A UAV control method for non-cooperative target tracking

Abstract

The present invention belongs to the field of electronic information technology and discloses a drone control method for non-cooperative target tracking: Step 1: A gimbal camera collects a real-time video stream of a target and extracts a target feature vector; Step 2: Define and calculate a normalized visual error vector, adjust the gimbal attitude based on the target feature vector obtained in Step 1, and keep the target at the center of the image; Output the gimbal yaw angle, gimbal pitch angle, and normalized visual error vector; Step 3: Based on the output of Step 2, design a tracking quality factor and a pitch angle scaling factor, dynamically adjust the drone's maximum speed to calculate a desired speed vector, and obtain the drone's speed and position at the next moment. The present invention decouples drone motion from gimbal rotation, controlling the drone's trajectory while ensuring the target remains within the drone's field of view, thereby achieving accurate and stable tracking of non-cooperative targets.

Description

Technical Field

[0001] The present invention belongs to the field of electronic information technology, and in particular relates to a drone control method for non-cooperative target tracking. Background Art

[0002] As highly maneuverable and intelligent autonomous flight platforms, drones have been widely used in fields such as battlefield reconnaissance and outdoor inspections. UAV control technology, as the core of intelligent flight systems, determines the drone's motion performance and mission execution capabilities in complex environments. For the challenging scenario of non-cooperative target tracking, traditional methods based on preset trajectories or manual manipulation are difficult to adapt to the unpredictable and highly dynamic characteristics of the target. In this context, visual servo control, a closed-loop control strategy based on real-time image feedback, captures target visual information through an onboard camera and performs feature extraction and analysis, thereby dynamically adjusting the drone's position and attitude, providing a technical foundation for autonomous tracking of non-cooperative targets.

[0003] Several currently published patents explore visual servo control. Patent No. CN119645109A proposes a decoupled system for drone and image dynamics using virtual plane feature points and image moment eigenvalues, achieving precise landing while taking field-of-view constraints into account. However, this approach primarily targets fixed scenarios, such as fixed-point landing, and struggles to predict and track fast-moving, non-cooperative targets. Patent No. CN111966133A tracks targets by adjusting the gimbal's posture, but fails to achieve coordinated optimization between the drone and gimbal. Furthermore, the coordinate solution based on perspective projection is susceptible to camera imaging principles and positioning errors. Patents No. CN113138608B and CN111984036B enhance system adaptability by introducing disturbance observers, nonlinear velocity observers, or state switching mechanisms, respectively. However, these approaches suffer from a single reliance on visual features, susceptibility of feature estimation to environmental influences, and high computational resource consumption. Furthermore, these approaches fail to fully exploit the multi-degree-of-freedom motion of the gimbal, severely limiting the system's effective tracking range.

[0004] In summary, there is an urgent need to design a control method for non-cooperative UAVs that has high accuracy and can stably track targets. Summary of the Invention

[0005] The purpose of the present invention is to provide a UAV control method for non-cooperative target tracking to solve the problems of large target tracking error and low stability in the prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for controlling a drone for non-cooperative target tracking comprises the following steps:

[0008] Step 1: The PTZ camera collects the real-time video stream of the target and extracts the target feature vector;

[0009] Step 2: Define and calculate the normalized visual error vector. Adjust the gimbal posture based on the target feature vector obtained in step 1 to keep the target in the center of the image. Output the gimbal yaw angle, gimbal pitch angle, and normalized visual error vector.

[0010] In step 3, based on the gimbal yaw angle, gimbal pitch angle, and normalized visual error vector output in step 2, the maximum speed of the drone is dynamically adjusted to calculate the desired speed vector by designing the tracking quality factor and pitch angle scaling factor. The sliding surface and position control law are defined, and acceleration constraints are applied to obtain the speed and position of the drone at the next moment.

