Intercepting unmanned aerial vehicle terminal guidance method and device based on ground visual servo guidance

Abstract

This application provides a terminal guidance method and apparatus for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance, relating to the field of UAV defense and countermeasure technology. The method includes: synchronously acquiring observation data of target UAVs based on ground-deployed visual sensors; calculating the estimated position values ​​and estimation error ranges of each target UAV; generating guidance reference points to guide at least one interceptor UAV to search for at least one target UAV; generating search guidance commands for each interceptor UAV; optimizing the search formation of the interceptor UAVs; generating control commands for the interceptor UAVs; and driving the interceptor UAVs to search for the target UAVs. The method and apparatus provided in this application construct an air-ground cooperative architecture combining ground-based visual servo guidance and airborne autonomous search, significantly improving the response speed and interception success rate against highly maneuverable targets in complex electromagnetic and low-altitude environments. The cost is controllable, possessing extremely high practical value and economic efficiency.

Description

Technical Field

[0001] This application relates to the field of drone defense and countermeasure technology, and in particular to a terminal guidance method and device for intercepting drones based on ground visual servo guidance. Background Technology

[0002] With the popularization of drone technology and the evolution of combat styles, especially the increasingly serious threat posed by low-altitude, slow-speed, and small (referred to as "low-slow-small") drones and first-person view (FPV) drones with low-altitude penetration capabilities, the existing air defense system is facing great challenges.

[0003] Such targets typically have small radar cross-sections, low flight altitudes, and flexible maneuvering trajectories, making them highly susceptible to evading detection and tracking by traditional radar systems by utilizing terrain or buildings for cover. In complex low-altitude environments, radar guidance systems are severely hampered by ground clutter and multipath effects, making it difficult to achieve stable and accurate tracking of these small targets. Faced with such low-cost, high-consumption asymmetric air threats, using expensive medium- and long-range air defense missiles for interception presents a serious cost-effectiveness imbalance, while conventional man-portable air defense systems are ineffective due to their short detection range and long reaction time.

[0004] Therefore, how to develop a terminal guidance technology based on a low-cost platform with high response speed and high precision to effectively destroy highly maneuverable UAV targets has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0005] This application provides a terminal guidance method and device for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance. It provides a terminal guidance technology based on a low-cost platform with high response speed and high precision, enabling effective destruction of highly maneuverable UAV targets.

[0006] This application provides a terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance, including: Based on the simultaneous acquisition of observation data of at least one target UAV by at least two ground-deployed visual sensors, the position estimates of each target UAV and the estimation error range characterizing the uncertainty of the position estimates are calculated. Based on the estimated position and the estimated error range, a guidance reference point is generated to guide at least one interceptor drone to search for the at least one target drone, and search guidance instructions for each interceptor drone are generated based on the guidance reference point. The search formation of the intercepting drone is optimized based on the estimated error range, and control commands for the intercepting drone are generated based on the search guidance commands to drive the intercepting drone to search for the target drone.

[0007] In some embodiments, the step of simultaneously acquiring observation data from at least one target UAV based on at least two ground-deployed visual sensors, and calculating the position estimates of each target UAV and the estimation error range characterizing the uncertainty of the position estimates, includes: Construct a matching cost matrix by integrating at least two of the following: geometric topology sorting constraints, back ray hard constraints, depth consistency constraints, and spatial distance threshold constraints; The line-of-sight relationships of the target UAV observed by the at least two visual sensors are matched based on the matching cost matrix to determine the line-of-sight correspondence; the line of sight is generated based on the observation data. Based on the line-of-sight correspondence, calculate the point pair with the closest spatial distance on each pair of matching lines of sight, and use the midpoint of the line connecting the point pairs as the estimated position of the target UAV; The estimated error range is determined based on the gap error of the matched line of sight and the physical measurement error of the vision sensor.

[0008] In some embodiments, the method further includes: In the event that matching fails due to occlusion of any target UAV, a primary visual sensor is determined among the at least two visual sensors, and a virtual guide point for the target UAV is generated along the observation direction of the primary visual sensor toward the target UAV.

[0009] In some embodiments, generating guidance reference points to guide at least one interceptor drone to search for the at least one target drone based on the position estimate and the estimation error range includes: An uncertainty sphere is generated with the estimated position as the center and the estimated error range as the radius. If the minimum distance between the intercepting drone and the uncertainty sphere is greater than a preset mode switching distance threshold, a close-range search mode is adopted, and the center of the uncertainty sphere is used as the guiding reference point. When the minimum distance between the intercepting drone and the uncertainty sphere is less than or equal to a preset mode switching distance threshold, a close-range detection mode is adopted to generate multiple uniformly distributed sub-reference points inside the uncertainty sphere, and the sub-reference points are used as guiding reference points.

