Lift wing unmanned aerial vehicle high maneuvering target interception method based on visual servo

Abstract

The invention provides a lift wing unmanned aerial vehicle high maneuvering target interception method based on visual servo, which is used for using a lift wing unmanned aerial vehicle to cope with an invasion target moving at a high speed and using a lift wing platform. Lift wing modeling based on a rotation matrix and high-precision and high-dynamic autonomous interception control based on visual servo are introduced to serve as key technologies for supporting interception of the lift wing unmanned aerial vehicle. The effectiveness of the algorithm under airborne perception and control is verified, and the feasibility of the algorithm in practical application is consolidated. The contributions jointly promote the development of an unmanned aerial vehicle autonomous interception technology, and an important tool and method are provided for effectively coping with long-distance high-speed target intrusion threats. In addition, an IMU (Inertial Measurement Unit) measurement model and an image delay measurement model continue to be used, and a delay state estimation problem is researched.

Description

Technical Field

[0001] The present invention relates to the key technologies for autonomous interception of lift-wing unmanned aerial vehicles (UAVs), including lift-wing UAV control, image-based visual servoing, autonomous interception, and anti-UAV systems. In particular, a control allocation strategy for lift-wing UAV control is introduced to solve the autonomous interception task of lift-wing UAVs. This covers fields such as lift-wing UAV control, image-based visual servoing, autonomous interception, and anti-UAVs, aiming to apply lift-wing UAVs for rental interception tasks. Background Art

[0002] In recent years, UAVs, popularly known as flying vehicles, have developed rapidly in various fields such as reconnaissance, photography, and entertainment. However, the threat posed by unauthorized or malicious flying vehicles to public safety has become increasingly prominent. In response to this problem, the interception of flying vehicles has become crucial in the low-altitude safety field.

[0003] UAV interceptors can be divided into radio frequency interference and hard-kill malicious flying vehicles. However, due to the small size and high flexibility of UAVs, the success rate of radio frequency interference is low, and it is a challenging task to quickly deploy and intercept intruders using ground equipment. One way to solve this problem is to use intercept UAVs that rely on on-board sensors to detect and force intruders to land.

[0004] Compared with traditional quadrotor UAVs, lift-wing UAVs are faster in flight speed, have a longer endurance, and have higher anti-interference ability while taking into account flexibility, so they are more suitable for performing interception tasks.

[0005] In previous work, the applicant developed a high-precision and high-dynamic interception algorithm based on the image of a forward monocular camera for quadrotor UAVs (IBVS) and applied it to real flight experiments at 10 m / s. However, due to the use of quadrotor aircraft, the flight range is limited, and in real flight tests, only targets within 4 km can be intercepted and returned, and it is easy to be damaged itself after colliding with a quadrotor UAV. Therefore, in contrast to designing an interception control algorithm for lift-wing, to achieve this goal, the present invention mainly presents two main challenges:

[0006] 1. The high-maneuver flight of lift-wing UAVs requires establishing a complete attitude model, and it is necessary to perform dynamic modeling on lift-wing UAVs for interception control so that its rotation matrix is continuous on the three-dimensional special orthogonal group, denoted as SO(3).

[0007] 2. Lift-wing has its unique dynamic characteristics, and it is necessary to redesign the interception controller and design a visual servo interception algorithm based on lift-wing UAVs to meet the requirements of the lift-wing platform.

[0008] Considering these challenges comprehensively, successfully achieving the lift-wing UAV interception mission requires focusing on how to adapt to the lift-wing platform and improve interception performance. This involves in-depth research in multiple fields such as autonomous control, computer vision, machine learning, and optimization algorithms to ensure the efficient and successful execution of the lift-wing interception mission. Summary of the Invention

[0009] The present invention proposes a high-maneuver target interception method for lift-wing UAVs based on visual servoing, which is used to use lift-wing UAVs to deal with high-speed moving intrusion targets and use the lift-wing platform. The present invention introduces lift-wing modeling based on the rotation matrix and high-precision and high-dynamic autonomous interception control based on visual servoing as the key technologies to support the interception of lift-wing UAVs. The present invention verifies the effectiveness of the algorithm under onboard perception and control, and consolidates its feasibility in practical applications. These contributions jointly promote the development of UAV autonomous interception technology and provide important tools and methods for effectively dealing with the threat of long-distance high-speed target intrusion.

