Intrusion unmanned aerial vehicle interception method and system based on vision and data fusion

Abstract

Embodiments of the present application disclose an intrusion UAV interception method and system based on visual and data fusion. Embodiments of the present application can acquire a current image of an intrusion UAV, and determine a current relative position between the intrusion UAV and an interception UAV. Based on the current relative position and current pose data of the interception UAV in a global coordinate system, a current absolute position of the intrusion UAV is determined. The current speed information of the intrusion UAV is calculated, and based on the current speed information and the current absolute position, a current observation motion state is obtained. Through a Kalman filtering algorithm, a state update is performed to obtain a current corrected motion state. Based on the current corrected motion state of the intrusion UAV and the historical corrected motion state corresponding to the intrusion UAV in a historical time period, a motion trajectory of the intrusion UAV in a future prediction time domain is predicted. Based on the motion trajectory, the interception UAV is controlled to perform an interception task on the intrusion UAV. Thus, embodiments of the present application improve the interception success rate.

Description

Technical Field

[0001] This application relates to the field of drone countermeasures technology, specifically to a method and system for intercepting intrusive drones based on vision and data fusion. Background Technology

[0002] With the widespread adoption of drone technology, the security threat posed by drones to airports, major event venues, and sensitive airspace is becoming increasingly prominent. Physical interception of drones, due to its high precision and minimal collateral damage, has become an effective countermeasure.

[0003] However, existing autonomous interception methods based on UAV platforms still face significant challenges in practical applications: First, when relying solely on airborne visual information to calculate target position, the obtained relative position and motion state of the target are subject to considerable uncertainty due to factors such as imaging scale, lens distortion, and changes in target scale, making it difficult to meet the positioning requirements for high-precision interception. Second, most interception strategies generate control commands based primarily on the instantaneous observation state of the target. When the intruding UAV performs high-speed or irregular maneuvers, it is difficult to adjust the interception trajectory in time, leading to interception failure. Third, existing systems lack effective coordination mechanisms in perception, state estimation, path planning, and low-level control. The processing results at each stage are difficult to form a closed-loop feedback, resulting in a significant decrease in overall performance under complex environments (such as strong wind disturbances, electromagnetic interference, or GNSS signal blockage), and insufficient system robustness. These factors collectively lead to a low success rate in intercepting intruding UAVs, especially those performing active evasion maneuvers. Summary of the Invention

[0004] This application provides a method and system for intercepting intrusive drones based on vision and data fusion, which improves the interception success rate.

[0005] This application provides a method for intercepting intrusive drones based on vision and data fusion, the method including:

[0006] The system acquires the current image of the intruding drone captured by the camera on the interceptor drone at the current time, and determines the current relative position between the intruding drone and the interceptor drone based on the pixel position information of the corresponding area of ​​the intruding drone in the current image, as well as the transformation relationship between the camera coordinate system and the interceptor drone's body coordinate system.

[0007] Based on the current relative position and the current pose data of the intercepting drone in the global coordinate system, determine the current absolute position of the intruding drone in the global coordinate system;

[0008] Based on the current absolute position and the absolute position of the intruding drone at the previous time, calculate the current speed information of the intruding drone, and based on the current speed information and the current absolute position, obtain the current observed motion state;

[0009] The current predicted motion state of the intruding drone at the current time is obtained, and the current predicted motion state is used as a prior estimate, and the current observed motion state is used as a measurement input. The state is updated through the Kalman filter algorithm to obtain the current corrected motion state.

[0010] Based on the current corrected motion state of the intruding drone and its historical corrected motion state in the historical time period, the motion trajectory of the intruding drone in the future prediction time domain is predicted.

[0011] Based on the motion trajectory, the interceptor drone is controlled to perform the task of intercepting the intruding drone.

[0012] This application also provides an intrusion drone interception system based on vision and data fusion, the system comprising:

[0013] The visual relative positioning unit is used to acquire the current image of the intruding drone captured by the camera on the intercepting drone at the current time, and determine the current relative position between the intruding drone and the intercepting drone based on the pixel position information of the corresponding area of ​​the intruding drone in the current image, as well as the transformation relationship between the camera coordinate system and the intercepting drone body coordinate system.

[0014] The absolute position determination unit is used to determine the current absolute position of the intruding drone in the global coordinate system based on the current relative position and the current pose data of the intercepting drone in the global coordinate system.

[0015] The observation state construction unit is used to calculate the current velocity information of the intruding drone based on the current absolute position and the absolute position of the intruding drone in the previous time, and to obtain the current observation motion state based on the current velocity information and the current absolute position.

[0016] The state correction unit is used to obtain the current predicted motion state of the intruding drone at the current time, and use the current predicted motion state as a prior estimate and the current observed motion state as a measurement input. The state is updated through the Kalman filter algorithm to obtain the current corrected motion state.

[0017] The trajectory prediction unit is used to predict the trajectory of the intruding drone in the future prediction time domain based on the current corrected motion state of the intruding drone and its historical corrected motion state in the historical time period.

[0018] The interception control unit is used to control the interception drone to perform the interception task against the intruding drone based on the motion trajectory.

[0019] This application also provides an electronic device, including a processor and a memory, wherein the memory stores multiple instructions; the processor loads instructions from the memory to execute the steps in any of the vision and data fusion-based intrusion drone interception methods provided in this application.

[0020] This application also provides a computer-readable storage medium storing multiple instructions adapted for loading by a processor to execute steps in any of the vision-and-data fusion-based intrusion drone interception methods provided in this application.

[0021] This application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps in any of the vision and data fusion-based intrusion drone interception methods provided in this application.

[0022] In this application, after obtaining the current image of the intruding drone captured by the camera on the interceptor drone at the current time, the current relative position between the intruding drone and the interceptor drone is determined based on the pixel position information of the corresponding area of ​​the intruding drone in the current image and the transformation relationship between the camera coordinate system and the interceptor drone's body coordinate system. This achieves high-precision close-range positioning of the intruding drone relying solely on airborne vision. Furthermore, based on the current relative position and the current pose data of the interceptor drone in the global coordinate system, the current absolute position of the intruding drone in the global coordinate system can be determined. This establishes a unified spatial reference for the perception results and the navigation system, providing a consistent coordinate framework for subsequent trajectory prediction and cooperative control. Based on this, the current velocity information of the intruding drone is calculated based on the current absolute position and the absolute position of the intruding drone at the previous time. The system obtains the current observed motion state by analyzing the velocity information and the current absolute position. Then, it acquires the current predicted motion state of the intruding drone at the current time and uses this predicted motion state as a priori estimate. Using the current observed motion state as the measurement input, a Kalman filter algorithm is applied to update the state, resulting in the current corrected motion state. This effectively suppresses the influence of visual observation noise, imaging errors, and model uncertainties, significantly improving the stability and accuracy of state estimation. Based on the current corrected motion state of the intruding drone and its historical corrected motion states in historical time periods, the system predicts the drone's trajectory in the future prediction time domain. The predicted trajectory not only conforms to the historical motion trend of the intruding drone but also adapts to its potential maneuvering behavior, exhibiting good foresight and robustness. Finally, based on the trajectory, the system controls an interceptor drone to perform the interception task. This achieves end-to-end coordination from visual perception, state estimation, trajectory prediction to closed-loop control, significantly improving the interception success rate and overall system robustness against proactively evasive intruding drones in complex dynamic environments. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1a This is a flowchart illustrating the intrusion drone interception method based on vision and data fusion provided in the embodiments of this application;

[0025] Figure 1b This is a schematic diagram of the modules of the intrusion drone interception method based on vision and data fusion provided in the embodiments of this application;

[0026] Figure 2 This is a schematic diagram of the structure of the intrusion drone interception system based on vision and data fusion provided in the embodiments of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] This application provides a method and system for intercepting intrusive drones based on vision and data fusion.

