracking performance. Therefore, the FAUKF-PHD filter can adapt to radar multi-target tracking in dense clutter environments.

[0194] Scenario 3: When the clutter number λ = 30, the detection probability p D,k As the detection probability decreases continuously, the change in average OSPA distance for different filters is as follows: Figure 8 As shown. Figure 8 The x and y coordinates represent the detection range and OSPA distance; from Figure 8 It can be seen that as the detection probability decreases, the average OSPA distance of each filter increases, but the FAUKF-PHD filter has the lowest average OSPA distance. Therefore, the FAUKF-PHD filter can adapt to radar multi-target tracking in low detection probability environments.

[0195] Scenario 4: When the detection probability p D,k =0.85, clutter number λ=150, the target number comparison of different filters is as follows: Figure 9 As shown. Figure 9 The horizontal and vertical axes represent time and the number of targets, respectively; from Figure 9 As can be seen, when the detection probability is low and the clutter quantity is high, the number of targets estimated by the FAUKF-PHD filter is close to the actual number of targets, while the standard UKF-PHD filter shows obvious missed detections. The potential estimation performance of the improved filter is better than that of the standard UKF-PHD filter.

[0196] Real-world testing scenario: The proposed method was validated using multi-target data acquired by a holographic staring radar system. The radar was installed at a high point along the coast to track two UAVs flying in an eccentric circle over the sea. The acquired real-time data was processed using a FAUKF-PHD filter, and the results are as follows. Figure 10 As shown. Figure 10 The x and y axes represent angle and radial distance, respectively; where Measurements in the legend represent measured values ​​and Estimates represent estimated values. It can be seen that despite the presence of ocean clutter, the filtering results are still uniformly distributed around the target measurement point, demonstrating the effectiveness of the proposed algorithm.

[0197] In summary, this invention employs a feature-assisted method to optimize the UKF-PHD filter, which is suitable for nonlinear scenarios and has low computational cost, to achieve multi-target tracking in clutter-dense scenarios.

[0198] Compared with the prior art, the present invention has the following characteristics:

[0199] 1. Improved UKF-PHD Filter: Combining the high Doppler resolution of holographic staring radar, a UKF-PHD filter based on Doppler feature assistance is proposed to improve tracking accuracy in dense clutter scenarios.

[0200] 2. Verification using measured data: The algorithm is used to process the nonlinear multi-target measured data collected by the holographic staring radar to verify that the algorithm balances tracking efficiency and tracking accuracy.

[0201] Compared with the prior art, the present invention has the following advantages:

[0202] Compared to optimizing the GM-PHD filter, this invention selects the UKF-PHD filter for improvement. While retaining the low computational complexity of the GM-PHD filter, it extends it to nonlinear scenarios, which is more in line with engineering reality. The Gaussian component weights are corrected by using high-precision Doppler information in the holographic staring radar echo, and the accuracy of the weights can be improved by using only one feature information. The effectiveness of the algorithm in multi-target tracking under dense clutter and low detection probability scenarios is verified by measured data.

[0203] Another aspect of this invention provides a multi-target tracking system for nonlinear scenes, comprising:

[0204] The first module is used to construct nonlinear motion and measurement models, specifically including: representing nonlinear state and measurement processes through nonlinear models; determining the motion equations of each target based on multi-target tracking in a two-dimensional plane, setting up a cooperative turning model, and determining the motion equations of the targets; constructing the system state transition matrix, process noise distribution matrix, and measurement equations.

[0205] The second module is used to extract the state estimation of multiple targets through the FAUKF-PHD filter based on the nonlinear motion model and measurement model, so as to obtain the state vector of each target and complete the tracking of each target.

[0206] The third module is used to verify the tracking results of each target based on multiple preset scenario information.

[0207] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0208] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned multi-target tracking method for nonlinear scenes. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0209] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0210] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described nonlinear scene multi-target tracking method.

[0211] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0212] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0213] It should be noted that in various specific embodiments of the present invention, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of the present invention require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to a confirmation page. Only after obtaining the user's separate permission or consent is the necessary user-related data for the normal operation of the embodiments of the present invention acquired.

[0214] The embodiments described in this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0215] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present invention, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0216] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0217] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0218] The terms "first," "second," "third," "fourth," etc. (if present) in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0219] It should be understood that in this invention, "at least one (item)" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0220] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0221] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0222] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0223] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0224] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present invention should be within the scope of the claims of the present invention.