[0010] The present invention first introduces the six-degree-of-freedom rigid body model, camera imaging model, and visual servo model of the lift-wing aircraft related to the present invention. Using these models, the present invention models the high-precision and high-dynamic autonomous interception control problem. In addition, the present invention also follows the IMU measurement model and image delay measurement model and studies the problem of delay state estimation.

[0011] Coordinate systems: For the interceptor aircraft, the present invention uses four coordinate systems, as Figure 1 and Figure 2 shown, including:

[0012] · {e} = {o e -x e y e z e} is the Earth-fixed coordinate system (EFCS), which is used to represent the positions and velocities of the interceptor and the target in the global inertial system;

[0013] · {b} = {o b -x b y b z b} is the body coordinate system (BCS), which is used to represent the variables of the current attitude of the four-rotor body of the lift-wing interceptor relative to the ground;

[0014] · {l} = {o l -x l y l z l} is the lift-wing coordinate system (LCS), which is used to represent the variables of the current attitude of the wing of the lift-wing interceptor relative to the ground;

[0015] · {w} = {ow -x w y w z w} is the airflow coordinate system (WCS), a variable used to represent the current attitude of the lift-wing interceptor relative to the airflow;

[0016] · {c} = {o c -x c y c z c} is the camera coordinate system (CCS);

[0017] · {i} = {o i -x i y i} is the image coordinate system (ICS), used to represent the image position of the target feature in the first-person view of the camera;

[0018] Matrix represents the rotation matrix from the coordinate system to where the three-dimensional special orthogonal group is given by the following, where R represents the transpose of R, and I represents the identity matrix: T represents the transpose of R, and I represents the identity matrix:

[0019]

[0020] Lift-wing interceptor model: The lift-wing UAV (interceptor) is modeled as a rigid body with mass m, and its flight control rigid model is summarized as:

[0021]

[0022] where:

[0023]

[0024] where Q represents the aerodynamic pressure of the aircraft in the cruise state, S represents the wing area, and respectively represent the coordinate rotation matrices from the airflow coordinate system to the body coordinate system, from the body coordinate system to the Earth-fixed coordinate system, and from the lift-wing coordinate system to the airframe coordinate system. b and c respectively represent the lift-wing span and the mean chord length, and respectively represent the velocity v and the derivative of the rotation matrix, represents the position of the aircraft under the EFCS; represents the velocity vector under the EFCS; is the acceleration under the EFCS; m is the mass of the aircraft; g is the local acceleration due to gravity, g = [0 0 g] T ; bf = [0 f ry -f rz T is the controllable force of the lift wing motor in the BCS; is the equivalent aerodynamic rudder surface deflection angle of the lift wing in the BCS, [δ ar δ al T is the left and right aileron deflections of the lift wing in the BCS; C D 、C Y 、C L are the aerodynamic force coefficients of the lift wing surface in the WCS; is the aerodynamic force coefficient of the equivalent aerodynamic rudder surface in the WCS; C l 、C m 、C n are the aerodynamic moment coefficients of the lift wing surface in the WCS; is the aerodynamic moment coefficient of the equivalent aerodynamic rudder surface in the WCS; is the angular velocity under the BCS; b m is the total moment in the body frame; J is the multi-rotor inertia; τ = [m rx m ry m rz T is the controllable moment generated by the propellers and ailerons on the body axis. The matrix b ω] × is an anti-symmetric matrix, which is the cross product mapping of the angular velocity b ω.

[0025] The cross product mapping [·] × : is defined as follows. For any (three-dimensional vector), there is [x]×y = x×y, and the inverse mapping of the cross product mapping is represented by the vex mapping vex(·): where the set of 3×3 anti-symmetric matrices For the cross product mapping and the vex mapping can be summarized as:

[0026]

[0027] For the lift wing aircraft, the controllable force and moment are generated by the propellers and aerodynamic rudder surfaces, as shown below:

[0028]

[0029] where η represents the motor tilt angle, d y represents the distance from the motor to the x-axis of the body frame, ​​​is the aerodynamic moment coefficient of the equivalent aerodynamic control surface in the WCS, κ represents the angle between the lifting wing and the motor plane, as Figure 2 shown, K y and K z represent the moment coefficients of the motor in the y-axis and z-axis directions respectively, T 1 , T 2 , T 3 , T 4 are the thrust magnitudes of the four motors.