[0011] The present invention adopts a two-stage control architecture. First, the gimbal posture is precisely adjusted based on the target visual feature information obtained by the gimbal camera. Then, the gimbal posture information is used to control the drone's motion and achieve stable tracking of non-cooperative targets. The present invention adopts the following technical solutions:

[0012] During the gimbal control phase, the visual information collected by the camera is first integrated with the drone's sensor data to extract the target feature vector. Based on this, a sliding mode control strategy is employed to achieve rapid convergence of the target visual error. A boundary layer saturation function is introduced to effectively suppress chattering, a common phenomenon in the control process. Furthermore, an intelligent dynamic gain adjustment mechanism is designed to adapt to different tracking scenarios, enabling the system to automatically adjust control parameters based on the error magnitude. Ultimately, while fully considering the physical constraints of the gimbal, the system achieves precise calculation and smooth update of the gimbal angle, laying the foundation for the next stage of drone position control.

[0013] During the drone's position control phase, the drone relies directly on the attitude information from the gimbal camera to achieve precise position updates. First, a tracking quality assessment factor is designed based on the aforementioned eigenvectors. Second, the control gain is dynamically adjusted based on the deviation between the gimbal angle and the reference value. The desired velocity vector is calculated based on the gimbal's yaw angle and the maximum velocity limit. Finally, precise flight control of the drone is achieved within the constraints of finite acceleration.

[0014] In summary, this control method processes visual errors through an optimized gimbal control algorithm, directly drives the gimbal control to maintain the target within the field of view, and indirectly guides the UAV to adjust its position accordingly to achieve stable tracking of the target.

[0015] Compared with the prior art, the present invention has the following technical effects:

[0016] (1) Fully utilizing the motion characteristics of the gimbal camera and the drone, the gimbal rotation is controlled by independently adjusting the pitch and yaw angles to achieve an expanded field of view, overcoming the limited viewing angle problem of traditional methods such as fixed cameras or virtual camera projections;

[0017] (2) By utilizing a cascaded architecture to decouple gimbal camera control from drone position control, a two-stage control method was developed in which the gimbal prioritizes visual errors and the drone motion is adjusted based on the gimbal attitude, making target tracking faster and more accurate.

[0018] (3) Combining the image plane features and the drone sensor parameters to form a four-dimensional feature vector, it overcomes the instability caused by relying solely on the gimbal image features and reduces the requirements for computing resources.

[0019] In summary, the method of the present invention decouples the UAV motion and gimbal rotation, and fully utilizes the high degree of freedom characteristics of the gimbal camera to control the UAV motion trajectory while ensuring that the target is always within the UAV's field of view, thereby achieving accurate and stable tracking of non-cooperative targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the scene;

[0021] Figure 2 It is the motion trajectory of non-cooperative target tracking;

[0022] Figure 3 This is a histogram comparing the mean values ​​of tracking distances of different algorithms;

[0023] Figure 4 is the change of the gimbal pitch angle;

[0024] Figure 5 are the visual characteristics of different controllers Change graph;

[0025] Figure 6 are the visual characteristics of different controllers Change graph.

[0026] The present invention is further explained below with reference to the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0027] To better understand the method of the present invention, the motion principle of the drone and the PTZ camera is first described here, and the PTZ camera control and the drone position control are decoupled. Figure 1 As shown in the figure, the present invention establishes a coordinate system to characterize the motion characteristics of the drone, the PTZ camera and the target position. The inertial coordinate system with the origin as the fixed global reference coordinate system; secondly, the center of mass of the drone is the origin, the coordinate axis 、 、 Represent the forward direction, lateral direction and vertical direction of the UAV respectively, and establish the UAV body coordinate system; then, use the gimbal camera body center of mass is the origin, the coordinate axis 、 、 Aligned with the gimbal-camera structure, where The axis corresponds to the optical axis direction of the gimbal camera, and the gimbal camera coordinate system is established; finally, the target image plane coordinate system is located in the imaging plane, and the origin is At the geometric center of the target image, the coordinate axis and Respectively represent the horizontal and vertical directions of the image.

[0028] The UAV control method for non-cooperative target tracking provided by the present invention specifically includes the following steps:

[0029] Step 1: The PTZ camera collects the real-time video stream of the target and extracts the target feature vector to achieve a complete description of the target state.

