Target tracking method, device, equipment and medium

Abstract

The embodiment of the invention provides a target tracking method and device, equipment and a medium, which are used for solving the problem that target tracking cannot be accurately and effectively realized in related technologies. According to the embodiment of the invention, the electronic equipment constructs the state vectors of the defender agent and the intruder agent based on the collected image; according to the target state vector of the defender agent at the preset number of moments, the first prediction state vector of the intruder agent at the preset number of moments and a first relation between the distance between the defender agent and the intruder agent, target control input at the preset number of moments when the distance is minimum is determined; and the defender agent is controlled based on the target control input of the preset number of moments, so that the defender agent can quickly and effectively approach the intruder agent, and target tracking is accurately and effectively realized.

Description

Technical Field

[0001] This application relates to the fields of multi-agent systems, game theory, and security defense, and particularly relates to a target tracking method, device, equipment, and medium. Background Technique

[0002] In the complex application environment of modern multi-agent systems, the importance of target tracking tasks has become increasingly prominent, especially in the dynamic confrontation scenarios between defender agents and intruder agents. Defender agents need to cooperate to ensure system security, while intruder agents attempt to avoid tracking. This game relationship poses extremely high requirements for the decision-making capabilities of agents. In a rapidly changing environment, defender agents must make precise decisions in real time, quickly approach and continuously track intruder agents, creating favorable conditions for subsequent expulsion or control operations. This process requires agents to have a high degree of autonomous decision-making ability and environmental adaptability. The core challenge lies in how to achieve complex strategy planning and real-time decision-making in a dynamic and uncertain environment.

[0003] In related technologies, non-cooperative target tracking methods mainly rely on predetermined search patterns. Although this centralized control method can achieve basic tracking functions in specific scenarios, its inherent limitations become increasingly apparent when faced with complex and changing actual environments. Specifically, it shows insufficient flexibility and difficulty in adapting to dynamic environmental changes; limited scalability and inability to effectively meet the tracking requirements of multiple targets and multiple scenarios; and limited intelligence, making it difficult to achieve accurate target state estimation and behavior prediction. These limitations severely restrict the overall performance and practical application effects of the target tracking system. Summary of the Invention

[0004] Embodiments of this application provide a target tracking method, device, equipment, and medium to solve the problem in related technologies that target tracking cannot be accurately and effectively achieved.

[0005] In a first aspect, embodiments of this application provide a target tracking method, and the method includes:

[0006] Identify the position information and attitude information of the defender agent and the intruder agent at the current moment in the image collected at the current moment through a pre-trained model; for each agent, generate a state vector of the agent at the current moment based on the position information and attitude information of the agent;

[0007] For a preset number of moments after the current moment, determine the target state vector of the defender agent at this moment according to the state vector of the defender agent at the current moment and the control input to be predicted by the defender agent at this moment, where the control input includes the instantaneous pitch angular velocity, instantaneous horizontal angular velocity, and instantaneous acceleration of the defender agent;

[0008] Determine the target control input at the preset number of moments when the distance is minimized according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent; control the defender agent based on the target control input at the preset number of moments.

[0009] In a second aspect, an embodiment of the present application further provides a target tracking device, and the device includes:

[0010] A determination and generation module, configured to identify the position information and attitude information of the defender agent and the intruder agent in the image collected at the current moment through a pre-trained model; for each agent, generate a state vector of the agent at the current moment based on the position information and attitude information of the agent; for a preset number of moments after the current moment, determine the target state vector of the defender agent at this moment according to the state vector of the defender agent at the current moment and the control input to be predicted by the defender agent at this moment, where the control input includes the instantaneous pitch angular velocity, the instantaneous horizontal angular velocity, and the instantaneous acceleration of the defender agent;

[0011] A processing module, configured to determine the target control input at the preset number of moments when the distance is minimized according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent; control the defender agent based on the target control input at the preset number of moments.

[0012] In a third aspect, an embodiment of the present application further provides an electronic device, and the electronic device at least includes a processor and a memory. When the processor executes a computer program stored in the memory, the steps of the target tracking method as described in any one of the above are implemented.

[0013] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the target tracking method as described in any one of the above are implemented.

[0014] In the embodiments of the present application, an electronic device constructs state vectors of a defender agent and an intruder agent based on the acquired images; and determines the target control inputs at a preset number of moments when the distance is minimized according to the target state vectors of the defender agent at a preset number of moments, the first predicted state vectors of the intruder agent at a preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent; and controls the defender agent based on the target control inputs at a preset number of moments, so that the defender agent can quickly and effectively approach the intruder agent and accurately and effectively achieve target tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for the description of the embodiments. Obviously, the following described drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0016] Figure 1 FIG.

[0017] Figure 2 FIG.

[0018] Figure 3 FIG.

[0019] Figure 4 FIG.

[0020] Figure 5 FIG.

[0021] Figure 6 FIG. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following will further describe the present invention in detail with reference to the drawings. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0023] To accurately and effectively perform target tracking, the embodiments of the present application provide a target tracking method, device, equipment, and medium.

[0024] The target tracking method includes: identifying in the image collected at the current moment through a pre-trained model to determine the position information and attitude information of the defender agent and the intruder agent at the current moment; for each agent, generating a state vector of the agent at the current moment based on the position information and attitude information of the agent; for a preset number of moments after the current moment, determining the target state vector of the defender agent at that moment according to the state vector of the defender agent at the current moment and the control input to be predicted for the defender agent at that moment, where the control input includes the instantaneous pitch angular velocity, the instantaneous horizontal angular velocity, and the instantaneous acceleration of the defender agent; determining the target control input at the preset number of moments when the distance is minimized according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship of the distance between the defender agent and the intruder agent; and controlling the defender agent based on the target control input at the preset number of moments.

[0025] Embodiment 1:

[0026] Figure 1 FIG. 1 is a schematic diagram of a target tracking process provided by an embodiment of the present application. The process includes the following steps:

[0027] S101: Identifying in the image collected at the current moment through a pre-trained model to determine the position information and attitude information of the defender agent and the intruder agent at the current moment; for each agent, generating a state vector of the agent at the current moment based on the position information and attitude information of the agent.

[0028] The target tracking method provided by an embodiment of the present application is applied to an electronic device, and the electronic device can be an intelligent device such as a PC or a server.

[0029] In order to accurately and effectively perform target tracking, the image collected at the current moment can be obtained first. The image can be collected by a collection device for monitoring, and the electronic device can locally store a pre-trained model. The electronic device can input the image into the model to obtain the position information and attitude information of the defender agent and the intruder agent at the current moment output by the model. The attitude information includes the instantaneous horizontal angle, the instantaneous pitch angle, and the instantaneous velocity. In a possible implementation manner, the electronic device can input the image and the previous frame image of the image into the model to obtain the position information and attitude information of the defender agent and the intruder agent at the current moment output by the model.