[0030] Target model: The motion of the target model is represented using a particle model, as follows:

[0031]

[0032] where, and represent the position, velocity, and thrust acceleration of the target under the EFCS respectively; the noise is a Gaussian random variable with zero mean.

[0033] Using e p r to indicate the relative position of the interceptor and the target in the EFCS, e p r = e p - e p t . In addition, the relative velocity is defined as e v r = e v - e v t , the relative acceleration is defined as e a r = e a - e a t . Note that it is challenging to measure the motion acceleration e a t of the target using a vision sensor. Therefore, in the design of the controller and the state observer, e a t is regarded as a perturbation acting on them.

[0034] Camera imaging model: As Figure 1 shown, perspective projection performs the mapping from 3D space to the image plane: Assume that the origin of the CCS coincides with the origin of the BCS. That is, their translation vector and the rotation matrix is a constant. The camera used in the present invention is integrated with the airframe. When the camera contacts the target, it is regarded as a successful interception. Therefore, there is a constant rotation between the optical axis and the direction of the aircraft head

[0035] Under the conditions of large maneuvers and high speeds, it is necessary to re-describe the interception problem, such as Figure 1 shown. e p and e p t represent the position of the interceptor and the position of the target respectively; are the coordinates of the target; n t represents the unit vector of the target along the line of sight in the EFCS; n c represents the unit vector of the optical axis in the EFCS; n td is defined as the unit vector designed in the EFCS, called the designed unit vector, called the designed LOS vector; i p td is the intersection point with the image plane in the ICS, called the interception point. The interception problem can be summarized as n t and n td are collinear, making e p and e p t approach each other.

[0036] When the target is within the FOV of the sensor, all the above types of sensors can be converted into perspective views through a unified imaging model. The vector n t can be expressed as the unit vector in the negative direction of p r , and can also be obtained through the image coordinates and the focal length f oc :

[0037]

[0038] FOV is the range of the observable area where the sensing sensor can image, including the horizontal field of view (HFOV) and the vertical field of view (VFOV). The parameters α hfov , α vfov are the angles of the HFOV and VFOV of the sensing sensor. Specifically, the FOV values of ordinary cameras and solid-state lidars are limited; the FOV value of a pan-tilt camera is constantly changing; panoramic cameras and omnidirectional lidars have a 360-degree FOV. FOV is defined as where, is the coordinate of x in the CCS. Camera imaging is restricted by the FOV, that is, the target is within the FOV of the sensor,

[0039] Visual servo (IBVS) model: The image Jacobian matrix can represent the relationship between camera motion and image feature motion as:

[0040]

[0041] Among them, represents the velocity and angular velocity quantities in CCS, including the linear velocity and angular velocity of a rigid body; represents the normalized image coordinates of features in ICS, and represents the z - coordinate of the feature point in CCS.

[0042] The rigid - body model of the aircraft and the camera imaging model describe the motion and perception of the intercepting aircraft. For the multi - rotor interception process under large - maneuver and high - speed conditions, the delay amounts of the IMU and the image also need to be considered.

[0043] System model: The logarithmic mapping ln(·): converts the rotation matrix into a rotation angle θ∈[-π,π) and a unit rotation vector That is,

[0044]

[0045] The rotation matrix of the interceptor is tightly represented by a four - dimensional vector called the direction quaternion Regarding the normalized image coordinates of the target center as feature points, denoted as State variables consist of some necessary states, including the direction quaternion q of the interceptor, the relative position p r and the relative velocity v r , the image feature point gyroscope bias b gyr =[b gyr,x b gyr,y b gyr,z T , the accelerometer bias b acc =[b acc,x b acc,y b acc,z T .