[0010] In some embodiments, generating search guidance instructions for each intercepting drone based on the guidance reference point includes: Construct a task allocation cost matrix that includes at least one of path cost, angle cost, and load penalty cost; the path cost is determined based on the distance between the current position of the intercepting UAV and the guidance reference point; the angle cost is determined based on the deviation between the current observation angle of the intercepting UAV and the observation angle corresponding to the guidance reference point; the load penalty cost is determined based on the number of intercepting UAVs allocated in the search area defined by the estimation error range. Based on the task allocation cost matrix, the guidance reference point is assigned to each intercepting drone, and the position information of the guidance reference point is used as the search guidance instruction for each intercepting drone.

[0011] In some embodiments, optimizing the search formation of the intercepting drone based on the estimated error range includes: A fitness function is constructed based on at least one of the following: the total distance between the intercepting UAV and the guiding reference point, the penalty for the number of times the search formation of the intercepting UAV overlaps, the penalty for the coverage uniformity of the search area defined by the estimation error range, and the penalty for the maneuver smoothness of the intercepting UAV. Based on the fitness function, a particle swarm genetic hybrid intelligent optimization algorithm is used to optimize the search formation of the intercepting drone.

[0012] In some embodiments, generating the control commands for intercepting the drone based on the search guidance commands includes: The search guidance command is sent to the controller of the intercepting drone. The controller determines the position information of the guidance reference point based on the search guidance command, determines the distance deviation based on the current position of the intercepting drone and the position information of the guidance reference point, and generates the control command based on the distance deviation. The controller is an enhanced robust proportional-integral-derivative controller.

[0013] This application provides a ground-based visual servo-guided terminal guidance device for intercepting unmanned aerial vehicles, comprising: The position estimation module is used to simultaneously acquire observation data of at least one target UAV based on at least two ground-deployed visual sensors, calculate the position estimate of each target UAV and the estimation error range characterizing the uncertainty of the position estimate; The air-ground coordination module is used to generate guidance reference points for at least one interceptor drone to search for at least one target drone based on the position estimate and the estimation error range, and to generate search guidance instructions for each interceptor drone based on the guidance reference points. The search guidance module is used to optimize the search formation of the intercepting drone based on the estimated error range, generate control commands for the intercepting drone based on the search guidance commands, and drive the intercepting drone to search for the target drone.

[0014] This application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance.

[0015] This application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance.

[0016] The terminal guidance method and device for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance provided in this application achieves precise tracking of the target UAV by acquiring observation data from a low-cost visual sensor deployed on the ground. It generates guidance reference points based on position estimates and estimation error ranges, and further generates search guidance commands for the intercepting UAV, achieving efficient conversion of ground detection information into airborne interception commands. By optimizing the search formation of the intercepting UAV through estimation error ranges, and generating control commands for the intercepting UAV from the search guidance commands, it drives the intercepting UAV to search for the target UAV, achieving lightweight air-ground coordination. Overall, by constructing an air-ground coordinated architecture combining ground-based visual servo guidance and airborne autonomous search, it successfully solves the dual dilemmas of difficult detection and tracking and low cost-effectiveness in guidance in traditional anti-UAV methods. It significantly improves the response speed and interception success rate of the interception system against highly maneuverable targets in complex electromagnetic and low-altitude environments, while maintaining controllable overall system cost, possessing extremely high practical value and economic efficiency. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance provided in this application.

[0020] Figure 2 This is an architecture diagram of the terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance provided in this application.

[0021] Figure 3 This is a schematic diagram of the hypothetical initial moment provided in this application.

[0022] Figure 4 This is one of the schematic diagrams of the group trajectory at the final state provided in this application.

[0023] Figure 5 This is the second schematic diagram of the group trajectory at the final state provided in this application.

[0024] Figure 6 This is the third schematic diagram of the group trajectory at the final state provided in this application.

[0025] Figure 7 This is the fourth schematic diagram of the group trajectory at the final state provided in this application.

[0026] Figure 8 This is a schematic diagram of the terminal guidance device for intercepting unmanned aerial vehicles based on ground-based visual servo guidance provided in this application.

