Anti-uav phase plane guidance and field of view constraint prediction interception control method

Abstract

The application discloses a kind of anti-UAV phase plane guidance and field of view restriction prediction interception control method, belong to the intersection field of unmanned aerial vehicle flight control and visual servoing.This method obtains pure translation optical flow by establishing optical flow de-rotation model and estimates visual time to collision TTC, based on phase plane proportional guidance constructs nonlinear model predictive control NMPC cost function with dynamic adaptive weight, simultaneously camera field of view angle boundary is converted into non-convex hard constraint, finally integrates dynamics equation and constraint condition completes online rolling optimization and closed-loop interception.The application solves the problems of traditional monocular visual servoing "centering deadlock", three-dimensional proportional guidance depth dependence, target easy to lose and end impact attitude inaccuracy, realizes the limit maneuver of high dynamic target under monocular depth-free condition, field of view preservation and accurate head-on interception.

Description

Technical Field

[0001] This invention relates to the field of cross-technology of UAV flight control and visual servoing, and in particular to a method for anti-UAV phase plane guidance and field-of-view constraint prediction interception control. Background Technology

[0002] In recent years, the demand for low-altitude security and the control of unauthorized drone flights has been increasing, making autonomous interception of quadcopter drones based on monocular vision a research hotspot. Existing interception technologies are mainly divided into two categories: image-based visual servoing (IBVS) and three-dimensional proportional guidance (PN). Traditional IBVS methods use target image centering as the control target, which is a passive tail-chasing logic. Since quadcopters are underactuated systems, forced centering prevents the drone from performing extreme maneuvers at large angles, severely limiting its dynamic performance, and the target easily slips out of the field of view, leading to tracking loss. Existing improved methods often set the field of view constraint as a soft constraint, which cannot guarantee that the target will not be lost under extreme maneuvers. Traditional three-dimensional proportional guidance relies on accurate relative depth information of the target, but monocular cameras cannot directly provide absolute depth. In high-dynamic combat scenarios, depth estimation is prone to divergence, leading to guidance command failure. Simply relaxing the field of view constraint to improve maneuverability can cause the drone's optical axis to fail to align with the target at the end of the interception, resulting in unexpected collisions between the propeller, frame, and target auxiliary structures, posing a safety hazard.

[0003] In summary, existing technologies cannot simultaneously achieve extreme maneuverability, full-range visibility, precise terminal aiming, and safe interception under monocular depth-free conditions, making it difficult to meet the requirements for high-dynamic anti-drone interception. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the purpose of this invention is to provide a method for anti-UAV phase plane guidance and field-of-view constraint prediction interception control.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] This invention provides a method for anti-UAV phase plane guidance and field-of-view constraint prediction interception control, comprising the following steps:

[0007] S1: Establish a monocular camera optical flow de-rotation model and a visual collision time estimation model, remove the optical flow that is disturbed by the body rotation, obtain the pure translational optical flow of the target, and estimate the remaining collision time based on the pixel area dilation rate of the target detection box.

[0008] S2: Based on the phase plane ratio guidance law, a nonlinear model prediction control cost function with dynamic adaptive weights for visual collision time is constructed to achieve adaptive mode handover between mid-section approximation and end-section alignment.

[0009] S3: Based on the camera's physical field of view boundary, construct a non-convex field of view hard constraint with a safety margin to ensure that the target is within the effective field of view throughout the entire process;

[0010] S4: Integrates quadrotor dynamics equations, actuator saturation constraints, field of view hard constraints and cost functions, performs rolling time-domain optimization through a nonlinear solver, and issues the optimal control command to the flight control system to execute closed-loop interception.

[0011] Furthermore, the pure translational optical flow mentioned in step S1 is obtained by removing the rotational optical flow corresponding to the body attitude angular velocity from the optical flow observed by the airborne camera. The pure translational optical flow is equivalent to the two-dimensional line-of-sight angular rate. The remaining collision time is calculated from the pixel area of ​​the target detection box and its time change rate.

[0012] Furthermore, the dynamic adaptive weights described in step S2 change nonlinearly with the visual collision time. When the collision time is long, the weights approach zero, allowing the target to wander at the edge of the field of view; when the collision time approaches zero, the weights increase sharply, forcing the target to align with the center of the image.