[0030] Specifically, the target feature vector The four components in are defined as follows:

[0031] (1) Set target feature points The coordinates on the image plane are , there are a total of target feature points; Indicates the horizontal distribution center of the target on the current image plane, and its value is equal to the horizontal coordinate of all target feature points on the current image The average value of Indicates the longitudinal distribution center of the target on the current image plane, and its value is equal to the ordinate of all target feature points The average value of:

[0032]

[0033] (2) is the flight altitude of the UAV, expressed as:

[0034]

[0035] in, is the flight altitude of the drone.

[0036] (3) is the shooting angle, that is, the pitch angle of the gimbal, expressed as:

[0037]

[0038] in, is the gimbal pitch angle, that is, the angle between the gimbal camera lens and the horizontal direction.

[0039] Step 2: Define and calculate the normalized visual error vector. Adjust the gimbal posture based on the target feature vector obtained in step 1 to keep the target in the center of the image. Output the gimbal yaw angle, gimbal pitch angle, and normalized visual error vector.

[0040] Step 2 specifically includes the following sub-steps:

[0041] Step 21: Define and calculate a normalized visual error vector. This vector is used to convert the pixel coordinate error of the image captured by the gimbal camera into a standardized error for easier control processing.

[0042] Specifically, the normalized visual error vector Expressed as:

[0043]

[0044] Where:

[0045] , —The horizontal and vertical distances between the center of the image and the image boundary;

[0046] , —The horizontal and vertical distribution centers of the target on the image plane;

[0047] In step 22, the normalized visual error vector calculated in step 21 is processed using a sliding mode control method to obtain a sliding surface for the gimbal controller. This operation can achieve rapid convergence of the visual error. A boundary layer saturation function is then calculated for the sliding surface. Step 22 can reduce chattering during control.

[0048] Specifically, the sliding surface of the PTZ controller is the visual error vector and its derivatives The linear combination of is expressed as:

[0049]

[0050] Where:

[0051] —Visual error vector;

[0052] — derivative of the visual error vector;

[0053] —Sliding mode parameters. The determination process follows the classical sliding mode control theory and is determined by analyzing the system response characteristics to adjust the convergence speed. In the present invention, .

[0054] Specifically, the boundary layer saturation function is expressed as follows:

[0055]

[0056] Where:

[0057] —Boundary layer thickness including yaw and pitch channels The component is conventionally determined by balancing the control accuracy and the degree of system chattering;

[0058] —Sliding surface of the PTZ controller The weight.

[0059] pass Realize continuous change of control signal within boundary layer thickness;

[0060] —Sign function. To achieve the guarantee of convergence speed beyond the boundary layer thickness, each degree of freedom is processed independently to generate the gimbal control command.

[0061] Step 23: Design a dynamic gain adjustment mechanism to obtain a control gain output after dynamic gain adjustment.

[0062] Specifically, the control gain output after dynamic gain adjustment is Expressed as:

[0063]

[0064] Where:

[0065] —Basic control gain;

[0066] —normalized visual error vector at the current moment;

[0067] —Fine control gain coefficient;

[0068] —Error sensitivity parameter;

[0069] —Fine control of threshold;

[0070] —Basic gain coefficient;

[0071] —rough adjustment factor;

[0072] — Rough control of the threshold.

[0073] Preferably, in the present invention, , , , , , , .

[0074] This design increases the control gain to improve accuracy when the error is small, and adjusts the gain moderately to ensure smooth convergence when the error is large.

[0075] Step 24: Introduce the integral term into the control law of the gimbal camera to obtain a new control law. This operation can eliminate the steady-state error and keep the target at the center of the image. Specifically, the control law is expressed as:

[0076]

[0077] Where:

[0078] —control laws;

[0079] 、 —Angular velocity of gimbal yaw and pitch angles;

[0080] 、 —Dynamic gains of yaw and pitch channels;

[0081] 、 —Sliding surface of the PTZ controller Components of the yaw and pitch channels of the variables;

[0082] 、 — boundary layer thickness in yaw and pitch channels;

[0083] —Integral gain, determined according to the system steady-state response characteristics , used to eliminate long-term deviations;

[0084] 、 —Normalized visual error components of the yaw and pitch channels.