[0030] The electronic device can generate a state vector of each agent at the current moment based on the position information and attitude information of the agent. Among them, the agent includes a defender agent and an intruder agent. Specifically, the electronic device can place the position information of the agent at the first preset position of a preset vector and place the attitude information of the agent at the second preset position of the preset vector.

[0031] The state vector of the agent is:

[0032] r k =[x k ,y k ,z k ,θ k ,β k ,v k T

[0033] Among them, (x k ,y k ,z k ) are the three-dimensional position coordinates of the agent, θ k is the instantaneous horizontal angle of the agent, β k is the instantaneous pitch angle of the agent, v k is the instantaneous speed v of the agent.

[0034] It should be noted that the target tracking scenario in the embodiments of this application is that the intruder agent performs an air mission. Specifically, a three-dimensional air mission area can be defined and two key areas, namely a warning area and a protected area, are set therein. The mission area itself can be regarded as the warning area, while the protected area is specifically defined as a spherical area.

[0035] Among them, in the actual scenario, there may be a situation where multiple defender agents track the same intruder agent. In this application, the case where two defender agents track the same intruder agent can be taken as an example for introduction.

[0036] Figure 2 This is a schematic diagram of a tracking and security defense scenario provided by the embodiments of this application.

[0037] As can be seen from Figure 2 , the air mission area W can be a spherical area. D1, D2, D3, D4, D5, D6 are defender agents (Defenders), is the intruder agent (Intruders), Figure 2 The shown P is the protected area (ProtectedArea). The center point of this protected area is r c , and the radius of this protected area is R P .​

[0038] Among them, it is possible to set that there are n defender agents and m intruder agents in the task space, and use and to represent the defender agent cluster and the intruder agent cluster respectively, and define For the defender agent, the intruder agent represents a non - cooperative group. The goal of the intruder agent is not necessarily the protected area, but may accidentally cross the protected area. Therefore, the goal of the defender agent is to drive away the intruder agent while tracking it, so that it leaves the protected area.

[0039] S102: For a preset number of moments after the current moment, according to the state vector of the defender agent at the current moment and the control input to be predicted for the defender agent at that moment, determine the target state vector of the defender agent at that moment, where the control input includes the instantaneous pitch angular velocity, instantaneous horizontal angular velocity, and instantaneous acceleration of the defender agent.

[0040] To accurately and effectively perform target tracking, the electronic device can, for a preset number of moments after the current moment, where the preset number can be 30, according to the state vector of the defender agent at the current moment and the control input to be predicted for the defender agent at that moment, determine the target state vector of the defender agent at that moment, where the control input includes the instantaneous pitch angular velocity, instantaneous horizontal angular velocity, and instantaneous acceleration of the defender agent.

[0041] Considering the discrete - time open - loop dynamic game, the kinematic models of both the defender agent and the intruder agent can be expressed as r k+1 = r k + ΔT·f(r k , u k , k). That is, the electronic device can determine the target state vector of the defender agent through the following formula:

[0042] r k+1 = r k + ΔT·f(r k , u k , k)

[0043] where r k+1 is the target state vector of the defender agent at the (k + 1)-th moment after the current moment, r k is the state vector of the defender agent at the k - th moment after the current moment, ΔT is the time interval between two adjacent moments, and u k is the control input to be predicted for the defender agent at the k - th moment after the current moment.

[0044] It should be noted that this time interval is a preset time interval. At the same time, the consistency of the states and controls of both the defender and the intruder is maintained.

[0045] The control input to be predicted for the defender agent at the k-th moment after the current moment is:

[0046] u k =[ω k , ξ k , a k T

[0047] where ω k is the predicted pitch angular velocity of the defender agent at the k-th moment after the current moment, ξ is the predicted instantaneous horizontal angular velocity of the defender agent at the k-th moment after the current moment, and a is the predicted instantaneous acceleration of the defender agent at the k-th moment after the current moment.

[0048] It should be noted that in the embodiments of the present application, subscripts i and j or D i and I j can be used to distinguish the defender agent and the intruder agent.

[0049] S103: Determine the target control input at the preset number of moments when the distance is minimized according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distances of the defender agent and the intruder agent; control the defender agent based on the target control input at the preset number of moments.

[0050] In the embodiments of the present application, the electronic device can obtain the first predicted state vectors of the intruder agent at a preset number of moments. In one possible implementation, the electronic device determines the preset control input as the control input of the intruder agent, and based on the preset control input and the state vector of the intruder agent at the current moment, determines the first predicted state vectors of the intruder agent at a preset number of moments. The electronic device can predict the first predicted state vectors of the intruder agent at a preset number of moments according to the following formula:

[0051] r k =r0 + kΔT·f(r0, u0, 0)

[0052] where r k is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, r0 is the state vector of the intruder agent at the current moment, and u0 is the preset control input.

[0053] ​In an embodiment of the present application, the electronic device may determine the target control input at a preset number of moments when the distance is minimized based on the target state vectors of the defender agent at a preset number of moments, the first predicted state vectors of the intruder agent at a preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent.

[0054] Among them, the first relationship specifically refers to the relationship of calculating the distance between the two by obtaining the position information in the state vectors of the defender agent and the intruder agent and based on this position information.

[0055] After determining the target control input at a preset number of moments, the electronic device may control the defender agent based on the obtained target control input at the preset number of moments.

[0056] Since in the actual scenario, the intruder agent will avoid colliding with the defender agent, during the tracking process of the defender agent, the intruder agent will move away from the defender agent and thus away from the corresponding protected area.

[0057] The embodiment of the present application is equivalent to providing a multi-agent target tracking and security defense technology framework based on dynamic game.

[0058] Since in the embodiment of the present application, the electronic device constructs the state vectors of the defender agent and the intruder agent based on the collected images; and determines the target control input at a preset number of moments when the distance is minimized according to the target state vectors of the defender agent at a preset number of moments, the first predicted state vectors of the intruder agent at a preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent; and controls the defender agent based on the target control input at a preset number of moments, so that the defender agent can quickly and effectively approach the intruder agent and accurately and effectively achieve target tracking.

[0059] Embodiment 2:

[0060] In order to accurately and effectively perform target tracking, based on the above embodiment, in the embodiment of the present application, determining the target control input at a preset number of moments when the distance is minimized according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent includes:

[0061] Determine the target control input at the preset number of moments when the defensive agent has a target state vector at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and a first relationship between the distance of the intruder agent, such that the defensive agent and each other defensive agent have a defensive distance greater than a preset minimum collision distance at the preset number of moments, and the target control input at the preset number of moments when the distance is minimized; wherein, the defensive distance between a defensive agent and each other defensive agent at a certain moment is determined according to the target state vector of the defensive agent at that moment, the second predicted state vector of each other defensive agent at that moment, and a second relationship between the defensive distance of the defensive agent and each other defensive agent.

