Target tracking distributed control method for time sequence space fusion optimization

Abstract

The invention relates to the technical field of aircraft perception, in particular to a target tracking distributed control method for time sequence space fusion optimization. In a pixel coordinate system, time sequence tracking is carried out on multiple targets by extracting time sequence apparent characteristics, and the time sequence state of the multiple targets is obtained; obtaining a multi-target spatial topological structure based on a prediction principle; and under the pixel coordinate system, performing fusion optimization on the time sequence state and the space structure of the multiple targets to obtain an optimized tracking result. Aiming at the inherent error problem existing in multi-target tracking in an aircraft visual perception system, the time sequence state and the space structure are fused and optimized, so that the target state estimation precision is improved, and effective perception track information is formed.

Description

Technical Field

[0001] The present invention relates to the field of aircraft perception technology, and in particular to a target tracking distributed control method with time-space fusion optimization. Background Art

[0002] With the rapid development of unmanned aerial vehicle (UAV) technology, aircraft swarms (i.e., multi-UAV systems) have been widely used in military reconnaissance, disaster monitoring, environmental protection, and logistics transportation. Swarm aircraft systems can improve mission execution efficiency and system flexibility through collaborative operations. However, in practical applications, multi-target tracking has become a key technical challenge, especially in complex and dynamic environments.

[0003] Traditional target tracking methods usually rely on centralized perception and processing, which requires all sensor data to be concentrated in a central node for analysis and decision-making. Although this centralized method can provide higher tracking accuracy, it also faces problems such as data transmission delay, excessive computing load and single point failure of the system. In order to deal with these problems, researchers have gradually turned to distributed perception methods in recent years, using each drone in the aircraft cluster to perform local data processing and target tracking, thereby achieving higher system robustness and flexibility.

[0004] Distributed perception methods usually rely on each drone to independently acquire and process information about the surrounding environment, and then share the processing results through a communication network. Although this method can effectively reduce the burden on individual nodes, reduce data transmission delays, and improve system reliability, it also faces some new challenges. For example, how to ensure that the information between drones can be effectively integrated in a dynamic and complex environment, and how to ensure tracking accuracy and real-time performance in a multi-target and high-density target environment are all issues that need to be solved urgently.

[0005] In distributed perception systems, temporal-spatial fusion is a key technology that involves how to reasonably integrate the observation data of multiple drones in time and space to obtain accurate target status information. Traditional temporal-spatial fusion methods usually ignore the timeliness issues and optimization of spatial distribution characteristics in the data fusion process, resulting in a significant decrease in the tracking performance of the system when facing fast-moving targets or complex changes in the environment.

[0006] In order to achieve efficient time-space fusion under limited computing power, it is necessary to strike a balance between complex mathematical models and a large amount of computing resources. Summary of the invention

[0007] The present invention provides a distributed control method for target tracking with time-space fusion optimization. Aiming at the inherent error problem of multiple object tracking (MOT) in the aircraft visual perception system, the accuracy of target state estimation is improved by fusing and optimizing the time state and space structure, thereby forming effective perception trajectory information.

[0008] The present invention is achieved through the following technical solutions:

[0009] A target tracking distributed control method with time-series spatial fusion optimization, the method comprising the following steps:

[0010] Step S1: Acquire the tracking data of all targets, generate multi-target motion trajectories according to the recognition image, and perform time-series tracking on multiple targets in the pixel coordinate system by extracting time-series appearance features and motion features to obtain the time-series states of multiple targets, where the target is an unmanned aerial vehicle;

[0011] Step S2: obtaining the spatial topological structure of multiple targets based on the prediction principle;

[0012] Step S3: In the pixel coordinate system, the temporal state of the multiple targets in step 1 and the spatial topological structure in step S2 are fused and optimized to obtain an optimized tracking result.

[0013] Furthermore, the step S1 is specifically as follows: initialize a pre-trained neural network to track the target, obtain the time series tracking result of the target, establish a time series tracker to realize the time series tracking of multiple targets, extract the time series appearance features of the target, select LSTM to realize the long-term memory of the time series appearance features of the target, and obtain the motion state of the tracked target through the collected deep features.

[0014] Furthermore, the step S2 is specifically as follows: extracting time series appearance features and time series motion features, selecting the Prime minimum spanning tree algorithm to meet the principle of minimum computational effort, using the historical trajectory of the sliding filter to predict the motion speed of the target, and forming a spatial topological structure of multiple targets.