[0027] Figure 9 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps, units, or modules is not necessarily limited to those explicitly listed, but may include other steps, units, or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0030] Currently, terminal interception guidance technologies for small unmanned aerial vehicle (UAV) targets mainly include radar guidance, pure airborne infrared / optical imaging guidance, and inertial navigation guidance. Among these technologies, radar guidance is severely affected by multipath effects and ground clutter in low-altitude environments, making stable tracking of small targets difficult. Pure airborne infrared or optical guidance is limited by the size and payload constraints of the interceptor itself; its optical system has a short focal length and limited field of view, resulting in a short interception range and insufficient correction time and energy for interception, making it highly susceptible to misses when the target is highly maneuvering. Furthermore, although some technologies have attempted to utilize ground-based optoelectronic equipment for assistance, existing collaborative methods are mostly limited to simple target indication, lacking guidance algorithms that deeply integrate high-precision ground-based visual measurement information with the interceptor's onboard maneuver control capabilities. This generally results in low information fusion utilization, poor guidance law adaptability, and difficulty in handling highly dynamic terminal engagement scenarios. These technologies have failed to effectively solve the core control challenge of how to efficiently convert high-precision ground-based detection information into terminal maneuver commands for the interceptor, leading to interception accuracy and success rates that fail to meet practical combat requirements.

[0031] In view of the problems existing in related technologies, this application proposes a terminal guidance method for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance. This application focuses not on the development of specific sensor hardware and its performance assurance, but rather on the innovation of guidance and control algorithms and system architecture, proposing an air-ground cooperative guidance architecture that combines ground-based visual servo guidance with airborne autonomous decision-making. This method leverages the high-precision advantage of ground-based visual servo systems in target tracking through algorithms, transforming it into remote and precise target information for the interceptor (intercepting UAV), and designs an airborne autonomous decision-making mechanism to achieve real-time response to the target's maneuvering status.

[0032] Figure 1 This is a flowchart illustrating the terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance provided in this application, as shown below. Figure 1 As shown, the method includes steps 110, 120 and 130.

[0033] Step 110: Based on at least two ground-deployed visual sensors, simultaneously acquire observation data of at least one target UAV, calculate the position estimate of each target UAV and the estimation error range characterizing the uncertainty of the position estimate.

[0034] Specifically, the execution entity of the ground-based visual servo-guided terminal guidance method for intercepting unmanned aerial vehicles (UAVs) provided in this application embodiment is the UAV terminal guidance system. This system can be implemented in software, such as a ground-based visual servo-guided terminal guidance program for intercepting UAVs; or it can be a device that executes the ground-based visual servo-guided terminal guidance method for intercepting UAVs, such as a terminal, computer, or server.

[0035] The target drone is the drone that needs to be intercepted, such as a low-altitude, slow-moving, small target, or highly maneuverable drone. The interceptor drone is the drone that intercepts the target drone. In this embodiment, the number of target drones can be one or more, and the number of interceptor drones can also be one or more. This embodiment does not specifically limit the number of target drones and interceptor drones.

[0036] The visual sensor can be any photoelectric detection device capable of acquiring information such as the orientation and angle of the target drone. To reduce system cost and deployment complexity, visible light cameras or industrial cameras can be used. To adapt to nighttime or inclement weather conditions, infrared thermal imagers can also be used. At least two visual sensors are deployed. In the following embodiments, the method provided in this application will be described using a visual sensor as an example of a camera.

[0037] The observation data mainly refers to the information acquired from various visual sensors that characterizes the orientation of the target UAV. For each target UAV appearing in the sensor's field of view, the observation data includes at least its azimuth and pitch angles in the sensor's coordinate system. This angular information can be extracted from the acquired images using image processing algorithms and calculated in conjunction with the sensor's intrinsic and extrinsic parameters.

[0038] To ensure that the observation data acquired by different visual sensors correspond to the state of the target UAV at the same time, synchronous data collection is required.

[0039] The calculation refers to determining the coordinates of the target UAV in three-dimensional space using synchronous observation data from at least two visual sensors. Its basic principle is triangulation. Given the three-dimensional spatial positions of the two visual sensors and the direction vectors (determined by azimuth and pitch angles) of their observations of the same target, these two direction vectors should intersect at the target's location in space. In practice, due to measurement errors, the two lines may be skewed. In this case, the midpoint of the pair of closest points along the line of sight can be calculated as the optimal estimate of the target's position.

[0040] When there are multiple target drones in the scene, the solution process also includes a matching or association step, that is, it is necessary to correctly determine the first target drone observed by visual sensor 1. The target drone and visual sensor 2 observed the first To determine whether each target drone is the same physical entity, the correct line of sight is paired for triangulation.

[0041] The position estimate is the direct output of the solution step, that is, the three-dimensional coordinates of the target UAV in the preset global coordinate system.