[0029] Specifically, the intrusion drone interception system based on vision and data fusion can be integrated into electronic devices, such as terminals and servers. The terminal can be an interception drone, a mobile phone, tablet, smart Bluetooth device, laptop, or personal computer (PC). When the terminal is an interception drone, the computation and control algorithms of the vision and data fusion-based intrusion drone interception system run on the terminal module (such as Xavier or 3588 boards) on the interception drone. The server can be a single server or a server cluster consisting of multiple servers.

[0030] The following sections provide detailed descriptions of each example. It should be noted that the sequence numbers of the following embodiments are not intended to limit the preferred order of the embodiments.

[0031] In this embodiment, a method for intercepting intrusive drones based on vision and data fusion is provided, such as... Figure 1a The specific process of this intrusion drone interception method based on vision and data fusion can be shown as follows:

[0032] 101. Obtain the current image of the intruding drone captured by the camera on the intercepting drone at the current time, and determine the current relative position between the intruding drone and the intercepting drone based on the pixel position information of the corresponding area of ​​the intruding drone in the current image, as well as the transformation relationship between the camera coordinate system and the intercepting drone's body coordinate system.

[0033] Intercepting drones refer to unmanned aerial vehicles used to actively approach, track, and physically intercept or drive away drones. For example, intercepting drones can be first-person view (FPV) drones, multi-rotor intercepting drones, fixed-wing intercepting drones, tilt-rotor / compound-wing drones, etc.

[0034] The camera refers to the airborne visual sensor installed on the interceptor drone, which is used to collect optical images of aerial targets in real time. The camera can be a visible light camera, an infrared thermal imager, or a depth camera. Its internal parameters (including focal length, principal point coordinates, distortion coefficients, etc.) and its installation pose (i.e., external parameters) relative to the interceptor body have been obtained through calibration.

[0035] The current time is synchronized with the timestamp of image acquisition to identify the temporal consistency of the data and ensure that information from multiple sources is fused and processed under a unified time reference.

[0036] Intruding drones refer to non-cooperative drone targets that enter protected airspace (such as airport clear zones, major event sites, military restricted areas, etc.) without permission. Their type, size, flight intentions and communication protocols are unknown, and they may have active evasion behaviors.

[0037] The current image refers to a single-frame digital image captured by the camera at the current time, containing the intruding drone, as the raw input data for visual perception.

[0038] In some embodiments, the current image may be obtained from a live video stream captured by a camera, specifically the latest frame of the video stream.

[0039] The region corresponding to the intrusion drone in the current image refers to the set of pixels that represent the projection of the intrusion drone onto the image plane, identified by target detection or segmentation of the current image. This region is usually represented by a bounding box, mask, or polygonal outline, and is used to extract the geometric features of the target.

[0040] Pixel location information refers to the numerical information extracted from the corresponding region that characterizes the spatial distribution of the intruding drone in the image coordinate system, including but not limited to the pixel coordinates (u,v) of the center point of the region, and the width and height of the region (in pixels).

[0041] The transformation relationship between the camera coordinate system and the interceptor body coordinate system refers to the mathematical model that describes the spatial pose relationship between the camera coordinate system and the interceptor drone body coordinate system. It usually consists of a 3×3 rotation matrix and a 3D translation vector, which is used to transform three-dimensional points in the camera coordinate system to the interceptor body coordinate system. This transformation relationship is determined by calibration before system deployment and is regarded as a known constant during operation.

[0042] The current relative position refers to the position vector of the intruding drone relative to the interceptor drone at the current time. It is expressed in the coordinate system of the interceptor drone and usually includes three components: the distance in the forward (x), lateral (y), and vertical (z) directions, in meters. This relative position is obtained by combining visual ranging and coordinate transformation and is the basic input for subsequent global positioning and interception control.

[0043] In some embodiments, based on monocular vision and combined with target physical size, focal length, and pixel information, high-precision relative positioning is achieved through internal / external parameter backprojection, without the need for depth sensors or cooperative signals, significantly improving the perception reliability and autonomy of close-range interception. The pixel position information of the intruding drone in the corresponding area of ​​the current image includes the center point pixel position and the area pixel width;

[0044] Based on the pixel location information of the intruding drone in the corresponding area of ​​the current image, and the transformation relationship between the camera coordinate system and the interceptor's body coordinate system, the current relative position between the intruding drone and the interceptor drone is determined, including:

[0045] Obtain the current focal length of the camera, as well as the physical dimensions of the intruding drone;

[0046] The intermediate ranging parameters are determined based on the product of the current focal length and physical size data;

[0047] Divide the intermediate ranging parameter by the region pixel width to obtain the distance information between the camera on the intercepting drone and the intruding drone.

[0048] Based on distance information and the camera's intrinsic parameter matrix, the center point pixel position is back-projected onto the camera coordinate system to obtain the three-dimensional position of the intrusion drone in the camera coordinate system. The intrinsic parameter matrix represents the camera's internal optical parameters, including focal length and principal point offset.

[0049] Based on the extrinsic parameter matrix of the camera relative to the interceptor's body coordinate system, the three-dimensional position is transformed to the interceptor's body coordinate system to obtain the current relative position between the intruding drone and the interceptor drone. The extrinsic parameter matrix represents the spatial pose relationship between the camera coordinate system and the interceptor's body coordinate system, including rotation and translation components.

[0050] The center point pixel position refers to the pixel coordinates of the geometric center of the corresponding area of ​​the intruding drone in the current image in the image coordinate system, usually represented as (u,v), where u is the horizontal pixel index and v is the vertical pixel index, used to characterize the projection center of the target on the image plane.

[0051] Region pixel width refers to the number of pixels occupied by the corresponding region of the intruding drone in the current image in the horizontal direction. It is usually taken as the width value of the bounding box and is used to reflect the apparent scale of the target in the image.

[0052] The current focal length refers to the effective focal length of the camera at the current imaging time, measured in pixels or millimeters. If the camera is a zoom lens, the focal length can be dynamically obtained based on the current zoom state. This parameter is a key optical parameter bridging the image scale and the real-world scale.

[0053] Physical size data refers to the known or pre-defined characteristic dimensions of the intrusion drone in the real world, such as rotor width, fuselage length, or height.

[0054] The intermediate ranging parameter is a constant obtained by multiplying the current focal length by the physical size data of the intruding drone.

[0055] Distance information refers to the straight-line distance (i.e., depth) between the optical center of the camera on the intercepting drone and the intruding drone, in meters; it is obtained by dividing the intermediate ranging parameter by the area pixel width, reflecting the longitudinal position of the intruding drone in the camera coordinate system.