[0062] To accurately determine the target control input that can achieve the minimum distance at a preset number of critical moments, the electronic device needs to first clarify the constraints that the defensive agent must follow and determine the optimal control input based on these constraints. Specifically, to ensure the safety of the target tracking process, the defensive agent must maintain a safe distance from other defensive agents. Therefore, this constraint aims to ensure that an appropriate safety distance is maintained between defensive agents.

[0063] The electronic device can determine the target control input at the preset number of moments when the defensive agent has a target state vector at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and a first relationship between the distance of the intruder agent, such that the defensive agent and each other defensive agent have a defensive distance greater than a preset minimum collision distance at the preset number of moments, and the target control input at the preset number of moments when the distance is minimized; wherein, the defensive distance between a defensive agent and each other defensive agent at a certain moment is determined according to the target state vector of the defensive agent at that moment, the second predicted state vector of each other defensive agent at that moment, and a second relationship between the defensive distance of the defensive agent and each other defensive agent.

[0064] Regarding the constraints on the defensive agent, in the embodiments of the present application, the target of the defensive agent is transformed from a soft constraint in the cost function into an inequality hard constraint. The hard constraint condition for safety defense between defensive agents can be expressed as:

[0065]

[0066] Wherein, is any other defensive agent except for the defensive agent with serial number i, is the state vector of the i-th defensive agent, is the state vector of any defensive agent other than the i-th defensive agent, is each defensive agent, Q aFor a custom matrix, The calculation result is the three-dimensional position vector of the i-th defender agent at the k-th moment after the current moment, d def,def is the preset minimum collision distance. It should be noted that the inequality constraint conditions of each defender agent include the state information of other defender agents, which highlights the interactive nature of the game. This design reflects the dynamic interaction within the defender agents, indicating that when formulating their own action plans, the defender agents must consider the possible behaviors and position changes of other defender agents. This consideration of interactivity ensures the collaboration and real-time nature of the strategy.

[0067] Embodiment 3:

[0068] In order to accurately and effectively determine the target control input, based on the above embodiments, in the embodiments of the present application, the determining the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agents at the preset number of moments, the first predicted state vectors of the intruder agents at the preset number of moments, and the first relationship between the distances of the defender agents and the intruder agents includes:

[0069] Determine the target control input at the preset number of moments when the distance is the smallest through the following formula:

[0070]

[0071] Where, is the target state vector of the defender agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, and N is the preset number.

[0072] In order to accurately and effectively determine the target control input, the electronic device can determine the target control input at the preset number of moments when the distance is the smallest through the following formula:

[0073]

[0074] Where, is the target state vector of the defender agent at the k-th moment after the current moment, r′ goal is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, and N is the preset number.

[0075] Specifically, the purpose of the embodiments of the present application is to determine When it is the smallest Among them, is the target state vector of the defender agent at the (k + 1)-th moment after the current moment, is the state vector of the defender agent at the k-th moment after the current moment, ΔT is the time interval between two adjacent moments, is the control input to be predicted for the defender agent at the k-th moment after the current moment.

[0076] It should be noted that, since there are multiple defender agents in the actual scenario, the electronic device can also determine the target control input at the preset number of moments when the distance is the smallest through the following formula:

[0077]

[0078] Among them, is the state vector of the l-th defender agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, N is the preset number, and n is the total number of defender agents.

[0079] It should be noted that for the defender agent, its target point is the real-time position of the intruder agent; in the operation cost of the defender agent, the design of determining the target control input can make the defender agent approach the target intruder agent, and at the same time anticipate the behavior that the intruder agent may move closer to the protected area.

[0080] Embodiment 4:

[0081] In order to accurately and effectively perform target tracking, based on the above embodiments, in the embodiments of the present application, the determining the target control input at the preset number of moments when the distance is the smallest according to the target state vector of the defender agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent includes:

[0082] Determining the target control input at the preset number of moments when the sum of the distance and the consumed energy is the smallest according to the target state vector of the defender agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, the first relationship between the distance between the defender agent and the intruder agent, the control input to be predicted for the defender agent, and the third relationship between the consumed energy of the defender agent.

[0083] Since the defender agent needs to minimize the energy consumption during operation, in order to accurately and effectively perform target tracking, the electronic device can determine the target control inputs at a preset number of moments when the sum of the distance and the energy consumption is minimized.

[0084] Specifically, the electronic device can determine the target control inputs at a preset number of moments when the sum of the distance and the energy consumption is minimized according to the target state vectors of the defender agent at a preset number of moments, the first predicted state vectors of the intruder agent at a preset number of moments, the first relationship between the distance between the defender agent and the intruder agent, the control inputs to be predicted of the defender agent, and the third relationship between the energy consumption of the defender agent. Specifically, the corresponding cost values can be determined according to the distance, the energy consumption, and the respective weights, and the target control inputs at a preset number of moments when the cost value is minimized can be determined.

[0085] Embodiment 5:

[0086] In order to accurately and effectively perform target tracking, based on the above embodiments, in the embodiments of the present application, the step of determining the target control inputs at a preset number of moments when the sum of the distance and the energy consumption is minimized according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, the first relationship between the distance between the defender agent and the intruder agent, the control inputs to be predicted of the defender agent, and the third relationship between the energy consumption of the defender agent includes:

[0087] Determine the target control inputs at a preset number of moments when the sum of the distance and the energy consumption is minimized through the following formula:

[0088]

[0089] Where, is the target state vector of the defender agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, N is the preset number, is the control input to be predicted of the defender agent at the k-th moment after the current moment, R i,h is the second preset matrix, is the weight corresponding to the energy consumption.

[0090] In order to accurately and effectively determine the target control inputs, the electronic device can determine the target control inputs at a preset number of moments when the distance is minimized through the following formula:

[0091]

[0092] Among them, is the target state vector of the defender agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, N is the preset quantity, is the control input to be predicted of the defender agent at the k-th moment after the current moment, R i,h is the second preset matrix, is the weight corresponding to the consumed energy.

[0093] Specifically, the purpose of the embodiment of the present application is to determine when it is the smallest Among them, is the target state vector of the defender agent at the (k + 1)-th moment after the current moment, is the state vector of the defender agent at the k-th moment after the current moment, ΔT is the time interval between two adjacent moments, is the control input to be predicted of the defender agent at the k-th moment after the current moment.

[0094] It should be noted that since there are multiple defender agents in the actual scenario, the electronic device can also determine the target control input at the preset number of moments when the distance is the smallest through the following formula:

[0095]

[0096] Among them, is the state vector of the l-th defender agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, N is the preset quantity, n is the total number of defender agents, is the control input to be predicted of the defender agent at the k-th moment after the current moment, R i,h is the second preset matrix, is the weight corresponding to the consumed energy.