[0013] Furthermore, the hard constraint on the field of view in step S3 is a non-convex inequality constraint based on the horizontal and vertical pixel boundaries of the camera and superimposed with a safety margin, and the target predicted pixel coordinates are strictly limited within the constraint range.

[0014] Furthermore, in step S4, the fourth-order Runge-Kutta method is used to discretize the dynamic equations, and a sequential quadratic programming scheme combined with a real-time iterative approach is used for online solution. The calculation is accelerated by starting the optimal solution from the previous cycle.

[0015] Compared with the prior art, the technical solution disclosed in this invention has the following beneficial effects:

[0016] This invention uses pure translational optical flow to replace three-dimensional spatial motion calculation, reducing the three-dimensional proportional guidance to a two-dimensional image plane. In principle, it eliminates the dependence on the absolute depth of the target, avoids guidance failure caused by depth estimation divergence under high maneuverability, and improves the robustness of the system.

[0017] This invention constructs the camera field of view boundary as a non-convex hard constraint, replacing the traditional soft penalty constraint. Under the premise of ensuring that the target is not lost, it allows the UAV to place the target at the edge of the field of view to achieve large tilt angle acceleration, completely breaking the "centering deadlock" of traditional visual servoing and releasing the ultimate dynamic maneuverability of the quadcopter.

[0018] This invention employs a visual collision time-driven dynamic adaptive weight. During mid-stage pursuit, the centering requirement is reduced to increase speed, while the centering weight is drastically increased before the final impact to force optical axis alignment. This ensures interception efficiency and avoids unexpected collisions at the end, achieving smooth modal transitions and precise head-on collisions.

[0019] This invention integrates phase plane guidance, hard field of view constraints, and nonlinear model predictive control, enabling high-dynamic target interception under monocular vision without depth input. The system is lightweight, highly versatile, and suitable for real-time interception of low-altitude anti-UAVs. This allows monocular UAVs to safely, accurately, and rapidly intercept targets without losing track of them, and without being restricted by image centering. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic flowchart of the anti-UAV phase plane guidance and field-of-view constraint prediction interception control method provided in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram illustrating the principle of the anti-UAV phase plane guidance and field-of-view constraint prediction interception control method provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] This invention provides a phase plane guidance and field-of-view constraint prediction interception control method for anti-UAVs, enabling monocular UAVs to safely, accurately, and at high speed intercept targets without losing track of them or being restricted by the centering of the image.

[0026] like Figures 1-2 As shown, this embodiment of the invention provides a method for anti-UAV phase plane guidance and field-of-view constraint prediction interception control, including the following steps:

[0027] S1: Establish a monocular camera optical flow de-rotation model and a visual collision time estimation model, remove the optical flow that is disturbed by the body rotation, obtain the pure translational optical flow of the target, and estimate the remaining collision time based on the pixel area dilation rate of the target detection box.

[0028] The pure translational optical flow mentioned in step S1 is obtained by removing the rotational optical flow corresponding to the body attitude angular velocity from the optical flow observed by the airborne camera. The pure translational optical flow is equivalent to the two-dimensional line-of-sight angular rate. The remaining collision time is calculated from the pixel area of ​​the target detection box and its time change rate.

[0029] Specifically, an optical flow de-rotation model based on a monocular camera and a visual collision time-to-traffic (TTC) estimation model are established to obtain pure relative guidance information. The formula for calculating the pure translational optical flow of the target on the camera image plane is shown below:

[0030]

[0031]

[0032] in The representation considers the pure translational optical flow (equivalent to a two-dimensional line-of-sight angular rate) after the UAV's own attitude interference, with units of . , This represents the pixel velocity of the target directly observed by the airborne monocular camera. Represents the high-frequency airframe attitude angular velocity acquired by the airborne inertial measurement unit (IMU), in units of... , The Jacobian matrix component representing the image corresponding to rotational motion is determined by the camera's focal length and the intrinsic parameters of the principal pixel coordinates. The specific analytical expression is shown below:

[0033]

[0034] In the formula, and These represent the horizontal and vertical equivalent focal lengths of the airborne camera, respectively; due to manufacturing limitations, there is a deviation of several to tens of pixels between the camera's principal point and geometric center. and The intrinsic parameter representing the principal pixel coordinates of the camera's optical center on the image plane.

[0035] Simultaneously, the remaining collision time is estimated based on the expansion rate of the target's detection box area in the image plane. As shown below:

[0036]

[0037] in This represents the pixel area occupied by the object detection bounding box in the current image frame. This represents the rate of change of the area of ​​the target detection box over time.