[0056] The intrinsic parameter matrix of a camera refers to a 3×3 matrix describing the internal optical characteristics of the camera, usually denoted as K, and its form is:

[0057] ,in, , The focal length (in pixels) is the focal length in the x and y directions. , The master point offset (i.e., image center offset) is a matrix used to project three-dimensional spatial points onto the image plane, or conversely, to back-project pixels onto the normalized camera coordinate system.

[0058] The camera coordinate system refers to a right-handed rectangular coordinate system with the camera's optical center as the origin, the optical axis as the z-axis, the x-axis to the right, and the y-axis downward. It is used to describe the spatial position of the target (i.e., the intruding drone) relative to the camera.

[0059] Three-dimensional position refers to the spatial coordinates of the intrusion drone in the camera coordinate system. ,in, For the aforementioned distance information, It is calculated by back-projection from the center pixel position using the intrinsic parameter matrix.

[0060] The interceptor's body coordinate system refers to the reference coordinate system fixed to the interceptor drone's body. It is usually based on the drone's center of gravity as the origin, with the x-axis pointing forward, the y-axis pointing to the right wing, and the z-axis pointing vertically downward (or as defined by aviation conventions). It is used to describe the relative spatial relationship between the airborne sensors and the target.

[0061] The extrinsic parameter matrix refers to the rigid body transformation matrix required to transform points in the camera coordinate system to the interceptor's body coordinate system. It consists of a 3×3 rotation matrix R and a 3D translation vector t, and is usually represented as a homogeneous transformation matrix.

[0062] This matrix is ​​determined through calibration before system deployment and is used to achieve multi-sensor coordinate alignment.

[0063] Understandably, this application employs a lightweight YOLO (YouOnly Look Once) V5s network model with deep pruning and quantization optimization as the object detector, which is used to detect intruding drones. This model significantly reduces computational complexity and memory usage while maintaining high recall. For each frame of the input image, the network can output the bounding box information of the intruding drone in the image pixel coordinate system in real time (i.e., the pixel position information of the corresponding region of the intruding drone in the current image), including the center pixel position. ,width and height The bounding box, along with a high confidence score, accurately defines the location of the intruding drone in the current two-dimensional image, providing a foundation for subsequent three-dimensional information processing.

[0064] After obtaining the center pixel position of the corresponding region of the intruding drone in the current image, it needs to be transformed into three-dimensional space. This transformation depends on an accurate camera model. This application uses a pinhole camera model, and the camera's intrinsic parameter matrix is ​​obtained through pre-calibration. and the extrinsic parameter matrix relative to the interceptor's body coordinate system Target depth information Estimating the distance from the optical center of the camera on the intercepting drone to the intruding drone is key to solving the monocular vision ranging problem. This application introduces the prior physical dimensions of the intruding drone (its average width) This involves using physical size data to overcome scale uncertainty. Based on the principle of similar triangles, the target's depth information... The region pixel width corresponding to the intruding drone in the current image Inversely proportional, that is ,in The focal length of the camera (included in the intrinsic parameter matrix) (In Chinese). This method provides sufficiently accurate depth estimation when the distance to the intruding drone is neither extremely far nor extremely close. Subsequently, using a camera model, the center point pixel position is determined. Three-dimensional position projected onto the camera coordinate system :

[0065] .

[0066] Then, using the calibrated extrinsic parameter matrix, the three-dimensional position is transformed to the interceptor's body coordinate system to obtain the current relative position. :

[0067] .

[0068] 102. Based on the current relative position and the current pose data of the intercepting drone in the global coordinate system, determine the current absolute position of the intruding drone in the global coordinate system.

[0069] The global coordinate system refers to an inertial reference coordinate system fixed on the Earth's surface or mission area, used to uniformly describe the absolute position and orientation of all aircraft in space. In this application, the global coordinate system can be a geographic coordinate system (such as WGS-84) or a local East-North-Up (ENU) rectangular coordinate system, whose origin is usually set at the mission start point or control center, with the Z-axis pointing vertically upward (or downward, as agreed), and the X and Y axes pointing east and north, respectively.

[0070] Current pose data refers to the complete spatial state information of the intercepting drone in the global coordinate system at the current time, including its current position (usually represented as three-dimensional coordinates). (or latitude and longitude) and current attitude (usually determined by three-axis attitude angles—roll angle) Pitch angle and yaw angle —or the corresponding rotation matrix / quaternion representation), this data is output in real time by the integrated navigation system (such as the Global Navigation Satellite System (GNSS) / Inertial Measurement Unit (IMU) fusion module) carried by the interceptor drone.

[0071] The current absolute position refers to the three-dimensional spatial coordinates of the intruding drone in the global coordinate system. The location is used to characterize the real geographical location of the intruding UAV in the mission airspace. This location is calculated by transforming the coordinate system by combining the current relative position of the intruding UAV with the current pose data of the intercepting UAV itself. It serves as a unified spatial reference for subsequent trajectory prediction, multi-aircraft coordination, and interception decisions.

[0072] In some embodiments, a precise mapping from local perception to the global coordinate system is achieved, providing a unified spatial reference and significantly improving positioning consistency and multi-aircraft collaborative capabilities over a large airspace. The current pose data of the interceptor UAV in the global coordinate system includes the current three-axis attitude angles of the interceptor UAV and its current position in the global coordinate system;

[0073] Based on the current relative position and the current pose data of the intercepting drone in the global coordinate system, determine the current absolute position of the intruding drone in the global coordinate system, including:

[0074] Based on the current three-axis attitude angles, construct the rotation matrix from the interceptor's body coordinate system to the global coordinate system;

[0075] Transform the current relative position to the global coordinate system using a rotation matrix to obtain the first global coordinate component;

[0076] Add the first global coordinate component to the current position to obtain the current absolute position of the intruding drone in the global coordinate system.

[0077] Among them, the current three-axis attitude angles refer to the three Euler angles of the interceptor UAV relative to the global coordinate system at the current time, including yaw (rotation about the vertical axis, representing the direction of the nose), pitch (rotation about the lateral axis, representing the nose-up / nose-down state), and roll (rotation about the forward axis, representing the left and right tilt state); these three angles together describe the spatial orientation of the interceptor's body coordinate system relative to the global coordinate system.

[0078] The rotation matrix is ​​a 3×3 orthogonal matrix calculated from the current three-axis attitude angles in a specific rotation order (such as ZYX or Yaw-Pitch-Roll). It is used to rotate and transform vectors in the interceptor's body coordinate system (such as the current relative position) to the global coordinate system; this matrix satisfies R ∈ R3×3, and This ensures the geometric fidelity of coordinate transformation.

[0079] The first global coordinate component refers to the displacement component in the global coordinate system direction obtained after transforming the current relative position vector of the intruding drone in the interceptor's body coordinate system through the above-mentioned rotation matrix (i.e., the relative position after direction alignment), which has not yet been superimposed on the global position of the interceptor drone itself; this component reflects the offset of the intruding drone relative to the interceptor drone in the global direction and is an intermediate variable for calculating the absolute position.