[0097] In the embodiment of the present application, can be expressed as Through recursion, the embodiment of the present application can substitute the dynamics of the agent into the cost function, and the cost function can be expressed as a function of the open-loop control input of the agent:

[0098]

[0099] The above equation reflects the coupling relationship of the control inputs between the defender agent and the intruder agent, that is, both sides will predict the control inputs of the other side to adjust their own strategies.

[0100] Meanwhile, the agent system is also restricted by constraint conditions, and the constraint conditions in the embodiments of this application can be expressed as:

[0101] C(u 1:(n+m) , r0) ≤ 0

[0102] The constraint conditions of the agent system are not unique, such as security defense constraints, etc. The constraint conditions also reflect the interaction and coupling relationship between the input of the agent itself and other agents.

[0103] In the embodiments of this application, the electronic device can solve the Nash equilibrium based on the quasi-Newton sequence quadratic game algorithm (GN-SQG) to determine the target control inputs at a preset number of moments after the current moment.

[0104] The optimization problem of the game system can be generally expressed as:

[0105]

[0106] s.t. c I (u 1:n ) ≥ 0

[0107] Where r in the input is omitted here.

[0108] The KKT conditions of the game optimization problem formula constitute the local Nash equilibrium of the dynamic game, specifically:

[0109]

[0110] Where is a diagonal matrix with the elements in as the main diagonal elements.

[0111] Specifically, how the KKT conditions of the game optimization problem formula constitute the local Nash equilibrium of the dynamic game is prior art and will not be elaborated here.

[0112] Next, linearization processing is performed. The KKT conditions are expanded by the first-order Taylor expansion at the point to obtain:

[0113]

[0114] Where M I,(k+1) is a diagonal matrix with the elements in as the main diagonal elements.

[0115] Construct a new quadratic game programming problem, and its optimal solution is the exact KKT condition of the above formula:

[0116]

[0117] By linearizing the KKT conditions to approximate the exact KKT conditions and solving the linearized KKT conditions, a quadratic game programming problem can be equivalently solved.

[0118] Specifically, how to approximate the exact KKT conditions by linearizing the KKT conditions and solve the linearized KKT conditions is prior art and will not be elaborated here.

[0119] The main challenge faced by the SQG problem is how to reduce the computational complexity when dealing with larger-scale problems, so as to improve the overall computational efficiency and reduce the demand for computing resources. On the other hand, the embodiments of the present application need to ensure the convergence of the algorithm, so that an effective optimization direction can be found in each iteration. However, traditional SQG methods only have local convergence, must be close enough to u *,1:n , and the condition for the problem to have a unique solution is positive definite. This prerequisite is often difficult to meet when solving games. Therefore, the embodiments of the present application propose a Nash equilibrium solving method for QN-SQG to improve the convergence range of the algorithm. Next, the embodiments of the present application will describe the improvement of the algorithm in detail.

[0120] (1) Quasi-Newton-based approximate correction

[0121] First, the embodiments of the present application discuss improving the computational efficiency of the algorithm by introducing the quasi-Newton method. In each iteration, the SQG algorithm needs to solve the quadratic programming subproblems of each player. This process requires calculating the Hessian matrix Due to its high computational complexity, it often becomes the main part of the time consumption in the optimization process, especially when dealing with large-scale multi-player problems. To solve this problem, the embodiments of the present application adopt the idea of the quasi-Newton method, add the Broyden-Fletcher-Goldfarb-Shanno (BFGS) correction formula, and approximate the Hessian matrix by establishing a positive definite matrix to avoid direct calculation of it and continuously update and correct it during the optimization process. Specifically, define a variable:

[0122]

[0123] where θ bi,k is defined as:

[0124]

[0125] Next, define the correction formula of

[0126]

[0127] It should be noted that at the initial moment, in the embodiments of the present application, is set to the identity matrix I. When dealing with optimization problems with limited computing resources or large problem scales, through this strategy algorithm, while avoiding high-cost Hessian matrix calculations, the accuracy and stability of the optimization process can still be maintained.

[0128] Thus, the quadratic game programming sub-problem can be approximately written as:

[0129]

[0130] (2) Benefit function design

[0131] In the embodiments of the present application, the benefit function is designed to transfer the quadratic programming sub-problem of the player solved in the iterative process to obtaining the optimal solution of the benefit function. Among them, the quadratic programming sub-problem of the player represents the descent direction of the current iteration point of the benefit function. The designed benefit function is as follows:

[0132]

[0133] Among them, p I is the penalty factor for inequality constraints. It can be seen that the benefit function includes a first-order optimal condition tracking term and an inequality constraint penalty term.

[0134] By solving the quadratic game programming sub-problem, the embodiments of the present application can obtain the descent direction merit (u 1:n , μ I , p I ) of the benefit function ψ at the iteration point In order to ensure that in each iteration process, the solution of the algorithm can take a step forward along the descent direction of ψ merit and gradually converge to the minimum value, the Armijo inexact search method is used in the present application to obtain the step size of the iteration. Through the combination of the above benefit function and line search strategy, the embodiments of the present application find the optimal iteration step size and realize the rapid advancement of the algorithm towards the optimal solution direction.

[0135] In the embodiments of the present application, a method based on the Quasi-Newton Sequence Quadratic Game (QN-SQG) can transform the safety distance constraint into specific control input constraints by introducing a control barrier function, and optimize the Nash equilibrium strategies of the defender agent and the intruder agent. This method continuously optimizes the solution through the Quasi-Newton approximation correction technique. Compared with traditional methods, it significantly improves the computational efficiency and the robustness of the algorithm. In addition, in the embodiments of the present application, a benefit function is established to ensure the stable convergence of the strategy during the iteration process, providing an effective and reliable security defense strategy path and an innovative solution for target tracking and security defense in practical applications.

[0136] Facing the strict constraint conditions in the defense task, the embodiments of the present application adopt the control barrier function theory to design an innovative non-cooperative target tracking and security defense method based on QN-SQG. On the basis of ensuring that key constraint conditions such as the safety distance are met, this method effectively realizes the tracking and defense of non-cooperative targets, and significantly improves the safety and reliability of the multi-agent system when performing defense tasks.

[0137] This method not only improves the computational efficiency problem of traditional methods in dealing with complex interactions between agents, but also ensures the robust convergence of the predicted trajectory during the iteration process. This method demonstrates an efficient and flexible method for protecting critical assets in a complex environment, enhancing the defense ability and adaptability of the multi-agent system in the face of evolving threats.