[0038] S2: Based on the phase plane ratio guidance law, a nonlinear model predictive control cost function with dynamic adaptive weights for visual collision time is constructed to achieve adaptive mode handover between mid-segment approximation and end-segment alignment.

[0039] The dynamic adaptive weights described in step S2 change nonlinearly with the visual collision time. When the collision time is long, the weights approach zero, allowing the target to wander at the edge of the field of view. When the collision time approaches zero, the weights increase sharply, forcing the target to align with the center of the image.

[0040] Specifically, based on the phase plane proportional guidance law, a nonlinear model predictive control (NMPC) cost function with dynamic adaptive weights is designed to achieve the mode transition of the system from "rapid approximation through mid-range edge wandering" to "precise alignment through forced centering at the end." The cost function expression is shown below:

[0041]

[0042] in Represents the total number of steps in the prediction time domain. The representative system is Control input variables at any given time (including main thrust) With triaxial torque ), The representative proportional guided penalty weight parameter is used to force the pure translation optical flow to approach zero; The target is Predicted pixel coordinates at time step The pixel coordinates representing the geometric center of the camera image. This represents the weight matrix for controlling energy smoothing penalty.

[0043] The centering adaptive penalty weight, which dynamically evolves with collision time, is expressed as a nonlinear function as follows:

[0044]

[0045] in This represents the end-impact gain constant, and its magnitude affects the aggressiveness of the end-impact alignment. A minimum positive constant is set to prevent [something] from occurring at the moment of physical impact. This prevents errors in division by zero singularity operations, ensuring the numerical stability of the system.

[0046] S3: Based on the camera's physical field of view boundary, construct a non-convex field of view hard constraint with a safety margin to ensure that the target is within the effective field of view throughout the entire process.

[0047] The hard constraint of the field of view in step S3 is a non-convex inequality constraint based on the horizontal and vertical pixel boundaries of the camera and superimposed with a safety margin. The target predicted pixel coordinates are strictly limited within the constraint range.

[0048] definition Representing drones The predicted dynamic state vector at time t is a 13-dimensional vector, whose analytical definition is as follows:

[0049]

[0050] in:

[0051] Represents the three-dimensional position coordinates of the UAV in the world coordinate system;

[0052] This represents the three-dimensional linear velocity of the drone in the world coordinate system.

[0053] The unit quaternion represents the attitude of the drone and is used to describe the rotation relationship between the body coordinate system and the world coordinate system.

[0054] This represents the three-axis angular velocity of the drone in the body coordinate system.

[0055] By defining the complete state variables mentioned above, and in conjunction with the quadcopter dynamics equations, the nonlinear solver can accurately predict the future pose and trajectory of the UAV in each control cycle, thus providing a basis for the calculation of hard constraints on the field of view.

[0056] Based on the camera's physical field of view boundary, a hard constraint on the non-convex field of view (FOV) is constructed to ensure that the target never leaves the effective field of view when the UAV performs extreme maneuvers at large tilt angles. The inequality equations for the hard field of view constraint are shown below:

[0057]

[0058] This represents the target's three-dimensional relative position variable after filtering estimation. and Represents the minimum and maximum physical pixel boundaries of the camera image in the horizontal direction. and Represents the minimum and maximum physical pixel boundaries of the camera image in the vertical direction. This represents the additional safety field of view and anti-loss pixel margin set to cope with high-frequency oscillations and tracking delays. and Let and represent the nonlinear perspective projection function of the target projected from three-dimensional relative space onto the camera's two-dimensional image plane, respectively. This function depends on the state vector. The attitude and rotation information is contained within. The parsing and construction process is as follows:

[0059] First, based on the drone's State vector at time step The attitude quaternion (or Euler angle) in the equation. The transformation formula from the world coordinate system to the body coordinate system is given:

[0060]

[0061] This matrix evolves in real time with the dynamic state of the UAV and is the core basis for the NMPC solver to calculate the field of view boundary in the prediction time domain.

[0062] Camera mounting extrinsic matrix This is a constant matrix determined by the physical mounting position of the camera on the drone. Typically, when a monocular camera is mounted forward, there is an axis remapping between the body coordinate system (front-left-top) and the camera coordinate system (right-bottom-front), which is typically in the form of:

[0063]

[0064] in This represents the preset mounting angle of the camera relative to the camera body axis. This allows the algorithm to adapt to different hardware structures.