[0080] Understandably, the current relative position of the intruding drone in the interceptor's coordinate system obtained solely from visual information is insufficient to support global interception decisions. This application converts the intruding drone's current relative position into global absolute coordinates by deeply fusing the interceptor drone's own global navigation information. The GNSS receiver onboard the interceptor drone provides its global position in the global coordinate system. The IMU provides its high-frequency three-axis attitude angles, namely roll angles. Pitch angle and yaw angle The rotation matrix from the body coordinate system constructed from the attitude angles to the global coordinate system is used. This allows the calculation of the intrusion drone's current absolute position in the global coordinate system. :

[0081] .

[0082] 103. Based on the current absolute position and the absolute position of the intruding drone at the previous time, calculate the current speed information of the intruding drone, and obtain the current observed motion state based on the current speed information and the current absolute position.

[0083] The absolute position of the intruding drone at the previous time refers to its three-dimensional spatial coordinates in the global coordinate system at the moment one sampling period (or control period) prior to the current time, usually denoted as . This position can be derived from the absolute position calculated in the previous cycle using the same method (such as visual fusion + pose transformation), and is used to jointly estimate the motion state of the intruding drone with the current absolute position.

[0084] Current velocity information includes the velocity vector and acceleration of the intruding drone at the current time.

[0085] It is understandable that by differentiating the absolute positions of consecutive frames in the global coordinate system, the velocity vector of the intruding drone can be preliminarily estimated. .

[0086] The current observed motion state consists of the intruding UAV's current absolute position, as well as its velocity vector and acceleration at the current time, typically expressed as: .

[0087] 104. Obtain the current predicted motion state of the intruding drone at the current time, use the current predicted motion state as a priori estimate, use the current observed motion state as a measurement input, and update the state through the Kalman filter algorithm to obtain the current corrected motion state.

[0088] The current predicted motion state refers to the estimated motion state of the intruding UAV obtained by extrapolation based on the historical corrected motion state of the previous moment and the system dynamic model (such as the constant velocity model) at the current time step, which serves as the prior input for the Kalman filter.

[0089] Prior estimation refers to the state vector predicted from the previous time state based solely on the system's dynamic model before incorporating newly observed motion states at the current time. It is denoted as... It is used to characterize the "predictive belief" about the current state.

[0090] Measurement inputs are used to correct prior estimates.

[0091] The Kalman filter algorithm is a recursive optimal estimation algorithm suitable for linear Gaussian systems. It outputs the optimal estimate of the state in the sense of minimum mean square error by fusing the system dynamic model (prediction) and sensor observations (update). Its core steps include time update (prediction) and measurement update (correction).

[0092] The current corrected motion state refers to the posterior state estimate obtained after updating the measurements using Kalman filtering, denoted as... It is the optimal state estimation result that integrates model predictions and current observation information, and is used for subsequent trajectory prediction and control decisions.

[0093] In some embodiments, while suppressing jitter caused by visual noise, scale jumps, and environmental disturbances, the system maintains a sensitive response to target maneuvers, significantly improving the accuracy, smoothness, and robustness of state estimation, providing a reliable foundation for high-confidence trajectory prediction and precise interception. Obtaining the current predicted motion state of the intruding drone at the current time includes:

[0094] Obtain the historical corrected motion state of the intruding drone at the previous time point;

[0095] Based on the constant velocity motion model, the position and velocity components in the historical corrected motion state are extrapolated to obtain the prior state estimate of the intruding drone at the current time, which is used as the current predicted motion state.

[0096] Based on a constant velocity motion model and combined with preset process noise covariance parameters, the state estimation error covariance of the intruding drone in the previous time is propagated to obtain the prior error covariance corresponding to the current predicted motion state.

[0097] Using the current predicted motion state as a priori estimate and the current observed motion state as measurement input, the state is updated using the Kalman filter algorithm to obtain the current corrected motion state, including:

[0098] Based on the preset observation model, the current predicted motion state is mapped to the same measurement space as the current observed motion state to obtain the predicted measurement value;

[0099] Calculate the measurement residual between the current observed motion state and the predicted measurement value;

[0100] The Kalman gain is calculated based on the prior error covariance, the observation model, and the preset measurement noise covariance parameters.

[0101] The prior state estimate is corrected by the product of the Kalman gain and the measurement residual to obtain the posterior state estimate, and the posterior state estimate is used as the current corrected motion state.

[0102] The error covariance of the state estimate is updated based on the Kalman gain and the prior error covariance.

[0103] Understandably, the current observed state of motion is defined as:

[0104] .

[0105] The constant velocity motion model is based on the assumption of constant velocity (CV), which states that the intruding drone moves at an approximately constant velocity over a short period of time. This assumption is reflected in the state transition matrix. superior. The matrix is ​​responsible for extrapolating the motion state (position and velocity) from the previous time step to the current predicted motion state. Simultaneously, a process noise covariance matrix is ​​introduced. This is used to quantify the uncertainty of this motion model (e.g., the real, unknown acceleration disturbances of the intruding drone). The larger the value, the more the filter trusts new measurements, resulting in a faster response but also higher noise. The smaller the value, the more the filter trusts the internal model, resulting in a smoother but potentially lagging output. Adjusting... This value allows for a balance between the newly measured response speed and the smoothness of the historical motion trajectory.

[0106] The observation model runs parallel to the constant velocity motion model. The current observed motion state... The information comes from the vision-navigation fusion module and typically contains the measured current absolute position of the intruding drone. Observation matrix Its function is to transform high-dimensional state vectors Mapped to a lower-dimensional observation space. In this example, The matrix structure "extracts" the position component from the state vector to match the current observed motion state. Compare them.

[0107] Based on the above model, the Kalman filter recursively performs the following two steps in each processing cycle:

[0108] 1. Prediction Steps: Utilize the optimal estimated state from the previous time step. That is, the historical corrected motion state and state transition matrix. Predict the prior state at the current time. and its error covariance The calculation formula is:

[0109] ;

[0110] ;

[0111] 2. Update steps: Compare the current predicted motion state (prior state estimate) with the current observed motion state at the current time. The fusion process is performed first. The Kalman gain is then calculated. It weighs the prediction error covariance and observation noise to determine the reliability of the current observed motion state in this update. Then, using... Multiply by the measurement residual ( This is used to correct the prior state and obtain the optimal estimate of the posterior state at the current time. and update the error covariance. This prepares for the next iteration. The calculation formula is:

[0112]

[0113]

[0114]

[0115] in It is the measurement noise covariance parameter, representing the measurement error of sensors (such as cameras and IMUs).

[0116] By continuously repeating the "prediction-update" cycle, the filter can effectively fuse multi-source sensor data that is sometimes delayed and noisy, and suppress measurement noise based on the motion model, ultimately outputting a smooth, accurate, and real-time target six-degree-of-freedom motion state matrix. That is, the optimal estimate of the position and velocity components in the current observed motion state Z. That is, the posterior state estimate is obtained, and the posterior state estimate is used as the current corrected motion state.

[0117] It contains the precise global position and instantaneous velocity vectors of the intruding drone, which can provide high-quality, unbiased input for the subsequent LSTM (Long Short Term Memory) neural network trajectory prediction module. It fundamentally solves the core pain points of inaccurate monocular vision ranging and large jitter in state estimation, and provides perception data for the entire autonomous interception system.