[0042] Due to various factors such as sensor measurement errors, calibration errors, and synchronization errors, the calculated position estimate will always deviate from its true value. This application introduces an estimation error range to quantify and characterize this uncertainty. This error range can be modeled as a spherical space with the position estimate as its center and a specific value as its radius—an uncertainty sphere. The radius of this sphere is related to several factors, such as the target distance, sensor angular resolution, and the angle between the lines of sight of the two sensors (i.e., the intersection angle; a small angle will significantly increase the error). This error range defines a specific, finite physical space for subsequent search missions to intercept unmanned aerial vehicles (UAVs).

[0043] Step 120: Based on the position estimate and the estimation error range, generate a guidance reference point to guide at least one interceptor drone to search for at least one target drone, and generate search guidance instructions for each interceptor drone based on the guidance reference point.

[0044] Specifically, the guidance reference point is one or more three-dimensional spatial coordinate points generated by the ground control station, which serve to indicate the target location during the interceptor UAV's flight and search. The strategy for generating the guidance reference point is closely related to the position estimate and estimation error range obtained in the first step.

[0045] In one specific embodiment, the estimated position of the target UAV can be directly used as a guidance reference point. In another specific embodiment, considering the existence of estimation error, one or more guidance reference points can be generated within the space defined by the estimation error range.

[0046] Search guidance instructions are specific commands sent from a ground control station to one or more interceptor drones. The core of these instructions is the position coordinates of the assigned guidance reference point. In one specific embodiment, to achieve lightweight and efficient air-to-ground information exchange, these instructions can be designed to be very concise, for example, containing only key information such as the target drone's estimated position, estimation error range, and guidance reference point. This avoids transmitting redundant image or process data, thereby saving communication bandwidth and reducing communication latency.

[0047] Step 130: Optimize the search formation of the intercepting drone based on the estimated error range, generate control commands for the intercepting drone based on the search guidance commands, and drive the intercepting drone to search for the target drone.

[0048] Specifically, when multiple (at least two) interceptor drones are collaboratively performing a mission, how to organize them to efficiently cover the search area defined by the estimated error range becomes a formation optimization problem. Optimization objectives typically include: maximizing search coverage, minimizing total search time, avoiding collisions between drones, and reducing overlapping areas of search paths. This optimization process can be implemented using intelligent optimization algorithms. The algorithm's inputs are the geometric description of the search area (i.e., the estimated error range) and the current state (position, speed, etc.) of each drone in the swarm. The output is the optimal search path or formation position planned for each drone. Through formation optimization, it can be ensured that the drone swarm conducts a comprehensive search of the area where the target may exist in the most efficient and safest manner.

[0049] Control commands refer to the low-level instructions generated by the interceptor drone's internal flight control system to drive actuators such as motors or servos. After receiving a search guidance command containing a guidance reference point, the interceptor drone's flight control system calculates the desired flight speed, acceleration, and attitude angles based on the deviation between that reference point and its current position using control algorithms. These desired states are then interpreted into specific low-level control commands such as motor speeds or control surface deflection angles.

[0050] Control commands are sent to the drone's power system and actuators, causing it to fly toward the guide reference point along the planned trajectory. During flight, its onboard sensors (such as cameras) continuously detect the surrounding airspace. Once it captures the target drone within its field of view, it can switch to autonomous tracking, countermeasures, or attacks.

[0051] The terminal guidance method for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance provided in this application achieves precise tracking of the target UAV by collecting observation data from a low-cost visual sensor deployed on the ground. It generates guidance reference points based on position estimates and estimation error ranges, and further generates search guidance commands for the intercepting UAV, achieving efficient conversion of ground detection information into airborne interception commands. By optimizing the search formation of the intercepting UAV through estimation error ranges, and generating control commands for the intercepting UAV through search guidance commands, it drives the intercepting UAV to search for the target UAV, achieving lightweight air-ground coordination. Overall, by constructing an air-ground coordinated architecture combining ground-based visual servo guidance and airborne autonomous search, it successfully solves the dual dilemmas of difficult detection and tracking and low cost-effectiveness in guidance in traditional anti-UAV methods. It significantly improves the response speed and interception success rate of the interception system against highly maneuverable targets in complex electromagnetic and low-altitude environments, while maintaining controllable overall system cost, demonstrating high practical value and economic efficiency.

[0052] It should be noted that each implementation method of this application can be freely combined, rearranged, or executed individually, and does not need to rely on or depend on a fixed execution order.