[0065] Rotation matrix from world coordinate system to body coordinate system Combined with the preset camera mounting extrinsic rotation matrix This allows us to obtain the target's three-dimensional relative position vector in the camera coordinate system. :

[0066]

[0067] In the formula, From the state vector The 3D position of the UAV was extracted. Then, using an ideal pinhole camera projection model, the specific form of the nonlinear perspective projection function was obtained:

[0068]

[0069] In the formula, , This refers to the equivalent focal length of the camera in both horizontal and vertical directions. , This is the camera's main internal reference.

[0070] S4: Integrates quadrotor dynamics equations, actuator saturation constraints, field of view hard constraints and cost functions, performs rolling time-domain optimization through a nonlinear solver, and issues the optimal control command to the flight control system to execute closed-loop interception.

[0071] In step S4, the fourth-order Runge-Kutta method is used to discretize the dynamic equations, and a sequential quadratic programming scheme combined with a real-time iterative approach is used for online solution. The calculation is accelerated by starting the optimal solution from the previous cycle.

[0072] Specifically, the state-space equations and boundary hard constraints are integrated to perform multi-dimensional online optimization and low-level closed-loop execution. The optimal control problem, which includes the quadrotor system dynamics equations, actuator thrust saturation limits, and the aforementioned non-convex hard constraints of the field of view, is fed into the nonlinear solver (such as CasaADi / acados) of the airborne edge computing platform for rolling time-domain computation. Within each control cycle, the first control command vector from the solved optimal control sequence is extracted and sent to the low-level flight control system to complete the predictive interception operation under the entire field of view constraint.

[0073] More specific examples are as follows:

[0074] S1 can be described as follows: In view of the shortcomings of the existing three-dimensional proportional guidance PN which relies heavily on absolute depth and is prone to guidance failure due to depth estimation divergence under monocular high-frequency maneuvering, this step extracts pure relative guidance information by establishing an optical flow derotation model and a visual collision time TTC estimation model.

[0075] Specifically, in order to eliminate depth dependency, this step first calculates the pure translational optical flow of the target:

[0076]

[0077] This formula incorporates the interference optical flow generated by the quadrotor's own attitude rotation (i.e. This is extracted from the total observed optical flow. Through this algebraic operation, the system can directly obtain pure translational guidance information equivalent to the "line-of-sight angular rate" on the two-dimensional image plane, thus successfully reducing the three-dimensional guidance law to a two-dimensional plane, completely bypassing the absolute depth of the target from a mathematical perspective. The computational dependency.

[0078] Meanwhile, to address the difficulty of monocular ranging, this step introduces area dilation rate to estimate the remaining collision time. :

[0079]

[0080] Utilizing scale-independent visual features (target pixel area) and its rate of change This constructs a depth-independent temporal prediction metric. Without relying on any absolute distance information, it provides a precise triggering time reference for subsequent adaptive modal handover.

[0081] S2 can be described as follows: In response to the fatal flaw that "the nose of the aircraft is not aligned with the target at the moment of terminal impact and is prone to unexpected collisions" caused by directly relaxing the field of view constraint (allowing the target edge to wander), this step designs an NMPC dynamic cost function based on phase plane proportional guidance and adaptive weights.

[0082] Specifically, in order to resolve the physical contradiction between "the need for extremely high-speed pursuit in the middle stage" and "the need for high-precision alignment in the final stage," this step constructs the following dynamic cost function:

[0083]

[0084] The specific function of this formula is to unify "pure translation optical flow to zero" and "pixel deviation to zero" into the same objective function, by optimizing variables. The optimal solution for the control torque is sought. Simultaneously, to precisely control the priority of the two tasks mentioned above, a nonlinear dynamic weighting function is specifically designed:

[0085]

[0086] In practical engineering implementation, in order to ensure the convergence of the numerical solver throughout the entire prediction time domain, the minimum positive constant... Preferred value range to (For example, take) The value of the terminal impact gain constant is tuned based on the maximum normal overload allowed by the interception system, with a typical reference range of 10 to 1000. This parameter configuration enables highly efficient pursuit of highly maneuverable targets while ensuring that the algorithm does not experience division-by-zero crashes.