[0138] To solve the problem that the target cannot be accurately and effectively tracked in the related technologies, a dynamic target tracking method based on Model Predictive Control (MPC) is proposed. It uses Kalman filtering to predict the target position and combines the MPC algorithm for tracking, but these methods do not cover the cooperative strategies of agent clusters or the safety distance problem. The progress of dynamic game theory provides new solutions to these challenges. Emerging algorithms, such as iterative best response and iterative linear quadratic algorithms, have greatly improved the computational efficiency. In particular, the sequence Quadratic Gaussian (SQG) method and the Iterative Linear Quadratic Gaussian (ILQG) method have improved the efficiency and convergence of problem solving.

[0139] Example 6:

[0140] For accurate and effective target tracking, based on the above embodiments, in the embodiments of the present application, the first predicted state vector of the intruder agent at the preset number of moments is determined as follows:

[0141] For a preset number of moments after the current moment, according to the state vector of the intruder agent at the current moment and the predicted control input of the intruder agent at the preset number of moments, determine the first predicted state vector of the intruder agent at this moment.

[0142] For accurate and effective target tracking, an electronic device can, for a preset number of moments after the current moment, determine the first predicted state vector of the intruder agent at this moment according to the state vector of the intruder agent at the current moment and the predicted control input of the intruder agent at the preset number of moments.

[0143] Specifically, the electronic device can determine the first predicted state vector of the intruder agent at the preset number of moments through the following formula:

[0144] r k+1 = r k + ΔT·f(r k , u k , k), r k+1 is the target state vector of the defender agent at the (k + 1)-th moment after the current moment, r k is the state vector of the defender agent at the k-th moment after the current moment, ΔT is the time interval between two adjacent moments, u k is the control input to be predicted for the defender agent at the k-th moment after the current moment.

[0145] For accurate and effective target tracking, based on the above embodiments, in the embodiments of the present application, the control input of the intruder agent at the preset number of moments is determined through the following formula:

[0146]

[0147] Wherein, is the state vector of the intruder agent at the k-th moment after the current moment, is the state vector of the intruder agent at the (k - 1)-th moment after the current moment, ΔT is the time interval between two adjacent moments, is a preset function, is the preset state vector corresponding to the pre-saved protected area, is the control input to be predicted for the intruder agent at the (k - 1)-th moment after the current moment, Q j,xis the third preset matrix, is the weight corresponding to the distance, is the control input to be predicted for the intruder agent at the k-th moment after the current moment, R j,h is the fourth preset matrix and is a positive definite parameter matrix, is the weight corresponding to the energy consumption, is the closest distance between the intruder agent and each defender agent at the k-th moment after the current moment, M is the set of each defender agent tracking the intruder agent, Q a is the fifth preset matrix, is the third predicted state vector of the j-th defender agent at the k-th moment after the current moment, is the state vector of the j-th defender agent at the current moment, u0 is the preset control input.

[0148] The electronic device can determine the control input of the intruder agent at a preset number of moments through the following formula:

[0149]

[0150] where, is the state vector of the intruder agent at the k-th moment after the current moment, is the state vector of the intruder agent at the (k - 1)-th moment after the current moment, ΔT is the time interval between two adjacent moments, is the preset function, is the preset state vector corresponding to the pre-saved protected area, is the control input to be predicted for the intruder agent at the (k - 1)-th moment after the current moment, Q j,x is the third preset matrix, is the weight corresponding to the distance, is the control input to be predicted for the intruder agent at the k-th moment after the current moment, R j,h is the fourth preset matrix, is the weight corresponding to the energy consumption, is the closest distance between the intruder agent and each defender agent at the k-th moment after the current moment, M is the set of each defender agent tracking the intruder agent, Q a is the fifth preset matrix, is the third predicted state vector of the j-th defender agent at the k-th moment after the current moment, is the state vector of the j-th defender agent at the current moment, u0 is the preset control input.

[0151] That is to say, the electronic device can assume that the intruder agent predicts the control input of the intruder agent at a preset number of moments after the current moment based on the preset control input and the position of the defender agent at the current moment, and then determines the state vector of the intruder agent at a preset number of moments after the current moment based on the predicted control input of the intruder agent at a preset number of moments after the current moment.

[0152] It should be noted that since there are multiple intruder agents in the actual scenario, the electronic device can also determine the control input of the intruder agent at a preset number of moments through the following formula:

[0153]

[0154] Where m is the total number of intruder agents. In the running cost of the intruder agent, the first term aims to push the intruder agent closer to the protected area, taking into account the tracking activities of the defender agent; the second term involves the energy consumption of the intruder agent; and the third term reflects the evasion behavior of the intruder agent attempting to avoid being tracked by the defender agent.

[0155] For the constraint conditions of the intruder agent, it is mainly the enemy avoidance constraint of maintaining a certain distance from the defender agent. Therefore, the constraint conditions of the intruder agent can be expressed as:

[0156]

[0157] The goals of the defender agent and the intruder agent are both to minimize their own cost functions through strategy selection. By finding the Nash equilibrium point, the embodiments of the present application can obtain the optimal strategy combination of the defender agent and the intruder agent in a given scenario. Combining the above, the optimization problem of the defender agent in discrete time can be generally expressed as:

[0158]

[0159] Similarly, the expression of the optimization problem of the intruder agent can be obtained.

[0160] Since the cost functions of the defender agent and the intruder agent can be expressed as functions of the control input, that is Therefore, the Nash equilibrium solution of this problem is {u 1,* ,u 1,* ,...,u (n+m),*}, which satisfies:

[0161] Where Denote any defender agent or intruder agent other than agent l. In the equilibrium state, no agent can improve the player's payoff by unilaterally adjusting its own strategy.

[0162] Figure 3 This is a schematic diagram of the trajectory of a dynamic target tracking and security defense task provided by an embodiment of the present application.

[0163] From Figure 3 it can be seen that under the condition that the intruder agent always maintains a certain distance from the defender agent due to its own escape intention and enemy avoidance constraint, the defender agent successfully completed the tracking and defense of the intruder agent within 30 time steps, preventing the intruder agent from entering the protected area.

[0164] Figure 4 This is a schematic diagram of the change in the distance between agents during the target tracking process provided by an embodiment of the present application.

[0165] From Figure 4 it can be seen that the defender agents always maintain a safe defense distance.

[0166] The embodiment of the present application discusses the tracking game in a 2V1 configuration, that is, the scenario of two defender agent drones and one intruder agent drone. The protected area is defined as a spherical area with the center located at r c = [15, 25, 30] and the radius R p = 5m. Set the target point r of the intruder agent g = r c , that is, in the absence of the intervention of the defender agent, the intruder agent will directly enter the center of the protected area. The initial positions of the defender agent and the intruder agent are both within the task area W. Specifically, the starting point of the defender agent is selected as [22, 22, 30], the starting point of the defender agent is selected as [25, 15, 30], and the starting point of the intruder agent is selected as [35, 25, 40]. The time interval of this example is set to 0.1 second, and the termination condition of the tracking task is that the time step reaches K tmax = 30 or the total amount of reconnaissance information (i.e., the quantitative total of the tracking effect) C = 200.