[0118] 105. Based on the current corrected motion state of the intruding drone and its historical corrected motion state in the historical time period, predict the motion trajectory of the intruding drone in the future prediction time domain.

[0119] In some embodiments, the limitations of traditional constant speed / uniform acceleration models are overcome, significantly improving trajectory prediction for high-speed irregular maneuvers such as sharp turns, dives, and hovering. By fusing wind field disturbance information, the robustness of the model under complex weather conditions is enhanced, providing a reliable basis for generating forward-looking and executable interception trajectories. The current speed information of the intruding UAV is represented in the form of a velocity vector.

[0120] Based on the current corrected motion state of the intruding drone and its historical corrected motion state in historical time periods, the trajectory of the intruding drone in the future prediction time domain is predicted, including:

[0121] Obtain wind field parameters of the intrusion environment;

[0122] Based on the wind field parameters of the intrusion environment, the current velocity vector, the current corrected motion state of the intrusion drone, and its historical corrected motion state in the historical time period, an input feature vector is constructed.

[0123] The input feature vector is fed into the Long Short-Term Memory Neural Network model. The Long Short-Term Memory Neural Network model extracts temporal features and updates the dynamic state of the input features through its internal forget gate, input gate and output gate, and outputs the motion trajectory of the intrusion drone in the future prediction time domain.

[0124] In some embodiments, without the need for external meteorological equipment, the interceptor drone's own sensors can be used to perceive the local wind field online, effectively compensating for the wind disturbance in trajectory prediction, improving the prediction accuracy of wind-driven drifting or maneuvering targets, thereby enhancing the foresight and reliability of interception decisions, and obtaining wind field parameters of the intrusion environment, including:

[0125] Obtain the current airspeed data output by the airspeed meter on the interceptor drone. The current airspeed data represents the speed of the interceptor drone relative to the surrounding air in the interceptor's own coordinate system.

[0126] Obtain the current ground speed data output by the positioning system on the interceptor drone. The current ground speed data represents the speed of the interceptor drone relative to the ground in the global coordinate system.

[0127] Obtain the current three-axis attitude angles of the interceptor drone, and construct the rotation matrix from the interceptor's body coordinate system to the global coordinate system based on the current three-axis attitude angles;

[0128] The current airspeed data is transformed to the global coordinate system based on the rotation matrix to obtain the global airspeed data;

[0129] The wind field parameters of the intrusion environment are calculated based on the vector difference between global airspeed data and current ground speed data.

[0130] Understandably, after obtaining the target UAV's six-DOF state sequence smoothed by Kalman filtering—that is, the current corrected motion state of the intruding UAV—it is necessary to predict its future trajectory. Traditional interception systems typically use simple linear dynamic models (such as uniform velocity or uniform acceleration models) for extrapolation prediction. This method is applicable to targets performing conventional flight, but when facing targets that realize they are being tracked and take high-intensity, nonlinear escape maneuvers (such as sudden turns, spiral climbs, or dives), the prediction error increases exponentially, leading to interception path failure. To address this core challenge, this application abandons the assumptions of traditional linear models and introduces a dynamic trajectory prediction model based on a Long Short-Term Memory (LSTM) neural network. This model can autonomously learn its complex maneuvering patterns from the target's historical motion data and generate a probabilistic trajectory distribution for future moments.

[0131] The input to this LSTM dynamic prediction model is a multi-dimensional, time-dependent feature sequence, designed to comprehensively capture the internal and external factors influencing the target's motion. The core of the input sequence is the historical corrected motion state of the intruding drone over a historical time period, specifically the past k time steps (e.g., the past 2 seconds, corresponding to 50 frames). Each state vector All of these include the absolute position and velocity of the intruding drone in the global coordinate system. Furthermore, this application also includes wind field parameters of the intrusion environment. As an important feature input model, wind field is an external factor that significantly affects the motion of UAVs. Its value can be estimated jointly by the difference between the intercepting UAV's own IMU data, airspeed indicator, and ground speed output by the navigation system, and is expressed as follows: The current velocity vector direction of the intruding drone. This is also incorporated into the input features as a key indicator reflecting the instantaneous motion intent. Therefore, at each time step t, the model's input feature vector... It is a fusion of state, environment, and intent information: .

[0132] LSTM networks are well-suited for handling time series prediction problems due to their unique gating mechanism. Its core lies in three gates: the forget gate... Input gate and output gate and cell state and hidden state The calculations for these doors are as follows:

[0133]

[0134]

[0135]

[0136]

[0137]

[0138]

[0139] in, It is the sigmoid activation function. This represents element-wise multiplication. and Let be the parameter matrix and vector to be trained. The forget gate determines which information is discarded from the previous cell state, the input gate determines which new information is stored in the cell state, and the output gate determines the output of the current hidden state based on the cell state. Through this mechanism, LSTM can selectively remember and forget long-term and short-term dependencies, thereby effectively capturing the complex transition patterns of target drones from level flight to escape.

[0140] This LSTM network is trained through supervised learning on a large amount of trajectory data containing various drone maneuvers (including evasion, escape, and hovering), based on the input historical sequences. This is used to predict the target state T time steps ahead (corresponding to the next 3 to 5 seconds). The network's output layer is typically a fully connected layer, and its output is not a single, deterministic trajectory, but rather a probability distribution of the future trajectory. There are two possible implementation methods here:

[0141] 1) Define the output as the Gaussian distribution parameters of the target location at a series of future time points, that is, for each future time step The network outputs a mean. and a covariance matrix The covariance matrix characterizes the uncertainty of the prediction. This means that the output of the model is a probability "spatiotemporal cone" that diverges over time.

[0142] 2) Use multiple parallel fully connected heads to predict N most likely future trajectories, forming a trajectory set. Each trajectory is a sequence of future state points.

[0143] 106. Based on the motion trajectory, control the intercepting drone to perform the interception mission against the intruding drone.

[0144] In some embodiments, the system effectively addresses the high-speed maneuvering and evasive behavior of intruding drones while also considering control smoothness and system energy consumption. This significantly improves the success rate and mission reliability of autonomous interception in complex dynamic environments. Based on the motion trajectory, the system controls the intercepting drone to perform the interception mission against the intruding drone, including:

[0145] Obtain the status data of the intercepted drone;

[0146] Construct candidate control sequences for intercepting drones in the future prediction time domain;

[0147] Using state data as initial conditions and candidate control sequences, the state evolution of the interceptor drone in the future prediction time domain is simulated to generate the corresponding predicted state sequence.

[0148] The motion trajectory, candidate control sequence, and predicted state sequence are substituted into the preset objective function to calculate the corresponding interception cost. The objective function is used to comprehensively evaluate the interception mission cost and control execution cost of intercepting the UAV.

[0149] The optimal control sequence is determined from the candidate control sequences based on the interception cost;

[0150] Based on the optimal control sequence, the interceptor drone is controlled to perform the interception mission against the intruding drone.

[0151] Understandably, after the trajectory prediction module generates a probabilistic spatiotemporal description of the intruding drone's future motion, it is necessary to consider how to utilize this information to calculate the pursuit path with the highest success rate under the premise of physical feasibility. This is a typical nonlinear, constrained optimization control problem. This application adopts a rolling optimization framework based on Model Predictive Control (MPC), which tightly couples prediction and decision-making to generate dynamically optimal interception commands.