[0015] Furthermore, the step S3 specifically includes: using an extended Kalman filter to fuse and optimize the temporal states and spatial structures of multiple targets to obtain an optimized tracking result.

[0016] Furthermore, the step S1 specifically includes the following steps:

[0017] Step S1.1: Initialize a pre-trained AeLsNet network to track all targets;

[0018] Step S1.2: In the time series tracking model, directly use the convolutional layer parameters of the AlexNet network pre-trained in step 1.1 to extract the highlighted depth features;

[0019] Step S1.3: Based on the deep features extracted in step S1.2, the deep features are concatenated into a one-dimensional vector and then passed through a multi-layer perceptron to obtain the temporal motion features.

[0020] Furthermore, step S1 also includes that the long-term memory of the target time series appearance features is realized by the mutual cooperation of the forget gate, input gate, and output gate of the LSTM.

[0021] Furthermore, the step 2 specifically includes the following steps:

[0022] Step S2.1: Generate an intra-frame tree structure topology according to the tracking result at time t;

[0023] Step S2.2: predict the motion state of each target;

[0024] Step S2.3: Predict the position of the target at a certain moment according to the motion speed predicted in step S2.2, and update the correlation function in the intra-frame tree structure topology of step 2.1 according to the predicted position to obtain the predicted spatial topology structure.

[0025] Furthermore, the step S3 specifically includes the following steps:

[0026] Step S3.1: root node optimization process in the spatial topology structure;

[0027] Step S3.2: The remaining targets optimize themselves using the position constraint relationship with the root node in the spatial topological structure;

[0028] Step S3.3: EKF linearizes the state equation and observation equation using Taylor expansion to enable it to be applied to nonlinear systems;

[0029] Step S3.4: Fusion optimization is performed on other targets, assuming that they are directly connected to the root node in the spatial topological structure, that is, other targets can use the relative position relationship of the root node to improve their own accuracy.

[0030] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the above method is implemented.

[0031] A computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0032] The beneficial effects of the present invention are:

[0033] The present invention improves the accuracy of target state estimation and forms effective perception trajectory information.

[0034] The present invention improves the data fusion mechanism to enhance the accuracy and real-time performance of multi-target tracking, so as to meet the application requirements of aircraft clusters in various complex environments.

[0035] The present invention can effectively improve the overall performance of the aircraft cluster system in the target tracking task.

[0036] The present invention can meet the future intelligent and networked combat and monitoring requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a simplified flow chart of the method of the present invention.

[0038] Figure 2 It is a flow chart of the method of the present invention.

[0039] Figure 3 It is a schematic diagram of the overall process of multi-target tracking of the present invention. DETAILED DESCRIPTION

[0040] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0041] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0042] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0043] The following is attached to this application specification Figure 1-2, the technical solutions in the embodiments of the present application are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0044] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0045] Implementation Method 1

[0046] This embodiment provides a target tracking distributed control method with time-series and space-time fusion optimization, and the method is specifically as follows:

[0047] In order to achieve time-series tracking of multiple targets, a pre-trained AeLsNet network is initialized in the pixel coordinate system to track all targets. Represents the parameter matrix of AeLsNet, represents the tracking result of the jth (total number is J) target at time t-1. Then, the time series tracking result of this target at time t can be expressed as

[0048]

[0049] Among them, |·| 1×l Represents 1 row and 1 column; This is the timing tracking result of 1 row and 5 columns The first four dimensions are the upper left corner horizontal axis coordinate, the upper left corner vertical axis coordinate, the target width, and the target height in the pixel coordinate system. The fifth dimension is the confidence level. Indicates that the result is the result of time series tracking. All J tracking targets in the image at time t-1 can be written as Denoted as matrix S t-1 The time series tracking results of these J targets are expressed as

[0050]

[0051] in, It includes all tracked targets at time t, that is, only one time series tracker is needed to realize the time series tracking of multiple targets.

[0052] In the time series tracking model (AeLsNet), the convolutional layer parameters of the pre-trained AlexNet network are directly used to extract the prominent deep features. tRepresents all the temporal appearance features that appear in the process of tracking the target, which can be expressed as

[0053]

[0054] in, is the temporal appearance characteristic of the tracked target j, is the apparent information of target j at time t (the deep features after the AlexNet convolution layer), and f is the mapping of the apparent information.