Claims

1. A multi-target tracking method for nonlinear scenes, characterized in that, Includes the following steps: The nonlinear motion model and measurement model are constructed, specifically including: representing the nonlinear state process and measurement process through the nonlinear model; determining the motion equation of each target based on multi-target tracking in a two-dimensional plane and setting it as a cooperative turning model to determine the motion equation of the target; and constructing the system state transition matrix, process noise distribution matrix, and measurement equation. Based on the nonlinear motion model and measurement model, the state estimation of multiple targets is extracted by using a feature-assisted UKF-PHD filter to obtain the state vector of each target, thereby completing the tracking of each target; Based on multiple preset scenario information, the tracking results of each target are verified; The step of extracting the state estimation of multiple targets based on the nonlinear motion model and measurement model using a feature-assisted UKF-PHD filter to obtain the state vector of each target includes the following steps: Construct a Gaussian mixture of the posterior intensities at time k-1 to obtain a new measurement set; Based on the new measurement set, the mean, covariance, and weight of the posterior intensity are calculated using the mean, covariance, and weight of the predicted intensity. Based on the feature-assisted UKF-PHD filter, the Gaussian component weights are corrected using high-precision Doppler information in the holographic staring radar echo. The minor components are pruned according to the Gaussian mixed pruning technique, and similar components are merged. Based on the final Gaussian mixture representation, extract the state vectors of all targets; In the step of constructing the Gaussian mixture form of the posterior intensity at time k-1 to obtain the new measurement set, the expression of the Gaussian mixture form is: Among them, v k-1 (x) represents the posterior strength at time k-1; J k-1 Represents the total number of given new target components; The weights, mean, and covariance of the i-th mixture component representing the intensity of newborn cells are respectively. This represents a Gaussian density with mean m and covariance P; The expression for the process of calculating the mean, covariance, and weight of the posterior intensity based on the new measurement set and the predicted intensity's mean, covariance, and weight is as follows: Among them, v k (x) represents the posterior strength at time k; p D,k Z represents the probability of object detection; k This represents information about a measurement point at time k; The step of correcting the Gaussian component weights using the high-precision Doppler information in the holographic staring radar echo based on the feature-assisted UKF-PHD filter includes the following steps: Using the location information in the measurement set, a Gaussian probability density function is used to describe the degree of matching between the location information in the measurement set and the location information contained in the predicted Gaussian component; Update the weights of the Gaussian components with the corrected likelihood values; Substitute the new weights into the posterior strength to complete the feature matching step; The step of extracting the state vectors of all targets based on the final Gaussian mixture representation includes the following steps: The number of targets is estimated by obtaining the sum of weights based on the Gaussian mixture expression of the posterior intensity. Select the component with the largest corresponding weight as the state estimate, or extract the mean of the Gaussian components with weights greater than a preset threshold as the state vector of all targets existing at time k.

2. The nonlinear scene multi-target tracking method according to claim 1, characterized in that, The construction of the nonlinear motion model and measurement model includes the following steps: Nonlinear state processes are represented by nonlinear models. Considering multi-target tracking in a two-dimensional plane, the motion equations of each target are set as a cooperative turning model.

3. The nonlinear scene multi-target tracking method according to claim 2, characterized in that, The expression for the nonlinear state process is: x k =f k|k-1 (x k-1 ,q k-1 ) z k =h k (x k ,r k ) Where, x k and z k Let f represent the target state and measurement set at time k, respectively; k|k-1 and h k These are the nonlinear state transition function and the measurement function, respectively; Gaussian noise q k-1 and r k The process noise and measurement noise are zero-mean, respectively, and their corresponding covariance matrices are Q. k-1 and R k ; The equation of motion for the target is expressed as follows: x k =F(ω k-1 )x k-1 +Gω k-1 Where F represents the system state transition matrix; G represents the process noise distribution matrix; ω k-1 This represents process noise.

4. The nonlinear scene multi-target tracking method according to claim 3, characterized in that, The expression for the system state transition matrix F is: Where T represents the sampling interval; The expression for the process noise distribution matrix G is: The expression for the process noise is: Among them, v k x is Gaussian white noise with zero mean. x,k and x y,k This represents the position and state of the target at time k.

5. The nonlinear scene multi-target tracking method according to claim 1, characterized in that, The verification of tracking results for each target based on multiple preset scene information includes the following steps: Based on the preset first detection probability and first clutter number, the tracking performance of the filter is compared with the average OSPA distance and the number of targets, and the real-time performance of the filter is compared with the running time. Based on the preset detection probability, the number of clutter is gradually increased to determine how the average OSPA distance of different filters changes with the increase of the number of clutter. Based on a preset clutter number, the detection probability is gradually reduced to determine how the average OSPA distance of different filters changes as the detection probability decreases. Based on the preset second detection probability and second clutter number, the number of targets for different filters is compared, and the tracking results of each target are verified.

6. A system for implementing the nonlinear scene multi-target tracking method as described in any one of claims 1-5, characterized in that, include: The first module is used to construct nonlinear motion and measurement models, specifically including: representing nonlinear state and measurement processes through nonlinear models; determining the motion equations of each target based on multi-target tracking in a two-dimensional plane, setting up a cooperative turning model, and determining the motion equations of the targets; constructing the system state transition matrix, process noise distribution matrix, and measurement equations. The second module is used to extract the state estimation of multiple targets by using a feature-assisted UKF-PHD filter based on the nonlinear motion model and measurement model, so as to obtain the state vector of each target and complete the tracking of each target. The third module is used to verify the tracking results of each target based on multiple preset scenario information.

7. An electronic device, characterized in that, Including the processor and memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 5.

Images