[0085] Therefore, during the sampling period The update equations for the gimbal yaw angle and gimbal pitch angle are:

[0086]

[0087]

[0088] Where:

[0089] 、 —The gimbal yaw angle and gimbal pitch angle at the next moment;

[0090] 、 —The current gimbal yaw angle and gimbal pitch angle;

[0091] 、 —Angular velocity of the gimbal’s yaw and pitch channels at the current moment;

[0092] 、 —The control law of the gimbal’s yaw and pitch channels at the current moment.

[0093] Preferably, in order to prevent the yaw angle and pitch angle of the gimbal from exceeding the physical limit and to protect the mechanical structure of the gimbal, angle constraints are introduced, specifically:

[0094]

[0095]

[0096] Where:

[0097] 、 —The physical limit of the gimbal yaw angle;

[0098] 、 —The physical limit of the gimbal pitch angle;

[0099] clip()—Standard clipping function. This function is used to achieve the gimbal yaw angle and gimbal pitch angle respectively. , In the present invention, the controllable rotation ranges of the yaw angle and the pitch angle of the gimbal are respectively , .

[0100] In step 3, based on the gimbal yaw angle, gimbal pitch angle, and normalized visual error vector output in step 2, the maximum speed of the drone is dynamically adjusted to calculate the desired speed vector by designing the tracking quality factor and pitch angle scaling factor. The sliding surface and position control law are defined, and acceleration constraints are applied to obtain the speed and position of the drone at the next moment.

[0101] Step 3 specifically includes the following sub-steps:

[0102] Step 31: Combine the normalized visual error vector Design Tracking Quality Factor :

[0103] .

[0104] Step 32: Based on the tracking quality factor and the gimbal pitch angle to define the pitch angle scaling factor :

[0105] .

[0106] Where:

[0107] —Pitch angle of the gimbal. When the gimbal pitch angle deviates far from the ideal state, the pitch angle proportional factor The setting can prevent the UAV from losing its target or exceeding the dynamic constraints due to violent movement;

[0108] —Reference pitch angle. Indicates the ideal pitch angle of the camera. In this invention, ;

[0109] - pitch angle normalization parameter, the present invention determines its value through experimental debugging, , used to map the angular difference to an appropriate scale factor;

[0110] Step 33, by tracking the quality factor and pitch angle scaling factor Dynamically adjust the maximum speed of the drone :

[0111] .

[0112] Where:

[0113] —The speed reference value in normal tracking mode can be set according to the general performance parameters of the drone.

[0114] From the above, based on the tracking quality factor and pitch angle scaling factor , when the target deviates from the center of the image, according to , we can get The value decreases, so The value decreases accordingly, thereby achieving the deceleration of the drone, giving priority to controlling the gimbal to ensure that the target is within the field of view. When the target is close to the center of the image, , the drone achieves full-speed tracking.

[0115] The method of the present invention automatically reduces the speed of the UAV when the target deviates from the image center to avoid loss of vision; when the target is stable, the control of the UAV is dynamically enhanced according to the difference between the gimbal yaw angle, the gimbal pitch angle and the expected value to enhance the trajectory tracking efficiency.

[0116] Step 34, based on the gimbal yaw angle and the maximum speed of the drone Get the desired velocity vector of the drone :

[0117] .

[0118] Where:

[0119] —The desired velocity vector of the UAV;

[0120] —Desired altitude of the drone;

[0121] - The flight altitude of the drone;

[0122] —The maximum vertical speed limit of the drone.

[0123] The operation of step 34 can ensure that the drone moves along the direction pointed by the camera.

[0124] Step 35, define the sliding surface of the UAV. Specifically, the sliding surface of the UAV is defined as the velocity error and position error The combination is expressed as:

[0125] .

[0126] Among them, the speed error is the expected speed and the current speed v The deviation is expressed as:

[0127] .

[0128] Is the adjustment position error The weight of is determined according to the experimental environment and actual physical constraints.

[0129] Since the horizontal movement direction of the UAV strictly follows the direction of the gimbal, the position error It is only necessary to express the position error in the height direction, which can be expressed as:

[0130] .

[0131] Step 36: Design the position control law and acceleration constraints.

[0132] Specifically, the position control law, as the core link between target tracking requirements and the UAV power system, is expressed in the present invention as:

[0133]

[0134]

[0135] Where:

[0136] is the position control law;

[0137] - The thickness of the boundary layer of the position control is used to adjust the smoothness of the control instruction; determined by system dynamic performance analysis and simulation adjustment, preferably, in the present invention .