[0046] The discrete state equation is written as:

[0047] x k =F k x k-1 +G k w k-1 (7)

[0048] Among them, and are from time t k-1 to time t k ​​The state transition matrix and the noise matrix. w k-1 is the process noise vector at time t k-1 , x k and x k-1 are the state variables at time t k and time t k-1 respectively, where the elements follow a Gaussian distribution with zero mean and are uncorrelated with each other. The vector w k is composed of the gyroscope bias noise and the accelerometer bias noise by .

[0049] IMU measurement model: The IMU (Inertial Measurement Unit) provides measurements of the gyroscope angular increment and the accelerometer velocity increment under the BCS. Combining these measurements with the rigid body model of the aircraft, the recursive expressions for the states q and v r are obtained.

[0050] · The derivative of the direction quaternion q is the angular velocity b ω, obtained from the gyroscope and measured as:

[0051] b ω = ω gyr - b gyr - n gyr (8)

[0052] where the gyroscope measurement noise n gyr is a Gaussian random variable with a mean of 0;

[0053] The bias b gyr follows a Wiener process as follows:

[0054]

[0055] · The derivative of the velocity v r , i.e., the acceleration e a, can be obtained from the accelerometer measurement as:

[0056]

[0057] where e 3 = [0 0 1] T , a acc represents the accelerometer measurement, the accelerometer measurement noise n acc is a Gaussian random variable with zero mean; the accelerometer bias b acc follows a Wiener process,

[0058]

[0059] Image delay measurement model: Image processing provides the image coordinates of the target point, and the state is obtained after normalization Since it takes time for camera imaging and image processing, the measured value obtained is data from a period ago. Assume that the delay based on the IMU update frequency is D cycles, and the delay time t D >0 is known. Then the image measurement value at time t is expressed as:

[0060]

[0061] where, represents the state at k-D time, and the image measurement noise n img is a Gaussian random variable with zero mean. Note that the delay time t of this aircraft D includes the camera imaging and image processing time, and two methods are used for estimation: ① Direct calculation based on camera parameters and internal program timer; ② Experimental measurement using the camera capture timer.

[0062] According to the camera fixed connection assumption mentioned above, assume that the multi-rotor state and image measurement are accurate and available. At the same time, the positions of the target and the interceptor in the inertial system are also unknown. For the interceptor model (1), design f, b ω controller to make the image tracking error within the set range, that is The interceptor continuously approaches the target, that is p r →0.

[0063] The present invention proposes a method based on visual servo and delayed Kalman filter to solve high-precision and high-dynamic autonomous interception. The flowchart is as Figure 2 shown. On the basis of the previous definition, the implementation steps are as follows:[[]]

[0064] Step 1: Controller based on visual servo

[0065] In step 1, the controller based on visual servo comprehensively calculates the interception control amount according to the collected image information and the aircraft's own state. The aircraft model, target model, and camera imaging model together constitute the mathematical model of step 1.

[0066] S11. Input image measurement Relative velocity v r and its own attitude

[0067] S12. Initialize a d and R d according to the current state of the interceptor, and a d and R d are the acceleration and attitude control commands respectively, specifically

[0068] S13. Calculate n based on the camera imaging model and the image features estimated in Step 2 Calculate n t , specifically as follows:

[0069] S14. During interception, according to the formula:

[0070]

[0071]

[0072] Calculate the attitude controller command, f d represents the thrust control command, e f drag represents the drag vector under EFCS, b ω 2 represents the angular velocity control command for the desired attitude transformation in the body frame.

[0073] S15. Determine whether the image features are updated. If they have been updated, jump to Step S16; if not, jump to Step S18.

[0074] S16. According to,

[0075]

[0076] a d =-k 2 v r -k 3 z 1 + e a t (16)

[0077]

[0078] Calculate the target collinearity controller command ω 1 and the attitude control set value a d , R d , where v r is the relative velocity between the interceptor and the target, is the target tracking error, e a t represents the acceleration of the target under EFCS, I is the identity matrix, r is the thrust axis, φ is the angle between the current thrust direction and the desired thrust, k 1 , k 2 , k 3 are constant coefficients that can be adjusted according to the control effect.