[0055] LSTM is used to achieve long-term memory of the target time series appearance features. The long-term memory of LSTM is achieved by the cooperation of the forget gate, input gate, and output gate. The weight coefficients and biases of the three gates are respectively represented by W f , W i , W o and b f , b i , b o The forget gate controls the cell state at time t-1. How many cell states are retained at time t? The input gate controls how much of the input at time t is retained in the cell state at time t The output gate controls the cell state at time t How many inputs at time t are output to the LSTM output at time t These three gates can be represented as

[0056]

[0057] in, are the direct output values ​​of the forget gate, input gate, and output gate, respectively. σ represents the sigmoid activation function. The cell state of LSTM is composed of the cell state at time t-1 and the features of the input information extracted at time t It consists of two parts and can be expressed as

[0058]

[0059] Among them, W c and b c Represent the weight matrix and bias of the computational unit state, is the output value of LSTM at time t-1. Finally, the output value of LSTM at time t By the output gate at time t and cell status Joint decision can be expressed as

[0060]

[0061] Among them, tanh represents the activation function.

[0062] Specifically, taking the jth tracked target as an example, the forget gate judges the temporal appearance characteristics from time 1 to time t-1 Does it match the apparent characteristics at time t? The output gate determines the characteristics of the input information at time t Whether to add to the extracted time series appearance features The output gate integrates the forget gate and the input gate to the apparent features Determine the tracking result at time t

[0063] LSTM-based temporal target tracking can memorize the irregular motion, morphological changes and other states of the target that have appeared in the entire sequence. When a similar state appears again, it directly awakens the long-term memory to maintain stable tracking, avoiding the defect of the traditional method of updating the template in real time and only memorizing the current state.

[0064] The deep features of two adjacent frames are obtained through AlexNet, and the deep features are connected into a 1-dimensional vector and then passed through a multi-layer perceptron to obtain the temporal motion feature m. t . Temporal motion characteristics m t It describes the motion state of the tracked target from time t-1 to time t, which can be expressed as

[0065]

[0066] in, is the motion feature of the tracked target i, and g is the mapping of the motion features of the target state value between two adjacent frames.

[0067] In order to obtain the spatial topological structure of multiple targets through prediction, the tree structure topology within the frame is generated according to the tracking results at time t. This process takes the principle of minimizing the amount of calculation as the principle, and the Prime minimum spanning tree method is selected. Among them, when there are N tracked targets at time t, the correlation function φ t The calculation method is: t The elements of

[0068] Predict the motion state of each target. Different from the temporal motion feature that focuses on the abstract motion properties of the target at time t, this motion state is based on the trajectory τ t As the carrier, the specific movement speed of target i at time t+1 is directly predicted using the historical trajectory based on sliding filtering.

[0069]

[0070] in, is a two-dimensional vector, representing the coordinates of target i at time t in the pixel coordinate system. In the pixel coordinate system, assume that the trajectory of target i at time t is The predicted value of the moving speed of target i at time t+1 by sliding filtering It can be expressed as

[0071]

[0072] Among them, l filter is the frame usage parameter of the sliding filter, the coefficient ζ is the weight coefficient, and Respectively size and direction.

[0073] According to the predicted movement speed Predict the position of the target at time t+1 and convert G t The correlation function in is updated according to the predicted position to obtain the predicted spatial topological structure

[0074] The spatial topological structure formed The edge E t+1 and vertex V t+1 The above three steps will be executed again to regenerate the spatial topology structure. When the number of targets increases, the newly appeared targets will be directly connected to the nearest targets, incrementally expanding the tree structure.

[0075] Finally, the temporal state and spatial structure of multiple targets are fused and optimized to obtain the optimized tracking result.

[0076] Based on the Extended Kalman Filter (EKF) and timing trace results Perform fusion optimization. The root node's observation value is its time series tracking result. The fusion optimization process of the root node is the standard EKF process. Applying the Taylor expansion equation, the final tracking result of the root node can be obtained. Remaining target exploits The position constraint relationship with the root node optimizes itself.

[0077] Based on the Extended Kalman Filter (EKF) and timing trace results Perform fusion optimization. High confidence tracking results have high accuracy, and the remaining targets use Spatial position constraints for medium and high confidence targets can further improve their accuracy.

[0078] The spatial-temporal fusion optimization can be divided into two steps. The first is the spatial topology structure. The root node (highest confidence) optimization process, followed by the remaining targets using The position constraint relationship with the root node optimizes itself.