[0167] The parameters of the cost function of the defender agent are and while the parameters of the cost function of the intruder agent are and The upper and lower limits of the control inputs of the defender agent and the intruder agent are both u = [-π / 2, -π / 2, -20]. For the defender agent, it is equipped with a directional sensor. In this example, the sensing radius R = 5m, the maximum sensing horizontal angle Γ = π / 4 rad, and the maximum sensing pitch angle Ψ = π / 4 rad. For the safety defense constraint and the enemy avoidance constraint parameters, this example selects d def,def = d def,int = 2m. It should be noted that considering that the primary goal of the intruder agent is to enter the protected area, the enemy avoidance constraint does not need to be strictly enforced. Therefore, during the execution of the algorithm, if the decision variable output of the intruder agent is an empty set due to the restriction of the enemy avoidance constraint, then at this moment, the inequality constraint of the intruder agent is discarded, and this iteration is re-executed.

[0168] Example 7:

[0169] Figure 5 A schematic structural diagram of a target tracking device provided by an embodiment of the present application. The device includes:

[0170] A determination and generation module 501, configured to identify the position information and attitude information of the defender agent and the intruder agent in the image collected at the current moment through a pre-trained model; for each agent, generate a state vector of the agent at the current moment based on the position information and attitude information of the agent; for a preset number of moments after the current moment, determine the target state vector of the defender agent at this moment according to the state vector of the defender agent at the current moment and the control input to be predicted of the defender agent at this moment, where the control input includes the instantaneous pitch angular velocity, the instantaneous horizontal angular velocity, and the instantaneous acceleration of the defender agent;

[0171] A processing module 502, configured to determine the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship of the distance between the defender agent and the intruder agent; control the defender agent based on the target control input at the preset number of moments.

[0172] In a possible implementation manner, the processing module 502 is specifically configured to determine, according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distances of the intruder agent, the target control inputs at the preset number of moments when the defense distance between the defender agent and each other defender agent is greater than a preset minimum collision distance and the distance is the smallest; wherein, the defense distance between the defender agent and each other defender agent at a certain moment is determined according to the target state vector of the defender agent at this moment, the second predicted state vector of each other defender agent at this moment, and the second relationship between the defense distances of the defender agent and each other defender agent.

[0173] In a possible implementation manner, the processing module 502 is specifically configured to determine the target control inputs at the preset number of moments when the distance is the smallest through the following formula:

[0174]

[0175] Wherein, is the target state vector of the defender agent at the kth moment after the current moment, is the first predicted state vector of the predicted intruder agent at the kth moment after the current moment, Q i,l is the first preset matrix, and N is the preset number.

[0176] In a possible implementation manner, the processing module 502 is specifically configured to determine the target control inputs at the preset number of moments when the sum of the distance and the consumed energy is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, the first relationship between the distances of the defender agent and the intruder agent, the control input to be predicted of the defender agent, and the third relationship between the consumed energy of the defender agent.

[0177] In a possible implementation manner, the processing module 502 is specifically configured to determine the target control inputs at the preset number of moments when the sum of the distance and the consumed energy is the smallest through the following formula:

[0178]

[0179] Wherein, is the target state vector of the defender agent at the kth moment after the current moment, is the first predicted state vector of the predicted intruder agent at the kth moment after the current moment, Qi,l is the first preset matrix, N is the preset quantity, is the control input to be predicted at the k-th moment after the current moment for the defender agent, R i,h is the second preset matrix, is the weight corresponding to the consumed energy.

[0180] In a possible implementation manner, the processing module 502 is further configured to determine the first predicted state vector of the intruder agent at the preset number of moments in the following manner:

[0181] For a preset number of moments after the current moment, according to the state vector of the intruder agent at the current moment and the predicted control input of the intruder agent at the preset number of moments, determine the first predicted state vector of the intruder agent at this moment.

[0182] In a possible implementation manner, the processing module 502 is further configured to determine the control input of the intruder agent at the preset number of moments through the following formula:

[0183]

[0184] where, is the state vector of the intruder agent at the k-th moment after the current moment, is the state vector of the intruder agent at the (k - 1)-th moment after the current moment, ΔT is the time interval between two adjacent moments, is a preset function, is the preset state vector corresponding to the pre-saved protected area, is the control input to be predicted at the (k - 1)-th moment after the current moment for the intruder agent, Q j,x is the third preset matrix, is the weight corresponding to the distance, is the control input to be predicted at the k-th moment after the current moment for the intruder agent, R j,h is the fourth preset matrix, is the weight corresponding to the consumed energy, is the shortest distance between the intruder agent and each defender agent at the k-th moment after the current moment, M is the set of each defender agent tracking the intruder agent, Q a is the fifth preset matrix, is the third predicted state vector of the j-th defender agent at the k-th moment after the current moment, is the state vector of the j-th defender agent at the current moment, u0 is the preset control input.

[0185] Embodiment 8:

[0186] Figure 4 A schematic structural diagram of an electronic device provided by the present invention. On the basis of the above embodiments, an embodiment of the present application further provides an electronic device, as Figure 6 shown, including: a processor 601, a communication interface 602, a memory 603, and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604;

[0187] A computer program is stored in the memory 603. When the program is executed by the processor 601, the processor 601 is caused to execute the following steps:

[0188] Identify the position information and attitude information of the defender agent and the intruder agent in the image collected at the current moment through a pre-trained model; for each agent, generate a state vector of the agent at the current moment based on the position information and attitude information of the agent;

[0189] For a preset number of moments after the current moment, determine the target state vector of the defender agent at this moment according to the state vector of the defender agent at the current moment and the control input to be predicted of the defender agent at this moment, wherein the control input includes the instantaneous pitch angular velocity, the instantaneous horizontal angular velocity, and the instantaneous acceleration of the defender agent;

[0190] According to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distances of the defender agent and the intruder agent, determine the target control input at the preset number of moments when the distance is the smallest; control the defender agent based on the target control input at the preset number of moments.

[0191] In a possible implementation manner, determining the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distances of the defender agent and the intruder agent includes:

[0192] Determine the target control input at the preset number of moments when the defensive agent has a defensive distance greater than the preset minimum collision distance from each other defensive agent at the preset number of moments, and the distance is the smallest, according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and the first relationship between the defensive agent and the distance of the intruder agent; wherein, the defensive distance between the defensive agent and each other defensive agent at a certain moment is determined according to the target state vector of the defensive agent at this moment, the second predicted state vector of each other defensive agent at this moment, and the second relationship between the defensive agent and the defensive distance of each other defensive agent.