[0138] —Position control gain. Determined through system dynamic performance analysis and simulation adjustment, where 、 and Used to realize drone x axis, y axis, z Independent adjustment of the axis control strength, preferably, in the present invention .

[0139] Step 37: Design acceleration constraints.

[0140] Specifically, calculate the expected acceleration of the drone :

[0141]

[0142] Considering the dynamic constraints of the UAV, the expected acceleration is limited to obtain the actual acceleration of the UAV :

[0143]

[0144] Where: It is the maximum acceleration limit of the UAV in all directions, which is determined by the dynamic performance.

[0145] Step 38, during the sampling period According to the current speed of the drone and acceleration , get the speed and position of the UAV at the next moment, and update the UAV motion state.

[0146] Specifically, the speed of the drone at the next moment :

[0147] ;

[0148] To ensure that the speed does not exceed the safety limit, a speed constraint is introduced to update the speed of the drone at the next moment:

[0149]

[0150] in, is the maximum speed in each direction.

[0151] Finally, according to the speed of the drone at the current moment Define the position of the drone at the next moment :

[0152]

[0153] in, They represent the three-dimensional position of the center of mass of the drone in the simulation environment at the current moment and the next moment respectively.

[0154] In order to verify the effect of the method given in this invention, a three-dimensional simulation environment was established, and a cooling period mechanism, state perception and motion constraints were introduced to construct a non-cooperative target model with active escape capability, so as to simulate the active escape behavior of the target in a complex dynamic environment. Finally, the application effect of this method was verified through multi-dimensional quantitative analysis and comparative experiments. Traditional model predictive control (MPC), proportional-integral-derivative (PID) control and deep reinforcement learning (DRL) control were selected as benchmark comparison algorithms. The initial experimental conditions set the drone reference coordinates to , the yaw azimuth and pitch angle are , the initial target position is , the initial velocities of the UAV and the non-cooperative target are both zero.

[0155] The escape model of non-cooperative targets is constructed as follows:

[0156] The target's escape strategy mainly includes direction maintenance during the cooling period, dynamic escape direction selection during the non-cooling period, and motion smoothness constraints.

[0157] The cooling mechanism is implemented through a cooling timer Controls how often the target's escape direction is updated. Cooldown timer changes with time step. Linearly decreasing, its update formula is:

[0158]

[0159] When the cooldown timer When the target maintains the escape direction of the previous moment , the target will not recalculate the escape direction, thus avoiding frequent direction switching. When the cooldown timer reaches zero ( ), the target recalculates the ideal escape direction based on the current relative position and speed.

[0160] Set the target's center of mass to the position in the world coordinate system ,speed and acceleration , so the relative position and velocity between the target and the UAV in the xy two-dimensional plane are:

[0161]

[0162]

[0163] Target's escape direction It is dynamically calculated based on these relative state quantities. and relative speed , the calculation of the ideal escape direction is divided into the following cases:

[0164] when When , the target moves away from the UAV, and the ideal escape direction is:

[0165]

[0166] when When , to avoid singularity, the target selects a random unit vector as the ideal escape direction:

[0167]

[0168] In general ( ), the strategy evaluates the relative speed Projection in the direction of normalized position difference:

[0169] .

[0170] if (in is the approach judgment threshold), indicating that the target is approaching quickly, at this time the vertical escape mechanism is triggered and the cooling mechanism is reset, i.e. The calculation of the vertical escape direction depends on the current speed state of the UAV. When approaching a standstill, the ideal escape direction is the direction perpendicular to the position difference:

[0171]

[0172] Otherwise, construct a vertical vector based on the drone's velocity direction, and the ideal escape direction is:

[0173]

[0174] After obtaining the ideal escape direction, the acceleration direction of the target is consistent with the ideal escape direction:

[0175]

[0176] in is the maximum two-dimensional acceleration of the target. The target velocity update follows the dynamic equation:

[0177]

[0178] and limited to a maximum speed :

[0179]

[0180] The target's position is updated as:

[0181]

[0182] The final target's velocity and position are updated as:

[0183]

[0184] The experimental results are as follows Figures 2 to 6 shown.