[0053] In some embodiments, observation data of at least one target UAV is simultaneously acquired by at least two ground-deployed visual sensors, and the position estimates of each target UAV and the estimation error range characterizing the uncertainty of the position estimates are calculated, including: Construct a matching cost matrix by integrating at least two of the following: geometric topology sorting constraints, back ray hard constraints, depth consistency constraints, and spatial distance threshold constraints; The line-of-sight relationships of a target UAV observed by at least two visual sensors are matched based on the matching cost matrix to determine the correspondence between the lines of sight; the lines of sight are generated based on the observation data. Based on the line-of-sight correspondence, calculate the point pair with the closest spatial distance on each pair of matched lines of sight, and use the midpoint of the line connecting the point pairs as the estimated position of the target UAV. The estimated error range is determined based on the gap error of the matched line of sight and the physical measurement error of the vision sensor.

[0054] Specifically, this application provides a ground-based dual-vision lightweight target position estimation and tracking method, mainly used to solve the problem of rapid position estimation and tracking continuity of target UAVs in complex low-altitude environments.

[0055] Step 1: Construct a vectorized estimation model of the orientation vectors of multiple target UAVs and perform data preprocessing.

[0056] Based on the mapping relationship between spherical coordinates (the coordinates of the target UAV) and Cartesian coordinates (the coordinates of the camera), the direction vectors of multiple target UAVs are estimated in batches, which is expressed by the formula: .

[0057] in, The pitch angle is measured in radians. It is the azimuth angle (in radians). The unit direction vector for observing the UAV.

[0058] Lightweight optimizations to ensure data validity can be implemented by designing observation data verification criteria to eliminate significant noise and outliers. L2 norm normalization is performed on the direction vector; when the vector magnitude is less than a given value, normalization is directly truncated, eliminating complex iterative calibration processes and ensuring efficient convergence of subsequent calculations.

[0059] Step 2: Perform geometric vectorization calculations on the observation data from the dual cameras.

[0060] Extracting camera coordinates from two cameras (Camera 1) (Camera 2) Calculate the baseline vector of the line connecting the two cameras. Based on the determined direction vectors, the geometric parameter matrices of all target line-of-sight pairs (line-of-sight pairs from two different cameras that the algorithm determines are pointing to the same target) are calculated using vectorization. Specifically, this includes: Calculate the dot product matrix of the direction vectors of the target UAV from the two cameras. ( , For camera 1 to observe the first The direction vector of the target UAV For camera 2 to observe the first The direction vector of the target UAV For camera 1 to observe the first The target drone and camera observed the second instance. (Line of sight angle of the target drone).

[0061] Calculate the dot product of the direction vector and the baseline vector. , , Let be the direction vector of camera 1. Let be the direction vector of camera 2. After expanding it into a matrix, the ray parameter matrix is ​​calculated based on geometric relationships, using the following formula: .

[0062] In the formula, For camera 1 to observe the first The dot product vector of the direction vector and the baseline vector of the target UAV. For camera 2 to observe the first The dot product vector of the direction vector and the baseline vector of the target UAV. Here is the ray parameter matrix for camera 1. Let be the ray parameter matrix of camera 2.

[0063] The elements in the ray parameter matrix represent the scalar distance (or depth) that needs to be traveled from the camera position along their respective observation lines (rays).

[0064] For parallel scenarios where the denominator approaches 0, the corresponding , The threshold is truncated to 0.0 to avoid computational errors; the closest point on two rays (the observation lines of the two cameras) is calculated based on the ray parameter matrix. and ; , And to quantize the ray spacing matrix for all line-of-sight pairs. .

[0065] Step 3: Integrate multiple constraints to construct a matching cost matrix, match the lines of sight of the two cameras observing the target UAV, and determine the line-of-sight correspondence.

[0066] Geometric topological ordering constraints refer to the high probability conservation of the relative left-right (or up-down) order of multiple target UAVs on different camera imaging planes under non-extreme deployment conditions. In practice, all target UAVs observed by two sensors can be ordered according to their azimuth angles to generate a geometric rank vector. , Construct the rank difference matrix , For camera 1 to observe the first The geometric rank vector of a target UAV For camera 2 to observe the first The smaller the rank difference between the geometric rank vectors of the target drones, the closer their relative order is from their respective perspectives, and the higher the probability of matching. This rank difference can be added as a cost term to the matching cost matrix to achieve matching guidance in ID-free scenarios.

[0067] Back-ray hard constraints are used to exclude physically impossible matching combinations. For a pair of candidate lines of sight, their intersection point must be in front of both cameras. If, during geometric calculations, a line of sight is found to have a negative calculated parameter, it means the intersection point is behind that sensor, which is physically impossible. Therefore, the cost of such matching combinations can be directly set to a maximum value (e.g., infinity), thus hard-excluding them from the matching process.