[0152] The core idea of ​​the MPC framework is to solve an open-loop optimal control problem within a "future prediction time domain" based on the intercepting drone's state data at the current time, in each control cycle (typically milliseconds). First, a mathematical model for interception optimization needs to be constructed. The system's state variables... Defined as the state of the intercepted drone itself, typically including its three-dimensional position. ,speed Attitude angle And angular velocity, etc. Controlled variables. These are the low-level inputs of the flight control system, such as throttle thrust. and the deflection angle of the three-axis control surfaces The dynamic model of the system is described by a nonlinear differential equation: This model encapsulates the aerodynamic characteristics and kinematic constraints of intercepting drones.

[0153] objective function The design of the control system is the soul of the decision-maker; it must quantify the concept of "optimal interception." This application designs a multi-objective cost function, aiming to simultaneously minimize the miss rate, maximize the interception probability, and ensure a smooth and efficient control process. The function takes the following form:

[0154]

[0155] in:

[0156] This is for predicting the time domain. The control input sequence within.

[0157] The term represents the predicted state sequence, indicating the mean distribution of the predicted location of the intercepting drone and the trajectory of the intruding drone in the k-th future step. The difference between (from LSTM output) and the weight matrix. To impose a price on it.

[0158] It is the target prediction location covariance matrix The trace represents the uncertainty of the prediction at that moment. Coefficient Punishing the intruder in order to drive the interceptor drone to not only head towards the mean point of the intruder drone's trajectory during the pursuit, but also to choose a path with lower predictability uncertainty, thereby actively avoiding the extreme escape directions that the intruder drone may take, which greatly improves the robustness of the interception.

[0159] The cost of control is represented by the cost of interception, which is used to avoid using control commands that drastically change the state of motion.

[0160] This represents the cost of control increments, i.e., the cost of control execution, used to ensure that the generated commands are smooth and continuous, avoid severe turbulence of the control surfaces, and ensure flight stability.

[0161] Solving this optimal control problem must be done under strict constraints. These constraints include:

[0162] 1) Dynamic constraints: The state transitions of the system must follow the dynamic model described above. .

[0163] 2) Control constraints: The control input must be within its physical limits, i.e. .

[0164] 3) State constraints: The interceptor's state cannot exceed the safety envelope, such as maximum speed. Maximum overload And the minimum flight altitude to avoid collisions with ground obstacles, etc.

[0165] In each control cycle, the decision-maker will determine the current status of the intercepted drone. Target Prediction Distribution With various constraints as input, the system solves the constrained optimization problem online, obtaining a set of optimal control sequences which are then sent to the flight control system for execution. In the next control cycle, the system updates the state estimate and trajectory prediction using new sensor observation data and recalculates the entire optimization process based on the new information. This "rolling optimization, feedback correction" mechanism enables the decision-maker to continuously adapt to unmodeled dynamic disturbances such as the intruding UAV's maneuvers and wind field changes, maintaining the real-time and optimal nature of its decisions. Ultimately, it guides the interceptor to cut into the target's future escape cone center along an energy-optimal and robust path, creating the best conditions for a physical collision or net capture.

[0166] In some embodiments, precise closed-loop control from high-level trajectory planning to low-level execution is achieved, improving the tracking accuracy and dynamic response capability of the interception action. Based on the optimal control sequence, the interception drone is controlled to perform the interception task against the intruding drone, including:

[0167] Obtain the desired attitude angular rate command from the optimal control sequence;

[0168] Based on the deviation between the desired attitude angular rate command and the currently measured body angular rate of the intercepting UAV, a proportional-integral-derivative control algorithm is used to adjust and generate a control signal for driving the servo motor.

[0169] Control signals are sent to the flight control system of the interceptor drone to drive the interceptor drone to perform interception actions according to the optimal control sequence.

[0170] Understandably, the interception scenarios involved in this application place extreme demands on the real-time performance, accuracy, and robustness of the interceptor drone control, especially in the terminal approach phase, where any delay or error could lead to mission failure. To address this, this application designs a hierarchical, high-frequency real-time closed-loop control architecture, seamlessly switching to a pure optical terminal guidance mode during the final approach phase to solve the problem of interception failure caused by insufficient positioning accuracy in the navigation system.

[0171] The flight control system is the final executor of the entire interception operation. The input to the system is the current optimal control command output by the decision-maker MPC framework. The desired attitude angular rate command is the desired throttle and control surface inputs. However, the MPC output is typically a high-level force and torque command in the body coordinate system, or a desired acceleration vector. The primary task of the flight control system is to decompose these commands and translate them into control signals for the lower-level servo motors and engines. This process is accomplished through a series of inner-loop controllers, the most crucial of which is the angular rate controller. This controller employs a proportional-integral-derivative (PID) control algorithm and operates at an extremely high frequency. Its goal is to rapidly and error-free track the desired body angular rate calculated from the higher-level commands. Taking pitch rate control as an example, its control law is:

[0172]

[0173] in, The current measured pitch rate, , , These parameters are tuned specifically for this aircraft type. The angular rate controller outputs a direct PWM signal to drive the control surfaces, generating aerodynamic torque that forces the aircraft's attitude angular rate to converge rapidly to the desired value. The outer loop position and velocity controller is responsible for tracking the desired trajectory given by the MPC, guiding the inner loop by generating the desired attitude angle commands. This hierarchical control structure ensures that the interceptor can accurately execute the complex, time-varying pursuit maneuvers calculated by the MPC with minimal delay and overshoot.

[0174] In some embodiments, visual closed-loop terminal guidance without additional sensors is achieved, which significantly improves the accuracy and real-time performance of close-range collision interception. The pixel position information of the intruding drone in the corresponding area of ​​the current image includes the center point pixel position and the area pixel width.

[0175] Also includes:

[0176] When the distance between the intercepting drone and the intruding drone is less than a preset distance, based on the timestamp corresponding to the current image, at least one historical image within a preset time range is obtained, as well as the center pixel position of the corresponding area of ​​the intruding drone in each historical image.

[0177] The relative movement speed of the intruding drone is calculated based on the change of the center point pixel position in all historical images and the current image.

[0178] The approach rate of the intruding drone is calculated based on the rate of change of the region pixel width across all historical images and the current image.

[0179] Based on the relative speed of movement and the approach speed, terminal guidance commands are generated to guide the intercepting drone to adjust its course and speed to achieve collision interception.

[0180] Understandably, as the distance between the intercepting drone and the intruding drone falls below the preset distance, entering the final approach phase (typically within tens of meters), the navigation system's meter-level positioning accuracy is insufficient to meet the centimeter-level collision accuracy requirements. At this point, the system's control shifts to the optical end-of-line guidance module. The end-of-line guidance algorithm continuously receives the center point pixel position from the lightweight YOLO network. and region pixel width By calculating the movement rate of the intruding drone in the image plane. and dimensional change rate It can accurately estimate approach speed and relative movement speed (such as line-of-sight angular rate).