[0079] S17. Composite control angular velocity command ω d is:

[0080] ω d = ω d + ω 1 (18)

[0081] S18. Composite control torque command m d is:

[0082]

[0083]

[0084] where J represents the moment of inertia, represents the proportionality coefficient, represents the integral coefficient, represents the proportionality coefficient, m d,1 represents the torque control component generated by the integral term, represents the saturation value of the integral term, represents the saturation value of the total torque command.

[0085] S19. If the interception mission is not over, jump back to step S14 after T time.

[0086] Step 2: Observer based on delayed Kalman filter

[0087] The aircraft model, target model, camera imaging model, and IBVS model together constitute the system model, IMU measurement model, and image delay measurement model, which are used for state prediction and measurement update of the delayed Kalman filter.

[0088] S21. Input the measurement values of the IMU: the angular velocity ω measured by the gyroscope gyr and the acceleration a measured by the accelerometer acc , and the image measurement output by the detection module

[0089] S22. Initialization

[0090] S23. Determine whether the image measurement of the detection module is updated. If it is updated, jump to step 24; if not, jump to step S25.

[0091] S24. According to the formula:

[0092]

[0093] Use the IMU measurement to recursively calculate the state, where represents t k-1State estimation at a moment denotes the state estimation at time t k-1 obtained by one-step recursion of the state estimation at time t k , F k and G k respectively denote the state transition matrix and the noise matrix from time t k-1 to time t k , P k1 denotes the covariance matrix of the state at time t k-1 , P k|k1 is the covariance matrix estimation at time t k-1 obtained by one-step recursion of the covariance matrix at time t k , Q k denotes the noise matrix of the recursion process.

[0094] S25. According to the formula:

[0095]

[0096] P k-D = I - KH P k-D|k-D-1 (24) K = P k-D|k-D-1 H T S -1 , S = HP k-D|k-D-1 H T + R (25)

[0097] Estimate the state at time t kD where z denotes the characteristic observation value at time t kD , H represents the observation matrix, k-D the one-step recursive state estimation, K represents the covariance fusion matrix, denotes the final state estimation at time t after fusing the residuals, and P k-D denotes the final covariance estimation at time t kD after fusing the residuals. k-D

[0098] S26. According to formulas (18) and (19), recursively backward the state, and repeat step S24 until the delayed state estimation is updated to the current moment.

[0099] S27. If the interception mission has not ended, jump to step S21.

[0100] The advantages and beneficial effects of the present invention are as follows:

[0101] The present invention introduces autonomous interception control of a lift-wing unmanned aerial vehicle (UAV) based on visual servoing as a key technology to support interception. The present invention verifies the effectiveness of the algorithm under airborne perception and control, consolidating its feasibility in practical applications. These contributions jointly promote the development of long-range high-dynamic autonomous interception technology for moving targets, providing important tools and methods for effectively coping with the threat of long-distance high-speed target intrusion. Description of the Drawings

[0102] Figure 1 It is a schematic diagram of a coordinate system and an imaging model.

[0103] Figure 2 It is the definition of the relationship between the body coordinate system and the lift-wing coordinate system and the layout angle.

[0104] Figure 3 It is a high-dynamic autonomous interception flowchart of a lift-wing based on visual servoing and delayed Kalman filtering.

[0105] Figure 4a It is a flight trajectory diagram of an aircraft during an interception process.

[0106] Figure 4b It is a comparison diagram of image observation values and measurement values under the action of delayed Kalman filtering. Detailed Implementation Manner

[0107] The technical solution of the present invention will be further described below in conjunction with the drawings and embodiments.

[0108] Embodiment: There is 1 UAV and 1 target in an unknown environment, and target estimation and autonomous interception control need to be completed. The implementation steps are as follows:

[0109] Step 1: Controller based on visual servoing

[0110] Flight experiment design: In the high-precision high-dynamic autonomous interception experiment of a moving target, the target makes a hovering motion at a speed of 7 m / s, and its motion is affected by the wind. The interceptor tracks and approaches the target with irregular motion, keeps it within the field of view until the interception is completed. The motion trajectory of the interceptor is shown in Figure 4a , showing that the multi-rotor can effectively lock the target and move flexibly, with a maximum speed of up to 12 m / s. Among them, the control quantity and calculation process at the moment of t = 3.21 s are as follows:

[0111] S11. Input image measurement, the pixel coordinates of the target are i p = [164 274] T , the camera focal length is f oc = 360, so the normalized image coordinates are Input relative velocity v r = [8.51 2.64 0.72]T and its own attitude

[0112] S12. State initialization a d = 0,

[0113] S13. Calculate n according to the camera imaging model and the image features estimated in step 2 calculate n t , specifically

[0114] S14. During interception, calculate the attitude controller command, the desired pulling force f d = 0.69mg = 8.11N, the desired angular velocity ω d = [-0.116 0.027 -0.064] T .