[0068] The depth consistency constraint leverages the characteristic that distances to the same target from different viewpoints should be consistent. For a pair of candidate matching lines of sight, the distances from the nearest points on each line of sight to their respective sensors can be calculated. Theoretically, these two depth values ​​should be similar. Their ratio or difference can be calculated to construct a depth penalty term.

[0069] Spatial distance threshold constraints are the most intuitive geometric constraints. Two lines of sight from different cameras, if corresponding to the same target, should theoretically have a very small distance in 3D space, ideally zero. In reality, due to errors, this distance is a non-zero value. A spatial distance threshold can be preset; if the calculated distance between the lines of sight exceeds this threshold, they are considered impossible to match, and the matching cost is directly set to infinity.

[0070] In a specific embodiment, a matching cost matrix can be constructed by fusing a depth penalty term (depth consistency constraint) and a rank difference penalty term (geometric topological sorting constraint) based on the ray spacing matrix. The formula is: .

[0071] in, The ray spacing matrix, This is a deep penalty item. For depth ratio, For geometric rank difference, and The corresponding weights can be determined based on the actual situation; a spatial distance threshold is set when... The cost is set to infinity when the threshold is exceeded.

[0072] A greedy strategy can be used to iterate through the sorted matching costs and quickly select the globally optimal matching pair. Each matching pair represents a set of line-of-sight correspondences.

[0073] Step 4: Based on the line-of-sight correspondence, calculate the point pair with the closest spatial distance on each pair of matched lines of sight, and take the midpoint of the line connecting the point pairs as the estimated position of the target UAV; and determine the estimation error range based on the gap error of the matched lines of sight and the physical measurement error of the visual sensor.

[0074] To address the pain point of unguided search areas for drones, this application proposes an uncertainty sphere model. This model integrates physical measurement errors from visual sensors with ray gap errors (gap errors in the matching line of sight) to define the estimation error range of the target position. The center of the sphere is the estimated position of the target drone, and the radius of the sphere is: .

[0075] in, For the error in the gap between rays, For physical measurement error ( This is the estimated depth of the target drone. To account for the measurement error of the two camera angles, (The angle between the observation directions of the two cameras).

[0076] The center of the uncertainty sphere is the midpoint of the line connecting the closest point pair in the matching line of sight, and the radius of the sphere is the estimated error range, which defines the search area for intercepting drones.

[0077] The terminal guidance method for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance provided in this application greatly improves the accuracy and robustness of target matching in multi-target, unmarked scenarios through dual-vision observation modeling, geometric parameter vectorization calculation, multi-constraint rapid matching, and uncertainty range characterization. It effectively avoids catastrophic positioning errors caused by mismatches, achieves efficient estimation and stable tracking of the target UAV's position, and provides a reliable basis for searching and guiding intercepting UAVs.

[0078] In some embodiments, the method further includes: In the event that matching fails due to occlusion of any target UAV, a primary visual sensor is determined among at least two visual sensors, and a virtual guide point for any target UAV is generated along the observation direction of the primary visual sensor for any target UAV.

[0079] Specifically, for a target UAV where matching fails due to occlusion, a primary vision sensor can be identified. The primary vision sensor is the one that can still observe the target UAV even with unilateral occlusion. In a binocular vision system, a sensor can be pre-designated as the primary vision sensor based on deployment location, observation angle, or historical data quality. More generally, when matching fails, the system checks which sensor still holds valid observation data from the target UAV; this sensor is dynamically determined as the primary vision sensor for the current situation.

[0080] A virtual guide point is generated in front of the main camera (main vision sensor). This can be expressed as a formula: .

[0081] in, The position of the main camera (with camera 1 as the main camera). For camera 1 to observe the first The direction vector of the target UAV (the target UAV that failed to match), This is for virtual guidance distance.

[0082] The virtual point is set as the center of an uncertain sphere with a radius of -1.0, and marked as a monocular observation mode. Combined with secondary angle calibration, tracking continuity is ensured, and the interception drone is guided to search the virtual point area.

[0083] The terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance provided in this application greatly enhances the tracking robustness of the entire guidance system by generating a virtual guidance point when the target unmanned aerial vehicle is obscured.

[0084] In some embodiments, based on position estimates and estimation error ranges, a guidance reference point is generated to guide at least one interceptor drone to search for at least one target drone, including: An uncertainty sphere is generated with the position estimate as the center and the estimation error range as the radius. When the minimum distance between the intercepting drone and the uncertainty ball is greater than the preset mode switching distance threshold, the close-range search mode is adopted, and the center of the uncertainty ball is used as the guiding reference point. When the minimum distance between the intercepting drone and the uncertainty sphere is less than or equal to the preset mode switching distance threshold, a close-range detection mode is adopted to generate multiple uniformly distributed sub-reference points inside the uncertainty sphere, and these sub-reference points are used as guiding reference points.