[0079] Next, the EKF linearizes the state equation and observation equation using Taylor expansion to make it applicable to nonlinear systems, expressed as

[0080]

[0081] P t|t-1 =F t|t-1 P t-1|t-1 (F t|t-1 ) T +Q t-1 ,

[0082]

[0083] P t|t ←(IG t H t ) t|t-1 ,

[0084] in, - , + They represent estimated value, non-optimal value, and optimal value respectively. f represents the nonlinearization of the state equation, h represents the nonlinearity of the observation equation, t represents the time step, x is the input state information, z t is the observed value, y t is the difference between the observation and the estimate, is the covariance matrix, P is the covariance matrix of the state error, F is the Jacobian matrix of the state equation, Q is the covariance matrix of the process noise, G is the Kalman gain, J is the Jacobian matrix of the observation equation, is the Jacobian matrix of the observation noise and I is the identity matrix.

[0085] For target i at time k, its motion equation (predicted position) can be expressed as

[0086]

[0087] Where ΔT is the time interval between adjacent time steps. When the root node (assuming target k) is fused and optimized based on EKF, its motion equation is:

[0088]

[0089] Among them, the matrix describes the motion of target i, is the input noise matrix of target j at time step t-1, σ propagation is the system noise, σ propagation ~N(0,M), we can get its covariance matrix for

[0090]

[0091] The root node's observation value is its time series tracking result The fusion optimization process of the root node is the standard EKF process. Applying the Taylor expansion equation, the final tracking result of the root node can be obtained.

[0092] Furthermore, other targets (target i) are optimized by fusion, and it is assumed that they are in the spatial topological structure with the root node (target j). , target i can use the relative position relationship of the root node to improve its own accuracy. This relative position is used as the observation of target i, that is, the observation matrix of target i relative to target j at time t

[0093]

[0094] in, is the optimal state estimate of the root node, that is, the final tracking result of target j at time t r~N(0,M) is the measurement error, and

[0095]

[0096] in, is the covariance matrix of the measurement error. The linearized observation equation can be expressed as

[0097]

[0098] in,

[0099]

[0100]

[0101] in, is the covariance matrix of the state error, and the Kalman gain is

[0102]

[0103] Then, the difference between the observed value and the estimated value is calculated, expressed as

[0104]

[0105] Then the location is updated, which is expressed as

[0106]

[0107] Finally, the update formula of the covariance matrix of the state error is

[0108]

[0109] Covariance matrix of target i The update formula is

[0110]

[0111] The covariance matrix of target j is The update matrix is

[0112]

[0113] After target j uses the relative position information with target i, it generates and The two coupling terms are expressed as

[0114]

[0115] This embodiment applies the proposed STFT method to a multi-aircraft scenario. Specifically, when performing multi-target tracking at time t, the tracking results S of all the targets of interest at time t-1 have been obtained. t-1 、Time series appearance characteristics C t-1 , motion model v of all targets t-1 , the spatial topological structure G at time t-1 t-1 , the process is: according to the motion model v t-1 Predict the possible arrival area of ​​each aircraft and obtain the candidate area of ​​the target of interest; use the tracking result at time t-1 and the target candidate area at time t as input to obtain the motion characteristics of the target The time series apparent characteristic C t-1 and the target candidate region at time t as input to obtain the temporal appearance feature C from the initial time to time t t ; Comprehensive sports characteristics m t and the temporal appearance characteristic C t , get the time series estimation results of all targets at time t According to the spatial topological structure G of all targets at time t-1 t-1 Its motion model t-1 Reasoning about the spatial topology at time t

[0116] Implementation Method 2

[0117] This embodiment provides a distributed perception system for target tracking with time-series and space-time fusion optimization. The system uses a distributed control method for target tracking with time-series and space-time fusion optimization as described in Embodiment 1. The system includes:

[0118] Time sequence state module: obtains the tracking data of all targets, generates multi-target motion trajectories based on the recognition images, and performs time sequence tracking on multiple targets by extracting time sequence appearance features and motion features to obtain the time sequence states of multiple targets; the targets are unmanned aerial vehicles;

[0119] Spatial topology structure module: obtains the spatial topology structure of multiple targets based on prediction principles;

[0120] Fusion optimization module: Fusion optimizes the multi-objective temporal states of the temporal state module and the spatial topological structure of the spatial topological structure module to obtain optimized tracking results.

[0121] Implementation Method 3

[0122] An embodiment of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory is used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory and the processor are connected via a bus. Specifically, the processor implements any step in the first embodiment above by running the computer program stored in the memory.

[0123] It should be understood that in the embodiments of the present invention, the processor referred to may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0124] The memory may include a read-only memory, a flash memory, and a random access memory, and provides instructions and data to the processor. A part or all of the memory may also include a nonvolatile random access memory.