[0087] when When the target is relatively large (mid-range pursuit), the weight calculated by this formula is... Minimal, causing the cost function to ignore the centering requirement, allowing the drone to pursue at full speed; when As the weights approach zero (on the eve of impact), the formula causes them to spike exponentially. This mathematical mechanism forces the NMPC solver to issue a "tail-swinging" command to the aircraft at the end of the flight path, overcoming the impact hazard caused by misalignment of the optical axis at the end of the flight path.

[0088] S3 can be described as follows: Addressing the "centering deadlock" defect of traditional image-based visual servoing (IBVS), which forces the target to be centered, causing underactuated UAVs to avoid large-angle acceleration and thus limiting their maneuverability, this step constructs a non-convex field of view (FOV) hard constraint based on the camera's physical field of view boundary. Specifically, to maximize the release of dynamic potential while ensuring the target is not lost, this step transforms the field of view redline into the following rigorous mathematical inequality:

[0089]

[0090] This changes the traditional IBVS approach of compromising "not losing the target" as a soft penalty. It utilizes a perspective projection function... Set at absolute physical pixel boundaries The purpose of this set of formulas is that, as long as the trajectory predicted by the solver does not cross this hard constraint line, the drone can use any extreme roll or pitch angle to extract the maximum acceleration. This design allows the drone to actively "force" the target to the physical edge of the field of view, thereby breaking the "centering deadlock" of traditional algorithms.

[0091] S4 can be described as: integrating state-space equations and hard boundary constraints, performing multi-dimensional online optimization and low-level closed-loop execution. In order to achieve high-frequency (e.g., 50Hz) nonlinear prediction and interception on edge computing platforms with limited onboard computing power, the above dynamic cost function and constraints are constructed as a standard nonlinear optimal control problem (OCP), and then discretized and executed.

[0092] The specific implementation process includes:

[0093] The dynamics formula for a continuous system of quadcopters is as follows:

[0094]

[0095] Since this formula cannot be directly solved by a numerical optimizer, this embodiment uses the Runge-Kutta 4th Order (RK4) method combined with a multiple-shot method to discretize the time. A step size is set... (e.g., 0.02 seconds) will predict the time domain. The internal dynamics are transformed into a sequence of equality constraints:

[0096]

[0097] In the formula, Represents the fourth-order Runge-Kutta discrete integral operator. For the first The system state of each prediction node (including position, velocity, attitude Euler angles or quaternions, and body angular velocity). For the first The control input for each node.

[0098] In the solver, in addition to configuring the dynamic cost function in the second step and the non-convex field of view (FOV) hard constraint in the third step, physical saturation constraints of the UAV actuators must also be added to ensure that the generated trajectory is absolutely executable in the real physical world. The added inequality constraints include:

[0099]

[0100] In the formula, The maximum combined thrust that a quadcopter motor can provide, This is the maximum control torque limit that the three axes of the machine can withstand.

[0101] The discretized cost function, equality constraints (system dynamics), and inequality constraints (FOV boundary and thrust saturation) are integrated to construct a standard nonlinear optimal control problem. This problem is then input into the nonlinear optimization solver of the airborne computing platform for high-frequency solving. To ensure real-time computation in high-dynamic adversarial scenarios, the solution process employs a Sequential Quadratic Programming (SQP) algorithm combined with a Real-Time Iteration (RTI) scheme. During each rolling solution iteration, a warm-start initialization is performed using the optimal solution from the previous control cycle, significantly reducing the iteration time of nonlinear optimization and meeting the system's high-frequency control update requirements.

[0102] After completing a single optimization, the solver outputs the predicted time domain. The optimal control input sequence within:

[0103]

[0104] The system extracts only the first control command vector from the sequence. (That is, the desired thrust and three-axis torque at the current moment), are sent to the underlying flight control module via the airborne communication bus. The underlying flight control module, in conjunction with its built-in attitude control loop and hybrid control logic, converts the above torque commands into underlying execution signals to drive each rotor motor. Subsequently, the system acquires the latest visual observation and inertial navigation status as the new initial state. The prediction window is scrolled forward one step, and the above process is repeated to achieve closed-loop interception control of "visual perception - state prediction - scrolling optimization - low-level execution".

[0105] Compared with the prior art, the beneficial effects of the embodiments of the present invention are as follows:

[0106] The present invention provides a phase plane guidance and field-of-view constraint prediction interception control method for anti-UAVs, which achieves several breakthroughs based on the prior art.