[0085] Specifically, the embodiments of this application construct a dual-mode guidance architecture that includes a close-range search mode and a close-range detection mode.

[0086] An uncertainty sphere is a geometric model of the spatial region where a target drone may exist. As described in the previous embodiments, after calculating the estimated position of the target drone and quantifying the range of uncertainty estimation errors using binocular vision, this embodiment visualizes it. The calculated estimated position of the target drone is used as the center of the uncertainty sphere. The range of estimation errors (e.g., the radius value determined by combining gap error and physical measurement error in the previous embodiments) is used as the radius of the uncertainty sphere. The uncertainty sphere generated in this way clearly delineates a target area in three-dimensional space that the intercepting drone needs to search for and detect.

[0087] When the minimum distance between the intercepting drone and the uncertainty sphere exceeds a preset mode switching distance threshold, a close-range search mode is adopted, using the center of the uncertainty sphere as a guiding reference point. Specifically, the center of the uncertainty sphere is used as the guiding reference point, and the coverage radius is determined to be the radius of the uncertainty sphere to define the search area. The intercepting drone is then guided to conduct a diffusion search centered on the guiding reference point and within the coverage radius, without the need for additional path planning.

[0088] When the minimum distance between the intercepting drone and the uncertainty sphere is less than or equal to a preset mode switching distance threshold, a close-range detection mode is adopted. Multiple sub-reference points are generated evenly distributed within the uncertainty sphere, and these sub-reference points serve as guiding reference points. Specifically, multiple dynamic sub-reference points are generated evenly distributed within the uncertainty sphere to guide the intercepting drone to perform a refined scan in sequence according to the sub-reference points.

[0089] During execution, minimal security constraints can be imposed on intercepting drones to prevent them from maneuvering violently and reduce the complexity of command execution.

[0090] The terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance provided in this application abandons the "one-size-fits-all" guidance logic from beginning to end. It can dynamically adjust the guidance strategy according to the real-time situation on the battlefield, so that the guidance command is highly matched with the phased requirements of the interception mission, avoiding resource waste and realizing the whole process optimization of highly maneuverable targets from rapid response to precise encirclement.

[0091] In some embodiments, search guidance instructions for each intercepting drone are generated based on a guidance reference point, including: Construct a task allocation cost matrix that includes at least one of path cost, angle cost, and load penalty cost; the path cost is determined based on the distance between the current position of the intercepting UAV and the guidance reference point; the angle cost is determined based on the deviation between the current observation angle of the intercepting UAV and the observation angle corresponding to the guidance reference point; the load penalty cost is determined based on the number of intercepting UAVs allocated in the search area defined by the estimation error range. Based on the task allocation cost matrix, guidance reference points are assigned to each intercepting drone, and the location information of the guidance reference points is used as the search guidance instructions for each intercepting drone.

[0092] Specifically, this application provides a method for allocating unmanned aerial vehicle (UAV) tasks, which enables lightweight air-to-ground information interaction and solves the technical problem of inefficient conversion of guidance commands.

[0093] Step 1: Construct a task allocation cost matrix.

[0094] Path cost reflects the physical flight cost of an interceptor drone performing a mission. Path cost is determined based on the Euclidean distance between the interceptor drone's current position and the guidance reference point to be assigned. A greater distance means more time and energy are required, thus resulting in a higher path cost. In some embodiments, this distance can also be weighted; for example, considering the complexity of low-altitude flight, the path cost can be multiplied by a coefficient greater than 1 (such as 1.1) to roughly account for the effects of terrain or obstacles.

[0095] The angle cost reflects the maneuvering costs and sensor pointing adjustments required when intercepting a drone during a mission. It is determined based on the deviation between the interceptor drone's current observation angle (i.e., the pointing of its onboard sensors) and the target observation angle required to point from its current position to the assigned guidance reference point. This deviation can be the absolute value of the combined deviation in azimuth and pitch angles. A larger deviation means the interceptor drone needs to make more drastic attitude adjustments, resulting in higher maneuvering costs and therefore a higher angle cost. Introducing this cost helps generate smoother, more energy-efficient flight paths. It can be expressed by the formula: .

[0096] in, For the first The intercepting drone against the first The angular cost of a target drone For the first The intercepting drone and the first Minimum azimuth deviation between target drones For the first The intercepting drone and the first The absolute value of the pitch angle deviation between the target drones For a fixed normalization factor (values) (To avoid multiple rounds of calibration).