[0185] Figure 2 The motion trajectories of all control methods for tracking non-cooperative targets are shown when the UAV is initially 50 meters away from the target in the horizontal direction. It can be seen that the method proposed in the present invention has better tracking performance than other control methods.

[0186] Figure 3 A quantitative evaluation of tracking performance was conducted by calculating the mean difference between the two-dimensional trajectories of the drone and the target in the xoy plane. The proposed method achieved an average position error of 5.72 meters, an improvement of 30.23%, 57.51%, and 17.28% over the MPC, PID, and DRL algorithms, respectively. These results demonstrate that the proposed method effectively achieves stable tracking of non-cooperative targets.

[0187] Table 1 lists the average processing time of all control methods. The proposed method takes 3.6×10 -4 s, which is only 2.10% of the MPC method, 85.71% of the PID algorithm, and 17.31% of the DRL method, showing a significant advantage in response speed.

[0188] Table 1 Comparison of response time of different algorithms

[0189]

[0190] Figures 4 to 6 is the change of visual feature error and gimbal parameters during tracking, where Figure 4 The changes of the gimbal pitch angle when the proposed method tracks non-cooperative targets are shown. Figure 5 and Figure 6 The ratio of the target's deviation from the image center in the horizontal and vertical directions under all control methods is shown. It can be seen that the method of the present invention can achieve rapid adjustment of the gimbal posture and better ensure that the target object is close to the center of the field of view, showing stronger stability and robustness.

Claims

1. A UAV control method for non-cooperative target tracking, characterized in that: The specific steps include: Step 1: The PTZ camera collects the real-time video stream of the target and extracts the target feature vector ,in, Indicates the horizontal distribution center of the target on the current image plane, Indicates the longitudinal distribution center of the target on the current image plane, is the flight altitude of the drone, is the shooting angle, i.e. the pitch angle of the gimbal; Step 2: Define and calculate the normalized visual error vector. Adjust the gimbal posture based on the target feature vector obtained in step 1 to keep the target in the center of the image. Output the gimbal yaw angle, gimbal pitch angle, and normalized visual error vector. The specific steps include the following: Step 21, defining and calculating a normalized visual error vector; Step 22: Processing the normalized visual error vector calculated in step 21 using a sliding mode control method to obtain a sliding surface of the pan / tilt controller; then calculating a boundary layer saturation function for the sliding surface; the sliding surface of the pan / tilt controller is a linear combination of the visual error vector and its derivative; Step 23, designing a dynamic gain adjustment mechanism to obtain a control gain output after dynamic gain adjustment; Step 24, introducing an integral term into the control law of the gimbal camera to obtain a new control law to keep the target at the center of the image; Step 3: Based on the gimbal yaw angle, gimbal pitch angle, and normalized visual error vector output from step 2, the desired velocity vector is calculated by dynamically adjusting the maximum velocity of the drone by designing the tracking quality factor and pitch angle scaling factor. The sliding surface and position control law are defined, and acceleration constraints are applied to obtain the velocity and position of the drone at the next moment. The specific steps include the following: Step 31, designing a tracking quality factor based on the normalized visual error vector; Step 32, defining a pitch angle scaling factor according to the tracking quality factor and the gimbal pitch angle; Step 33, dynamically adjust the maximum speed of the drone by tracking the quality factor and the pitch angle scale factor; Step 34, obtaining the desired velocity vector of the UAV based on the gimbal yaw angle and the maximum velocity of the UAV; Step 35, defining the sliding surface of the UAV as a combination of velocity error and position error; Step 36, designing position control law and acceleration constraints; Step 37, design acceleration constraints; Step 38: During the sampling period, the speed and position of the drone at the next moment are obtained based on the speed and acceleration of the drone at the current moment, so as to update the motion state of the drone.

2. The UAV control method for non-cooperative target tracking according to claim 1, wherein: In step 1, the target feature vector The four components in are defined as follows: (1) Set target feature points The coordinates on the image plane are , there are a total of target feature points; Indicates the horizontal distribution center of the target on the current image plane, and its value is equal to the horizontal coordinate of all target feature points on the current image The average value of Indicates the longitudinal distribution center of the target on the current image plane, and its value is equal to the ordinate of all target feature points The average value of: (2) is the flight altitude of the UAV, expressed as: in, is the flight altitude of the UAV; (3) is the shooting angle, that is, the pitch angle of the gimbal, expressed as: in, is the gimbal pitch angle.