[0107] First, by mapping the three-dimensional proportional guidance law in missile engineering to a two-dimensional image plane and using "pure translational optical flow" to replace the calculation of relative motion in three-dimensional space, this method completely eliminates the dependence of monocular vision systems on absolute depth information, achieves optimal interception under pure azimuth conditions, and effectively avoids system collapse caused by depth estimation divergence in highly maneuverable scenarios, thus significantly improving guidance robustness.

[0108] Meanwhile, unlike traditional image visual servoing methods based on centering error, this invention constructs the physical boundary of the camera field of view as a strict non-convex inequality hard constraint, enabling the nonlinear model predictive control solver to autonomously search for the optimal "tilt balance point" in multi-dimensional space; in order to pursue the ultimate straight-line pursuit acceleration, the system dares to actively push the target to the edge of the field of view, thereby completely releasing the dynamic maneuver potential of the quadcopter and solving the limitation of "centering deadlock" on maneuverability.

[0109] In the final stage of interception, this embodiment of the invention further introduces a nonlinear dynamic weight based on visual collision time: when the collision time is relatively large during mid-stage interception, the centering weight approaches zero, and the UAV pursues at full speed; while when the collision time in the final stage approaches zero, the centering weight is sharply amplified, forcing the UAV's attitude to remain centered and aligned with the optical axis. This adaptive modal transition mechanism effectively compensates for the blind spot problem existing in pure edge tracking at the end, ensuring both accurate collision between the interceptor and the target and avoiding catastrophic risks caused by collisions with auxiliary structures.

[0110] The basic principles of the present invention have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in the present invention are merely examples and not limitations, and should not be considered as essential features of each embodiment of the present invention. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the present invention to the necessity of employing the aforementioned specific details.

[0111] The block diagrams of devices, apparatuses, devices, and systems involved in this invention are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0112] It should also be noted that in the apparatus, device, and method of the present invention, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of the present invention.

[0113] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the invention. Therefore, the invention is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0114] It should be understood that the qualifying terms "first", "second", "third", "fourth", "fifth" and "sixth" used in the description of the embodiments of the present invention are only used to more clearly illustrate the technical solutions and are not intended to limit the scope of protection of the present invention.

[0115] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the invention to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for anti-UAV phase plane guidance and field-of-view constraint prediction interception control, characterized in that, Includes the following steps: S1: Establish a monocular camera optical flow de-rotation model and a visual collision time estimation model, remove the optical flow that is disturbed by the body rotation, obtain the pure translational optical flow of the target, and estimate the remaining collision time based on the pixel area dilation rate of the target detection box. S2: Based on the phase plane ratio guidance law, a nonlinear model prediction control cost function with dynamic adaptive weights for visual collision time is constructed to achieve adaptive mode handover between mid-section approximation and end-section alignment. S3: Based on the camera's physical field of view boundary, construct a non-convex field of view hard constraint with a safety margin to ensure that the target is within the effective field of view throughout the entire process; S4: Integrates quadrotor dynamics equations, actuator saturation constraints, field of view hard constraints and cost functions, performs rolling time-domain optimization through a nonlinear solver, and issues the optimal control command to the flight control system to execute closed-loop interception.

2. The method according to claim 1, characterized in that, The pure translational optical flow mentioned in step S1 is obtained by removing the rotational optical flow corresponding to the body attitude angular velocity from the optical flow observed by the airborne camera. The pure translational optical flow is equivalent to the two-dimensional line-of-sight angular rate. The remaining collision time is calculated from the pixel area of ​​the target detection box and its time change rate.

3. The method according to claim 1, characterized in that, The dynamic adaptive weights described in step S2 change nonlinearly with the visual collision time. When the collision time is long, the weights approach zero, allowing the target to wander at the edge of the field of view. When the collision time approaches zero, the weights increase sharply, forcing the target to align with the center of the image.

4. The method according to claim 1, characterized in that, The hard constraint of the field of view in step S3 is a non-convex inequality constraint based on the horizontal and vertical pixel boundaries of the camera and superimposed with a safety margin. The target predicted pixel coordinates are strictly limited within the constraint range.

5. The method according to claim 1, characterized in that, In step S4, the fourth-order Runge-Kutta method is used to discretize the dynamic equations, and a sequential quadratic programming scheme combined with a real-time iterative approach is used for online solution. The calculation is accelerated by starting the optimal solution from the previous cycle.

Images