[0181] These visual features are directly input into a dedicated end-effector guidance law. This application preferably employs an improved proportional navigation (PN) method, whose instructions are perpendicular to the instantaneous line-of-sight (LOS). Its core formula is:

[0182] ;

[0183] in, For command acceleration perpendicular to the line of sight, For an effective navigation ratio constant greater than 2, Based on image size change rate Estimated approach rate, For pixel-based movement rate The estimated line-of-sight angular rate. The flight control system then sends this acceleration command. The interceptor drone is decomposed into roll and pitch commands. It always approaches the intruding drone tangentially along the optimal path to ensure the highest success probability and the lowest relative kinetic energy at the moment of final collision, thereby achieving autonomous interception of the intruding drone by the interceptor drone.

[0184] like Figure 1bAs shown, this application proposes a vision-data fusion-based method for intercepting intruding drones. First, a lightweight YOLO target detection algorithm is used to process the video stream from the drone's camera. This video stream is then combined with pose data provided by GPS / IMU, and a Kalman filter is used for data fusion to construct a precise current observed motion state of the intruding drone—a six-degree-of-freedom motion state matrix—to overcome the large ranging error of monocular vision. Subsequently, an LSTM neural network is used to establish a dynamic prediction model. Based on the intruding drone's historical trajectory, current velocity vector, and wind field parameters of the intrusion environment, the model predicts the drone's trajectory for the next few seconds, which can be understood as an escape path probability distribution. Based on this, a Model Predictive Control (MPC) framework is employed to continuously solve for the optimal pursuit path of the intercepting drone within its predicted future trajectory in the time domain, generating control commands. Upon receiving the commands, the flight control system drives the servo mechanism to execute interception maneuvers. During the terminal approach phase, it switches to pure optical terminal guidance (such as proportional guidance) and makes final precise adjustments based on image information, forming a real-time closed-loop control system from perception to execution.

[0185] Its core lies in constructing a complete closed-loop chain of "perception-decision-execution". Using interceptor drones as the physical carrier, it integrates multi-source sensors, edge computing units, and an adaptive flight control system to achieve dynamic tracking and intelligent interception of intruding drones through data-driven methods. Compared to traditional solutions, this system has significant advantages such as fast response speed, strong environmental adaptability, and high interception accuracy, making it particularly suitable for protecting critical facilities in complex electromagnetic environments.

[0186] In summary, this application improves the interception success rate.

[0187] To better implement the above methods, this embodiment will take a vision and data fusion-based intrusion drone interception system specifically integrated into an electronic device as an example to provide a detailed description of the method of this application embodiment.

[0188] For example, such as Figure 2 As shown, the intrusion drone interception system based on vision and data fusion may include a visual relative positioning unit 201, an absolute position determination unit 202, an observation state construction unit 203, a state correction unit 204, a trajectory prediction unit 205, and an interception control unit 206, as follows:

[0189] (a) Visual relative positioning unit 201.

[0190] The visual relative positioning unit is used to acquire the current image of the intruding drone captured by the camera on the intercepting drone at the current time, and determine the current relative position between the intruding drone and the intercepting drone based on the pixel position information of the corresponding area of ​​the intruding drone in the current image, as well as the transformation relationship between the camera coordinate system and the intercepting drone's body coordinate system.

[0191] (ii) Absolute position determination unit 202.

[0192] The absolute position determination unit 202 is used to determine the current absolute position of the intruding drone in the global coordinate system based on the current relative position and the current pose data of the intercepting drone in the global coordinate system.

[0193] (III) Observation state construction unit 203.

[0194] The observation state construction unit 203 is used to calculate the current speed information of the intruding drone based on the current absolute position and the absolute position of the intruding drone in the previous time, and to obtain the current observation motion state based on the current speed information and the current absolute position.

[0195] (iv) Status correction unit 204.

[0196] The state correction unit 204 is used to obtain the current predicted motion state of the intruding UAV at the current time, and use the current predicted motion state as a prior estimate and the current observed motion state as a measurement input. The state is updated by the Kalman filter algorithm to obtain the current corrected motion state.

[0197] (v) Trajectory prediction unit 205.

[0198] The trajectory prediction unit 205 is used to predict the trajectory of the intruding drone in the future prediction time domain based on the current corrected motion state of the intruding drone and its historical corrected motion state in the historical time period.

[0199] (vi) Interception control unit 206.

[0200] The interception control unit 206 is used to control the interception drone to perform the interception task against the intruding drone based on the motion trajectory.

[0201] Therefore, the embodiments of this application improve the interception success rate.

[0202] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0203] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the vision and data fusion-based intrusion drone interception methods provided in embodiments of this application.

[0204] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0205] Since the instructions stored in the storage medium can execute the steps in any of the vision and data fusion-based intrusion drone interception methods provided in the embodiments of this application, the beneficial effects that any of the vision and data fusion-based intrusion drone interception methods provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0206] According to one aspect of this application, a computer program product or computer program is provided, comprising a computer program / instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer program / instructions from the computer-readable storage medium and executes the computer program / instructions, causing the electronic device to perform the method provided in the above embodiments regarding the interception of intrusive drones based on vision and data fusion.

[0207] The foregoing has provided a detailed description of a method and system for intercepting intrusive drones based on vision and data fusion, as provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and its core ideas. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for intercepting intrusive drones based on vision and data fusion, characterized in that, The method includes: The current image of the intruding drone captured by the camera on the intercepting drone at the current time is obtained, and the current relative position between the intruding drone and the intercepting drone is determined based on the pixel position information of the corresponding area of ​​the intruding drone in the current image and the transformation relationship between the camera coordinate system and the intercepting drone body coordinate system. Based on the current relative position and the current pose data of the intercepting drone in the global coordinate system, the current absolute position of the intruding drone in the global coordinate system is determined; Based on the current absolute position and the absolute position of the intruding drone at the previous time, calculate the current speed information of the intruding drone, and based on the current speed information and the current absolute position, obtain the current observed motion state; The current predicted motion state of the intruding drone at the current time is obtained, and the current predicted motion state is used as a prior estimate, and the current observed motion state is used as a measurement input. The state is updated by Kalman filtering algorithm to obtain the current corrected motion state. Based on the current corrected motion state of the intruding drone and its historical corrected motion state in the historical time period, the motion trajectory of the intruding drone in the future prediction time domain is predicted. Based on the motion trajectory, the intercepting drone is controlled to perform the interception task against the intruding drone.

2. The method as described in claim 1, characterized in that, The step of obtaining the current predicted motion state of the intruding drone at the current time includes: Obtain the historical corrected motion state of the intruding drone at the previous time point; Based on the constant velocity motion model, the position and velocity components in the historical corrected motion state are extrapolated to obtain the prior state estimate of the intruding drone at the current time, which is used as the current predicted motion state. Based on the constant speed motion model and combined with the preset process noise covariance parameter, the state estimation error covariance of the intruding drone in the previous time is propagated to obtain the prior error covariance corresponding to the current predicted motion state. The step of using the current predicted motion state as a prior estimate and the current observed motion state as a measurement input, and updating the state through a Kalman filter algorithm to obtain the current corrected motion state includes: Based on a preset observation model, the current predicted motion state is mapped to the same measurement space as the current observed motion state to obtain the predicted measurement value; Calculate the measurement residual between the current observed motion state and the predicted measurement value; The Kalman gain is calculated based on the prior error covariance, the observation model, and the preset measurement noise covariance parameters. The prior state estimate is corrected based on the product of the Kalman gain and the measurement residual to obtain the posterior state estimate, and the posterior state estimate is used as the current corrected motion state. The error covariance of the state estimate is updated based on the Kalman gain and the prior error covariance.