3. The UAV control method for non-cooperative target tracking according to claim 2, wherein: In step 21, the normalized visual error vector : Where: , —The horizontal and vertical distances between the center of the image and the image boundary; , —The horizontal and vertical distribution centers of the target on the image plane; In step 22, the sliding surface of the pan / tilt controller Expressed as: Where: —Visual error vector; — derivative of the visual error vector; — sliding mode parameters, ; The boundary layer saturation function is expressed as follows: Where: —Boundary layer thickness including yaw and pitch channels The component is conventionally determined by balancing the control accuracy and the degree of system chattering; —Sliding surface of the PTZ controller The weight; pass Realize continuous change of control signal within boundary layer thickness; —Symbol function; In step 23, the control gain output after the dynamic gain adjustment is Expressed as: Where: —Basic control gain; —normalized visual error vector at the current moment; —Fine control gain coefficient; —Error sensitivity parameter; —Fine control of threshold; —Basic gain coefficient; —rough adjustment coefficient; — Rough control threshold; In step 24, the control law is expressed as: Where: —control laws; 、 —Angular velocity of gimbal yaw and pitch angles; 、 —Dynamic gains of yaw and pitch channels; 、 —Sliding surface of the PTZ controller Components of the yaw and pitch channels of the variables; 、 — boundary layer thickness in yaw and pitch channels; —Integral gain, ; 、 —normalized visual error components of yaw and pitch channels; In the sampling period The update equations for the gimbal yaw angle and gimbal pitch angle are: Where: 、 —The gimbal yaw angle and gimbal pitch angle at the next moment; 、 —The current gimbal yaw angle and gimbal pitch angle; 、 —Angular velocity of the gimbal’s yaw and pitch channels at the current moment; 、 —The control law of the gimbal’s yaw and pitch channels at the current moment.

4. The UAV control method for non-cooperative target tracking according to claim 3, wherein: In step 24, angle constraints are introduced, specifically: Where: 、 —The physical limit of the gimbal yaw angle; 、 —The physical limit of the gimbal pitch angle; clip()—Standard clipping function.

5. The UAV control method for non-cooperative target tracking according to claim 4, characterized in that: In step 31, the normalized visual error vector Design Tracking Quality Factor : ； In step 32, according to the tracking quality factor and the gimbal pitch angle to define the pitch angle scaling factor : Where: —Pitch angle of the gimbal; —reference pitch angle, ; — normalized pitch angle parameter, ; In step 33, by tracking the quality factor and pitch angle scaling factor Dynamically adjust the maximum speed of the drone : Where: —Speed ​​reference value in normal tracking mode; In step 34, based on the gimbal yaw angle and the maximum speed of the drone Get the desired velocity vector of the drone : Where: —The desired velocity vector of the UAV; —Desired altitude of the drone; - The flight altitude of the drone; — Maximum vertical speed limit of the drone; In step 35, the sliding surface of the drone is defined as the velocity error and position error The combination is expressed as: Among them, the speed error is the expected speed and the current speed v The deviation is expressed as: Where: —Adjust position error The weight of Since the horizontal movement direction of the UAV strictly follows the direction of the gimbal, the position error Only the position error in the height direction needs to be expressed as: In step 36, the position control law is expressed as: Where: is the position control law; — position-controlled boundary layer thickness, ; —Position control gain, ; In step 37, the acceleration constraint is specifically designed as follows: Calculate the expected acceleration of the drone : Get the actual acceleration of the drone : Where: It is the maximum acceleration limit of the drone in each direction; In step 38, the speed of the drone at the next moment : ； Introduce the speed constraint to update the speed of the drone at the next moment: in, is the maximum speed in each direction; Finally, according to the speed of the drone at the current moment Define the position of the drone at the next moment : in, They represent the three-dimensional position of the center of mass of the drone in the simulation environment at the current moment and the next moment respectively.

Images