[0115] S15. Because the image features have been updated, jump to step S16.

[0116] S16. Calculate the collinear controller command ω 1 = [-0.013 0.122 -0.007] T and the attitude control set value a d = [-2.31 -2.44 0.85] T ,

[0117] S17. The comprehensive control instruction ω d is: ω d = ω d + ω 1 = [-0.129 0.159 -0.071] T .

[0118] S18. The comprehensive control instruction ω d is: m d = [-0.251 0.177 -0.162] T .

[0119] S19. After 20 milliseconds, jump to step S14 again until the interception task ends.

[0120] Step 2: Observer based on delayed Kalman filter

[0121] Flight data analysis: Figure 4bShows the data recorded in the flight experiment using the filter estimates. The image coordinate plot shows the effectiveness of the delayed Kalman filter in smoothing image measurement anomalies and correcting delays.

[0122] The state observer based on the delayed Kalman filter can compensate for the delays in the imaging and processing of the sensing module. This integration enables the controller to obtain more accurate real-time image measurements, thereby improving control performance under lower imaging frame rates and potential target loss, and further improving the interception accuracy.

[0123] S21. Measured quantities input to the IMU: Angular velocity ω measured by the gyroscope gyr =[-0.33 -0.02 -0.10] T , acceleration a measured by the accelerometer acc =[-1.98 0.64 8.92] T , image measurement output by the detection module

[0124] S22. Initialization

[0125] S23. Since the image measurement of the detection module has been updated, jump to step 24.

[0126] S24. According to formulas (18)-(19): Use the IMU measurements to recursively estimate the state and obtain

[0127]

[0128] S25. According to formulas (20)-(22): Estimate the state at time t k-D at the moment

[0129]

[0130] S26. According to formulas (18), (19), recursively estimate the state backward, and repeat step S24 until the delayed state estimate is updated to the current moment,

[0131]

[0132] S27. Jump to step S21 until the interception task ends.

Claims

1. A method for intercepting high-maneuverable targets of lifting-wing UAVs based on visual servoing, characterized in that: The steps include: Step 1: Controller based on visual servoing The visual servo-based controller calculates the interception control amount based on the collected image information and the aircraft's own status; include: S11. Input image measurement Relative speed v r and your own attitude S12. Initialize a according to the current state of the interceptor d and R d , a d and R d They are acceleration and attitude control instructions respectively; S13, based on camera imaging model and image features Calculate n t ; S14. In interception, according to the formula: Calculate the attitude controller command, f d Indicates the tension control command. e f drag represents the drag vector under EFCS, b ω2 represents the angular velocity control command of the desired attitude transformation under the machine system; S15. Determine image features Whether to update, if updated, jump to step S16, if not, jump to step S18; S16. According to a d =-k2v r -k3z1+ e a t (4) Calculate the target colinear controller command ω1 and attitude control setting value a based on visual servoing d ,R d , where v r is the relative speed between the interceptor and the target, is the target tracking error, e a t It represents the acceleration of the target under EFCS, I is the unit matrix, r is the thrust axis, φ is the angle between the current thrust direction and the expected thrust, k1, k2, k3 are constant coefficients, which are adjusted according to the control effect; S17, comprehensive control angular velocity command ω d for: oh d =ω d +ω1 (6) S18, comprehensive control torque command m d for: Where J represents the moment of inertia, represents the proportionality coefficient, represents the integral coefficient, Represents the proportionality coefficient, m d,1 represents the torque control component generated by the integral term, represents the saturation value of the integral term, m d,max Indicates the saturation value of the total torque command; S19, if the interception task is not completed, jump to step S14 again after T time; Step 2: Observer based on delayed Kalman filter S21, input IMU measurement: angular velocity ω measured by gyroscope gyr and the acceleration a measured by the accelerometer acc , the image measurement output by the detection module S22, Initialization S23, judging the image measurement of the detection module i p img Whether to update, if updated, jump to step S24, if not, jump to step S25; S24, according to the formula: The recursive state is measured using IMU, where Indicates t k-1 The state estimate at time Indicated by t k-1 The state estimate at time t is obtained by one step recursion k The state estimate at time F k and G k Respectively represent from t k-1 to k The state transfer matrix and noise matrix, P k1 Indicates t k-1 The covariance matrix of the state at each moment, P k|k1 Because k-1 The moment covariance matrix is ​​obtained by one-step recursion. k Moment covariance matrix estimation, Q k The noise matrix representing the recursive process; S25, according to the formula: P k-D =I-KH P k-D|k-D-1 (12) K=P k-D|k-D-1 H T S -1 ,S=HP k-D|k-D-1 H T +R (13) Estimation t k-D Status at all times Among them, z k-D Indicates t k-D The characteristic observation value at the moment, H represents the observation matrix, One-step recursive state estimation, K represents the covariance fusion matrix, Represents the pair t after fusion residual k-D The final state estimate at time t, P kD Represents the residual after fusion k-D The final covariance estimate at time t; S26, according to formula (18), (19), recursively push the state backward, repeat step S24, until the delayed state is estimated Update to the current time; S27: If the interception task is not completed, jump to step S21.

2. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1 is characterized in that: In step S12, 3. The method for intercepting high-maneuverable targets of lift-wing UAVs based on visual servoing according to claim 1 is characterized in that: In step S13, 4. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: Coordinate Systems: For intercept vehicles, four coordinate systems are used, including: ●{e}={o e -x e y e z e } is the Earth Fixed Coordinate System (EFCS), which is used to express the positions and velocities of the interceptor and target in the global inertial system; ●{b}={o b -x b y b z b } is the body coordinate system (BCS), a variable used to represent the current attitude of the lift-wing interceptor quadrotor body relative to the ground; ●{l}={o l -x l y l z l } is the lift wing coordinate system (LCS), a variable used to represent the current attitude of the lift wing interceptor wing relative to the ground; ●{w}={o w -x w y w z w } is the airflow coordinate system (WCS), a variable used to represent the current attitude of the lift wing interceptor relative to the airflow; ●{c}={o c -x c y c z c } is the camera coordinate system (CCS); ●{i}={o i -x i y i } is the image coordinate system (ICS), which is used to represent the image position of the target feature in the first-person view of the camera; matrix Indicates the coordinate system arrive The rotation matrix of The three-dimensional special orthogonal group is given by T represents the transpose of R, and I represents the identity matrix:

5. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: Lifting wing interceptor model: The lifting wing UAV is modeled as a rigid body with a mass of m, and its flight control rigidity model is summarized as: in: Among them, Q represents the air dynamic pressure of the aircraft in the cruising state, S represents the wing area, and They represent the coordinate rotation matrices from the airflow coordinate system to the body coordinate system, from the body coordinate system to the earth fixed coordinate system, and from the lift wing coordinate system to the body system. b and c represent the lift wing span and the average chord length, respectively. and They represent the speed v and The derivative of the rotation matrix, Indicates the position of the aircraft under the EFCS; represents the velocity vector under EFCS; is the acceleration under EFCS; m is the mass of the aircraft; g is the local gravitational acceleration, g = [00g] T ; b f=[0f ry -f rz ] T is the controllable force of the lift wing motor in the BCS; is the equivalent aerodynamic control surface deflection angle of the lift wing in BCS, [δ ar δ al ] T is the left and right aileron deflection angle of the lift wing in BCS; C D , C Y , C L is the aerodynamic coefficient of the lifting surface in the WCS; is the aerodynamic coefficient of the equivalent aerodynamic control surface in WCS; C l , C m , C n is the aerodynamic moment coefficient of the lifting surface in the WCS; is the aerodynamic moment coefficient of the equivalent aerodynamic control surface in WCS; is the angular velocity under BCS; b m is the total torque of the machine system; J is the moment of inertia of the multirotor; τ = [m rx m ry m rz ] T is the controllable torque generated by the propeller and aileron on the fuselage axis; the matrix [ b ω] × is an antisymmetric matrix, which is the angular velocity b The cross product mapping of ω.

6. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: The motion of the target model is represented by a mass point model as follows: in, and They represent the position, velocity and thrust acceleration of the target under EFCS respectively; noise is a Gaussian random variable with zero mean.

7. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: Camera Imaging Model: Perspective projection performs a mapping from 3D space to the image plane: Assume that the origin of CCS coincides with the origin of BCS; that is, their translation vectors And the rotation matrix is a constant; when the camera contacts the target, it is considered a successful interception; Therefore, there is a constant rotation between the optical axis and the direction of the aircraft head. e p and e p t Respectively represent the position of the interceptor and the position of the target; i p=[p xi p yi ] T are the coordinates of the target; n t represents the target unit vector along the line of sight in EFCS; n c represents the optical axis unit vector in EFCS; n td Defined as the unit vector designed in EFCS, is the designed LOS vector; i p td is the intersection point with the image plane in ICS, called the interception point; the interception problem is summarized as n t and n td are collinear, so e p and e p t Close to each other.

8. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: Visual servoing (IBVS) model: The image Jacobian matrix represents the relationship between camera motion and image feature motion as follows: in, Represents the velocity and angular velocity in CCS, including the linear velocity and angular velocity of the rigid body; Represents the normalized image coordinates of the features in ICS, p zc Indicates the z-coordinate of the feature point in CCS.

9. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: System model: Logarithmic mapping ln(·):SO(3)→so(3) transforms the rotation matrix R b e Converted to a rotation angle θ∈[-π,π) and a unit rotation vector Right now A four-dimensional vector compactly represents the rotation matrix of the interceptor. Orientation Quaternion The normalized image coordinates of the target center are regarded as feature points and recorded as State variables It consists of some necessary states, including the interceptor's direction quaternion q, relative position p r and relative velocity v r , image feature points i p, gyro bias b gyr =[b gyr,x b gyr,y b gyr,z ] T , accelerometer bias b acc =[b acc,x b acc,y b acc,z ] T ; The discrete state equation is written as: x k =F k x k-1 +G k w k-1 (18) in, and From time t k-1 To time t k The state transfer matrix and noise matrix of w k-1 is the time t k-1 The process noise vector, x k and x k-1 The time t k and time t k-1 The state variables of , whose elements follow a Gaussian distribution with zero mean and are uncorrelated, Vector k Gyroscope bias noise and accelerometer partial noise Depend on composition.

10. The method for intercepting a high-maneuverability target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: IMU measurement model: The inertial measurement unit (IMU) provides gyroscope angle increment and accelerometer velocity increment measurements under BCS; the measurements are combined with the aircraft rigid body model to obtain the states q and v r The recursive expression of ; ●The derivative of the directional quaternion q is the angular velocity b ω, obtained from the gyroscope, is measured as: b oh = oh gyr -b gyr -n gyr (19) Where, the gyroscope measurement noise n gyr is a Gaussian random variable with mean 0; Bias b gyr Follow the Weiner process as follows: ●Speedv r The derivative of acceleration e a, obtained from the accelerometer measurements: Where, e3 = [001] T , a acc represents the accelerometer measurement value, and the accelerometer measurement noise n acc is a zero-mean Gaussian random variable; the accelerometer bias b acc Following the Wiener process, 11. The method for intercepting a high-maneuverable target of a lifting-wing UAV based on visual servoing according to claim 1, 2 or 3, characterized in that: Image delay measurement model: Assume that the measured delay based on the IMU update frequency is D cycles, and the delay time t D >0 is known; then the image measurement value at time t is expressed as: in, represents the state at time kD, image measurement noise n img is a zero-mean Gaussian random variable; the delay time t of this aircraft D The camera imaging and image processing times are included and estimated using two methods: ① direct calculation based on camera parameters and internal program timers; ② experimental measurement using camera capture timers.

Images