3. The method as described in claim 1, characterized in that, The pixel position information of the intrusion drone in the corresponding area of ​​the current image includes the center point pixel position and the area pixel width; The step of determining the current relative position between the intrusion drone and the interceptor drone based on the pixel position information of the corresponding region of the intrusion drone in the current image, and the transformation relationship between the camera coordinate system and the interceptor body coordinate system, includes: Obtain the current focal length of the camera and the physical size data of the intruding drone; The intermediate ranging parameters are determined based on the product of the current focal length and the physical size data; Divide the intermediate ranging parameter by the region pixel width to obtain the distance information between the camera on the intercepting drone and the intruding drone. Based on the distance information and the intrinsic parameter matrix of the camera, the position of the center point pixel is back-projected onto the camera coordinate system to obtain the three-dimensional position of the intruding drone in the camera coordinate system. The intrinsic parameter matrix represents the internal optical parameters of the camera, including focal length and principal point offset. Based on the extrinsic parameter matrix of the camera relative to the interceptor's body coordinate system, the three-dimensional position is transformed to the interceptor's body coordinate system to obtain the current relative position between the intruding drone and the interceptor drone. The extrinsic parameter matrix represents the spatial pose relationship between the camera coordinate system and the interceptor's body coordinate system, including rotation and translation components.

4. The method as described in claim 1, characterized in that, The current pose data of the intercepting drone in the global coordinate system includes the current three-axis attitude angles of the intercepting drone and its current position in the global coordinate system; Determining the current absolute position of the intruding drone in the global coordinate system based on the current relative position and the current pose data of the intercepting drone in the global coordinate system includes: Based on the current three-axis attitude angles, construct the rotation matrix from the interceptor's body coordinate system to the global coordinate system; The current relative position is transformed to the global coordinate system using the rotation matrix to obtain the first global coordinate component; The first global coordinate component is added to the current position to obtain the current absolute position of the invading drone in the global coordinate system.

5. The method as described in claim 1, characterized in that, The current speed information of the intruding drone includes the current speed vector; The prediction of the intrusion drone's trajectory in the future prediction time domain based on its current corrected motion state and its historical corrected motion state in a historical time period includes: Obtain wind field parameters of the intrusion environment; Based on the wind field parameters of the intrusion environment, the current velocity vector, the current corrected motion state of the intrusion drone, and its historical corrected motion state in a historical time period, an input feature vector is constructed. The input feature vector is input into a long short-term memory neural network model. The long short-term memory neural network model extracts temporal features and updates the dynamic state of the input features through its internal forget gate, input gate and output gate, and outputs the motion trajectory of the intrusion drone in the future prediction time domain.

6. The method as described in claim 5, characterized in that, The acquisition of wind field parameters in the intrusion environment includes: The current airspeed data output by the airspeed meter on the interceptor drone is obtained, and the current airspeed data represents the speed of the interceptor drone relative to the surrounding air in the interceptor body coordinate system; The current ground speed data output by the positioning system carried by the intercepting drone is obtained, and the current ground speed data represents the speed of the intercepting drone relative to the ground in the global coordinate system; Obtain the current three-axis attitude angles of the interceptor drone, and construct a rotation matrix from the interceptor drone's body coordinate system to the global coordinate system based on the current three-axis attitude angles; Based on the rotation matrix, the current airspeed data is transformed to the global coordinate system to obtain global airspeed data; Based on the vector difference between the global airspeed data and the current ground speed data, the wind field parameters of the intrusion environment are calculated.

7. The method as described in claim 1, characterized in that, The step of controlling the interceptor drone to perform the interception task against the intruding drone based on the motion trajectory includes: Obtain the status data of the intercepted drone; Construct candidate control sequences for the intercepting drone in the future prediction time domain; Using the state data as initial conditions, and based on the candidate control sequence, the state evolution of the intercepting drone in the future prediction time domain is simulated to generate a corresponding predicted state sequence. The motion trajectory, the candidate control sequence, and the predicted state sequence are substituted into a preset objective function to calculate the corresponding interception cost. The objective function is used to comprehensively evaluate the interception task cost and control execution cost of the intercepting drone. Based on the interception cost, the optimal control sequence is determined from the candidate control sequences; Based on the optimal control sequence, the intercepting drone is controlled to perform the interception task against the intruding drone.

8. The method as described in claim 7, characterized in that, The step of controlling the interceptor drone to perform the interception task against the intruding drone based on the optimal control sequence includes: Obtain the desired attitude angular rate command from the optimal control sequence; Based on the deviation between the desired attitude angular rate command and the currently measured body angular rate of the intercepting UAV, the control signal for driving the servo motor is generated by adjusting the proportional-integral-derivative control algorithm. The control signal is sent to the flight control system of the intercepting drone to drive the intercepting drone to perform interception actions according to the optimal control sequence.

9. The method as described in claim 1, characterized in that, The pixel position information of the intrusion drone in the corresponding area of ​​the current image includes the center point pixel position and the area pixel width; Also includes: When the distance between the intercepting drone and the intruding drone is less than a preset distance, based on the timestamp corresponding to the current image, at least one historical image within a preset time range is obtained, as well as the center pixel position of the corresponding area of ​​the intruding drone in each historical image; The relative movement speed of the intruding drone is calculated based on the changes in the center point pixel position across all historical and current images. The approach rate of the intruding drone is calculated based on the rate of change of the region's pixel width across all historical and current images. Based on the relative movement rate and the approach rate, terminal guidance commands are generated to guide the intercepting drone to adjust its course and speed to achieve collision interception.

10. An intrusion drone interception system based on vision and data fusion, characterized in that, The system includes: The visual relative positioning unit is used to acquire the current image of the intruding drone captured by the camera on the intercepting drone at the current time, and determine the current relative position between the intruding drone and the intercepting drone based on the pixel position information of the corresponding area of ​​the intruding drone in the current image, and the transformation relationship between the camera coordinate system and the intercepting drone body coordinate system. An absolute position determination unit is used to determine the current absolute position of the intruding drone in the global coordinate system based on the current relative position and the current pose data of the intercepting drone in the global coordinate system. The observation state construction unit is used to calculate the current speed information of the intruding drone based on the current absolute position and the absolute position of the intruding drone in the previous time, and to obtain the current observation motion state based on the current speed information and the current absolute position. The state correction unit is used to obtain the current predicted motion state of the intruding drone at the current time, and use the current predicted motion state as a prior estimate, the current observed motion state as a measurement input, and update the state through the Kalman filter algorithm to obtain the current corrected motion state. The trajectory prediction unit is used to predict the trajectory of the intruding drone in the future prediction time domain based on the current corrected motion state of the intruding drone and its historical corrected motion state in the historical time period. An interception control unit is used to control the interception drone to perform an interception task against the intruding drone based on the motion trajectory.

Images