[0193] In a possible implementation manner, the determining of the target control input at the preset number of moments when the distance is the smallest according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and the first relationship between the defensive agent and the distance of the intruder agent includes:

[0194] Determine the target control input at the preset number of moments when the distance is the smallest through the following formula:

[0195]

[0196] wherein, is the target state vector of the defensive agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, and N is the preset number.

[0197] In a possible implementation manner, the determining of the target control input at the preset number of moments when the distance is the smallest according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and the first relationship between the defensive agent and the distance of the intruder agent includes:

[0198] Determine the target control input at the preset number of moments when the sum of the distance and the consumed energy is the smallest according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, the first relationship between the defensive agent and the distance of the intruder agent, the control input to be predicted of the defensive agent, and the third relationship between the defensive agent and the consumed energy.

[0199] In a possible implementation manner, determining the target control input at the preset number of moments when the sum of the distance and the energy consumption is the smallest according to the target state vector of the defender agent at the preset number of moments, the first relationship between the first predicted state vector of the intruder agent at the preset number of moments and the distance between the defender agent and the intruder agent, the control input to be predicted of the defender agent, and the third relationship between the energy consumption of the defender agent includes:

[0200] Determining the target control input at the preset number of moments when the sum of the distance and the energy consumption is the smallest through the following formula:

[0201]

[0202] Wherein, is the target state vector of the defender agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, N is the preset number, is the control input to be predicted of the defender agent at the k-th moment after the current moment, R i,h is the second preset matrix, is the weight corresponding to the energy consumption.

[0203] In a possible implementation manner, the first predicted state vector of the intruder agent at the preset number of moments is determined by the following method:

[0204] For the preset number of moments after the current moment, according to the state vector of the intruder agent at the current moment and the predicted control input of the intruder agent at the preset number of moments, determine the first predicted state vector of the intruder agent at that moment.

[0205] In a possible implementation manner, the control input of the intruder agent at the preset number of moments is determined through the following formula:

[0206]

[0207] Wherein, is the state vector of the intruder agent at the k-th moment after the current moment, is the state vector of the intruder agent at the (k - 1)-th moment after the current moment, ΔT is the time interval between two adjacent moments, is a preset function, is the preset state vector corresponding to the pre-saved protected area, is the control input to be predicted for the intruder agent at the (k-1)-th moment after the current moment, Q j,x is the third preset matrix, is the weight corresponding to the distance, is the control input to be predicted for the intruder agent at the k-th moment after the current moment, R j,h is the fourth preset matrix, is the weight corresponding to the energy consumption, is the shortest distance between the intruder agent and each defender agent at the k-th moment after the current moment, M is the set of each defender agent tracking the intruder agent, Q a is the fifth preset matrix, is the third predicted state vector of the predicted j-th defender agent at the k-th moment after the current moment, is the state vector of the j-th defender agent at the current moment, u0 is the preset control input.

[0208] The communication bus mentioned in the above server can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity in representation, only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0209] The communication interface is used for communication between the above electronic device and other devices.

[0210] The memory can include a Random Access Memory (RAM), and can also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory can also be at least one storage device located far from the aforementioned processor.

[0211] The above processor can be a general-purpose processor, including a central processor, a Network Processor (NP), etc.; it can also be a Digital Signal Processing (DSP), an application-specific integrated circuit, a field-programmable gate array, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0212] Example 9:

[0213] Based on the above embodiments, an embodiment of the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program executable by an electronic device. When the program runs on the electronic device, the electronic device is caused to execute the following steps when executed:

[0214] A computer program is stored in the memory. When the program is executed by the processor, the processor is caused to execute the following steps:

[0215] Identify the position information and attitude information of the defender agent and the intruder agent in the image collected at the current moment through a pre-trained model; for each agent, generate a state vector of the agent at the current moment based on the position information and attitude information of the agent.

[0216] For a preset number of moments after the current moment, determine the target state vector of the defender agent at that moment according to the state vector of the defender agent at the current moment and the control input to be predicted for the defender agent at that moment, where the control input includes the instantaneous pitch angular velocity, instantaneous horizontal angular velocity, and instantaneous acceleration of the defender agent.

[0217] According to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distances of the defender agent and the intruder agent, determine the target control input at the preset number of moments when the distance is the smallest; control the defender agent based on the target control input at the preset number of moments.

[0218] In a possible implementation manner, determining the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distances of the defender agent and the intruder agent includes:

[0219] Determine the target control input at the preset number of moments when the defensive agent has a defensive distance greater than the preset minimum collision distance from each other defensive agent at the preset number of moments, and the distance is the smallest, according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and the first relationship of the distance between the defensive agent and the intruder agent; wherein, the defensive distance between the defensive agent and each other defensive agent at a certain moment is determined according to the target state vector of the defensive agent at this moment, the second predicted state vector of each other defensive agent at this moment, and the second relationship of the defensive distance between the defensive agent and each other defensive agent.

[0220] In a possible implementation manner, the determining of the target control input at the preset number of moments when the distance is the smallest according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and the first relationship of the distance between the defensive agent and the intruder agent includes:

[0221] Determine the target control input at the preset number of moments when the distance is the smallest through the following formula:

[0222]

[0223] wherein, is the target state vector of the defensive agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, and N is the preset number.

[0224] In a possible implementation manner, the determining of the target control input at the preset number of moments when the distance is the smallest according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, and the first relationship of the distance between the defensive agent and the intruder agent includes:

[0225] Determine the target control input at the preset number of moments when the sum of the distance and the consumed energy is the smallest according to the target state vector of the defensive agent at the preset number of moments, the first predicted state vector of the intruder agent at the preset number of moments, the first relationship of the distance between the defensive agent and the intruder agent, the control input to be predicted of the defensive agent, and the third relationship of the consumed energy of the defensive agent.

[0226] In a possible implementation manner, determining the target control input at the preset number of moments when the sum of the distance and the energy consumption is the smallest according to the target state vector of the defender agent at the preset number of moments, the first relationship between the first predicted state vector of the intruder agent at the preset number of moments and the distance between the defender agent and the intruder agent, the control input to be predicted of the defender agent, and the third relationship between the energy consumption of the defender agent includes:

[0227] Determining the target control input at the preset number of moments when the sum of the distance and the energy consumption is the smallest through the following formula:

[0228]

[0229] Wherein, is the target state vector of the defender agent at the kth moment after the current moment, is the first predicted state vector of the predicted intruder agent at the kth moment after the current moment, Q i,l is the first preset matrix, N is the preset number, is the control input to be predicted of the defender agent at the kth moment after the current moment, R i,h is the second preset matrix, is the weight corresponding to the energy consumption.