[0125] It should be understood that if the above-mentioned integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned implementation method, and can also be completed by instructing the relevant hardware through a computer program. The above-mentioned computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method implementations when executed by the processor. Among them, the above-mentioned computer program includes computer program code, and the above-mentioned computer program code can be in source code form, object code form, executable file or some intermediate form. The above-mentioned computer-readable medium may include: any entity or device capable of carrying the above-mentioned computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the above-mentioned computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0126] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest range consistent with the principles and novel features disclosed herein.

[0127] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the above-mentioned device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the implementation method can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method implementation method, which will not be repeated here.

[0128] It should be noted that the methods and detailed examples provided in the above embodiments can be combined with the devices and equipment provided in the embodiments, and references can be made to each other, and no further details will be given.

[0129] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0130] In the embodiments provided by the present invention, it should be understood that the disclosed device / terminal equipment and method can be implemented in other ways. For example, the device / equipment implementation described above is only illustrative, for example, the division of the above modules or units is only a logical function division, and in actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0131] The above-mentioned embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above-mentioned embodiments, or replace some of the technical features therein by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A target tracking distributed control method with time-space fusion optimization, characterized in that: The method comprises the following steps: Step S1: Acquire the tracking data of all targets, generate multi-target motion trajectories according to the recognition image, and perform time-series tracking on multiple targets in the pixel coordinate system by extracting time-series appearance features and motion features to obtain the time-series states of multiple targets, where the target is an unmanned aerial vehicle; Step S2: obtaining the spatial topological structure of multiple targets based on the prediction principle; Step S3: In the pixel coordinate system, the temporal state of the multiple targets in step 1 and the spatial topological structure in step S2 are fused and optimized to obtain an optimized tracking result, thereby realizing a distributed control method for target tracking with temporal-spatial fusion optimization.

2. The target tracking distributed control method according to claim 1, characterized in that: The step S1 is specifically as follows: initializing a pre-trained neural network to track the target, obtaining the target's time series tracking result, establishing a time series tracker to achieve time series tracking of multiple targets, extracting the target's time series appearance features, selecting LSTM to achieve long-term memory of the target's time series appearance features, and obtaining the motion state of the tracked target through the collected deep features.

3. The target tracking distributed control method according to claim 2, characterized in that: The step S2 specifically includes: extracting time series appearance features and time series motion features, selecting the Prime minimum spanning tree algorithm to meet the principle of minimum computational complexity, using the historical trajectory of the sliding filter to predict the motion speed of the target, and forming a spatial topological structure of multiple targets.

4. The target tracking distributed control method according to claim 1, characterized in that: The step S3 specifically includes: using extended Kalman filtering to fuse and optimize the temporal states and spatial structures of multiple targets to obtain an optimized tracking result.

5. The target tracking distributed control method according to claim 4, characterized in that: The step S1 specifically includes the following steps: Step S1.1: Initialize a pre-trained AeLsNet network to track all targets; Step S1.2: In the time series tracking model, directly use the convolutional layer parameters of the AlexNet network pre-trained in step S1.1 to extract the highlighted depth features; Step S1.3: Based on the deep features extracted in step S1.2, the deep features are concatenated into a one-dimensional vector and then passed through a multi-layer perceptron to obtain the temporal motion features.

6. The target tracking distributed control method according to claim 5, characterized in that: Step S1 also includes that the long-term memory of the target time series appearance features is realized by the mutual cooperation of the forget gate, input gate, and output gate of the LSTM.

7. The target tracking distributed control method according to claim 1, characterized in that: The step 2 specifically includes the following steps: Step S2.1: Generate an intra-frame tree structure topology according to the tracking result at time t; Step S2.2: predict the motion state of each target; Step S2.3: Predict the position of the target at a certain moment according to the motion speed predicted in step S2.2, and update the correlation function in the intra-frame tree structure topology of step 2.1 according to the predicted position to obtain the predicted spatial topology structure.

8. The target tracking distributed control method according to claim 7, characterized in that: The step S3 specifically comprises the following steps: Step S3.1: root node optimization process in the spatial topology structure; Step S3.2: The remaining targets optimize themselves using the position constraint relationship with the root node in the spatial topological structure; Step S3.3: EKF linearizes the state equation and observation equation using Taylor expansion to enable it to be applied to nonlinear systems; Step S3.4: Fusion optimization is performed on other targets, assuming that they are directly connected to the root node in the spatial topological structure, that is, other targets can use the relative position relationship of the root node to improve their own accuracy.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images