[0230] In a possible implementation manner, the first predicted state vector of the intruder agent at the preset number of moments is determined by the following method:

[0231] For a preset number of moments after the current moment, according to the state vector of the intruder agent at the current moment and the control input of the predicted intruder agent at the preset number of moments, determine the first predicted state vector of the intruder agent at that moment.

[0232] In a possible implementation manner, the control input of the intruder agent at the preset number of moments is determined through the following formula:

[0233]

[0234] Wherein, is the state vector of the intruder agent at the kth moment after the current moment, is the state vector of the intruder agent at the (k - 1)th moment after the current moment, ΔT is the time interval between two adjacent moments, is a preset function, is the preset state vector corresponding to the pre - saved protected area, is the control input to be predicted for the intruder agent at the (k-1)-th moment after the current moment, Q j,x is the third preset matrix, is the weight corresponding to the distance, is the control input to be predicted for the intruder agent at the k-th moment after the current moment, R j,h is the fourth preset matrix, is the weight corresponding to the energy consumption, is the shortest distance between the intruder agent and each defender agent at the k-th moment after the current moment, M is the set of each defender agent tracking the intruder agent, Q a is the fifth preset matrix, is the third predicted state vector of the predicted j-th defender agent at the k-th moment after the current moment, r0 j is the state vector of the j-th defender agent at the current moment, u0 is the preset control input.

[0235] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0236] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0237] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1The functions specified in one or more boxes.

[0238] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing in the process Figure 1 One process or more processes and / or boxes Figure 1 The steps of the functions specified in one or more boxes.

[0239] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to include these changes and modifications.

Claims

1. A target tracking method, characterized in that, The method includes: Identifying, in the image collected at the current moment through a pre-trained model, the position information and attitude information of the defender agent and the intruder agent at the current moment; for each agent, generating a state vector of the agent at the current moment based on the position information and attitude information of the agent; For a preset number of moments after the current moment, determining the target state vector of the defender agent at this moment according to the state vector of the defender agent at the current moment and the control input to be predicted for the defender agent at this moment, where the control input includes the instantaneous pitch angular velocity, the instantaneous horizontal angular velocity, and the instantaneous acceleration of the defender agent; Determining the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship of the distance between the defender agent and the intruder agent; controlling the defender agent based on the target control input at the preset number of moments.

2. The method according to claim 1, wherein Determining the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship of the distance between the defender agent and the intruder agent, includes: Determining the target control input at the preset number of moments when the distance is the smallest and the defense distance between the defender agent and each other defender agent at the preset number of moments is greater than the preset minimum collision distance according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship of the distance between the defender agent and the intruder agent; where the defense distance between the defender agent and each other defender agent at a certain moment is determined according to the target state vector of the defender agent at this moment, the second predicted state vector of each other defender agent at this moment, and the second relationship of the defense distance between the defender agent and each other defender agent.

3. The method according to claim 1, characterized in that, The determining the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship of the distance between the defender agent and the intruder agent, includes: Determining the target control input at the preset number of moments when the distance is the smallest through the following formula: wherein, is the target state vector of the defender agent at the (k)-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the (k)-th moment after the current moment, Q i,l is the first preset matrix, and N is the preset quantity.

4. The method according to claim 1, wherein The determining the target control input at the preset number of moments when the distance is the smallest according to the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship of the distance between the defender agent and the intruder agent, includes: Determine the target control inputs at the preset number of moments when the sum of the distance and the energy consumption is minimized, based on the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, the first relationship between the distance between the defender agent and the intruder agent, the control input to be predicted of the defender agent, and the third relationship between the energy consumption of the defender agent.

5. The method according to claim 4, characterized in that The step of determining the target control inputs at the preset number of moments when the sum of the distance and the energy consumption is minimized, based on the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, the first relationship between the distance between the defender agent and the intruder agent, the control input to be predicted of the defender agent, and the third relationship between the energy consumption of the defender agent, includes: Determine the target control inputs at the preset number of moments when the sum of the distance and the energy consumption is minimized through the following formula: Among them, is the target state vector of the defender agent at the k-th moment after the current moment, is the first predicted state vector of the predicted intruder agent at the k-th moment after the current moment, Q i,l is the first preset matrix, N is the preset quantity, is the control input to be predicted of the defender agent at the k-th moment after the current moment, R i,h is the second preset matrix, is the weight corresponding to the consumed energy.

6. The method according to any one of claims 1-5, characterized in that, The first predicted state vectors of the intruder agent at the preset number of moments are determined by the following method: For a preset number of moments after the current moment, based on the state vector of the intruder agent at the current moment and the predicted control inputs of the intruder agent at the preset number of moments, determine the first predicted state vector of the intruder agent at that moment.

7. The method according to claim 6, wherein Determine the control inputs of the intruder agent at the preset number of moments through the following formula: Wherein, r k x is the state vector of the intruder agent at the k-th moment after the current moment, is the state vector of the intruder agent at the (k - 1)-th moment after the current moment, ΔT is the time interval between two adjacent moments, is a preset function, is the preset state vector corresponding to the pre-saved protected area, is the control input to be predicted for the intruder agent at the (k - 1)-th moment after the current moment, Q j,x is the third preset matrix, is the weight corresponding to the distance, is the control input to be predicted for the intruder agent at the k-th moment after the current moment, R j,h is the fourth preset matrix, is the weight corresponding to the consumed energy, is the shortest distance between the intruder agent and each defender agent at the k-th moment after the current moment, M is the set of each defender agent tracking the intruder agent, Q a is the fifth preset matrix, is the third predicted state vector of the j-th defender agent at the k-th moment after the current moment, is the state vector of the j-th defender agent at the current moment, u0 is the preset control input.

8. A target tracking device, characterized in that, The device includes: A determination and generation module, configured to identify the position information and attitude information of the defender agent and the intruder agent in the image collected at the current moment through a pre-trained model; for each agent, generate the state vector of the agent at the current moment based on the position information and attitude information of the agent; for a preset number of moments after the current moment, determine the target state vector of the defender agent at that moment based on the state vector of the defender agent at the current moment and the control input to be predicted of the defender agent at that moment, where the control input includes the instantaneous pitch angular velocity, instantaneous horizontal angular velocity, and instantaneous acceleration of the defender agent. A processing module, configured to determine the target control inputs at the preset number of moments when the distance is minimized based on the target state vectors of the defender agent at the preset number of moments, the first predicted state vectors of the intruder agent at the preset number of moments, and the first relationship between the distance between the defender agent and the intruder agent; control the defender agent based on the target control inputs at the preset number of moments.

9. An electronic device, characterized in that, The electronic device includes at least a processor and a memory. When the processor executes the computer program stored in the memory, it implements the steps of the target tracking method according to any one of the above claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by the processor, it implements the steps of the target tracking method according to any one of the above claims 1-7.

Images