[0097] The load penalty cost is used to achieve task load balancing among interceptor drone clusters. When multiple guidance reference points belong to the same target drone's search area (i.e., the same uncertainty sphere), the load penalty cost is determined based on the number of interceptor drones already assigned to that search area. The load penalty value is dynamically allocated based on the ratio of the number of interceptor drones to the number of target drone search areas (e.g., the load penalty cost doubles when the number of drones assigned to a search area is greater than or equal to 3), eliminating the need for complex load balancing iterations.

[0098] Normalize the path cost and angle cost, and fuse them with fixed weights to obtain the task allocation cost matrix. : .

[0099] in, For the first The first intercept drone search Task allocation cost per target drone For the first The first intercept drone search The normalized value of the path cost of a target UAV; For the first The first intercept drone search The normalized value of the angle cost of a target drone; For the first The first intercept drone search The cost of load balancing for a single target drone; , and These are the calculation weights for path cost, angle cost, and load balancing cost, respectively. =1. The calculation weights can be set according to the priority of each cost (e.g., path cost has the highest priority, angle cost has a medium priority, and load balancing cost has the lowest priority), omitting multiple rounds of weight optimization iterations and greatly improving modeling efficiency.

[0100] Step 2: Perform adaptive task allocation for multiple target drones and multiple interceptor drones.

[0101] Initialize the assignment flag for each interceptor drone (e.g., 0 for unassigned, 1 for assigned). Assign an interceptor drone to each of the multiple target drones one by one. Path conflicts can be determined by a simplified rule that checks if the distance between the midpoints of line segments is greater than a safety threshold (omitting the complex geometric calculations of high-precision collision warning). Select the interceptor drone with the lowest task assignment cost to complete the assignment, and update the assignment flag and load information.

[0102] For unassigned interceptor drones, they can be quickly assigned according to the principle of proximity. Specifically: Prioritize screening target areas (search areas for target drones) where the load has not reached the upper limit (less than the rated number of drones), and allocate them in order of cost from smallest to largest; If no target area is available, the unassigned interceptor drones will be directly assigned to the assigned area to perform collaborative supplementary scanning. Only the load penalty value will be updated, and the task allocation cost matrix will not be recalculated, ensuring the rapid generation of allocation instructions.

[0103] Step 3: Achieve lightweight air-to-ground information interaction and closed-loop guidance and correction.

[0104] Design a lightweight air-to-ground data interaction mode, limit the content to be transmitted, and avoid redundant data occupying the link.

[0105] The ground control station transmits only three types of core data to the intercepting drone: target position estimate, radius of the uncertainty sphere, and guidance reference point (including pattern markers: 0 = area search / 1 = close-range detection).

[0106] Intercepting drones only need to report two types of status data to the ground control station: the drone's current position and mission completion status (0 = not completed / 1 = completed), without needing to report redundant information such as attitude and speed.

[0107] Lightweight closed-loop correction refers to performing simplified correction based on synchronization information, as detailed below: If the interceptor drone fails to complete its feedback and its position deviates from the reference point by more than the set distance (e.g., 20 meters): only the drone's guidance reference point is corrected, and the global allocation scheme is not recalculated; if the interceptor drone reports that it has detected the target drone: immediately send a command to all interceptor drones to focus on the area, stop searching other areas, and quickly form a coordinated detection posture.

[0108] By performing the above operations, complex multi-round data verification and iterative optimization are omitted. With rapid correction and meeting task requirements as the core, an efficient closed loop of estimation-allocation-guidance-simplification correction is formed, improving the timeliness of information transformation.

[0109] The terminal guidance method for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance provided in this application can ensure that all intercepting UAVs perform search tasks in the most coordinated manner with the lowest overall cost, thereby significantly improving the collaborative combat effectiveness and system resource utilization in complex multi-target scenarios.

[0110] In some embodiments, optimizing the search formation for intercepting drones based on the estimated error range includes: A fitness function is constructed based on at least one of the following: the total distance between the intercepting UAV and the guidance reference point, the penalty for the number of times the intercepting UAV's search formation overlaps, the penalty for the coverage uniformity of the search area defined by the estimation error range, and the penalty for the smoothness of the intercepting UAV's maneuvering. Based on the fitness function, a particle swarm genetic hybrid intelligent optimization algorithm is used to optimize the search formation for intercepting UAVs.

[0111] Specifically, the fitness function is the core of intelligent optimization algorithms; it is used to quantitatively evaluate the quality of a candidate solution (in this case, a specific search pattern). A good search pattern should have the shortest path, no overlapping coverage, uniform distribution, and smooth maneuverability.

[0112] The total distance between the interceptor drone and the guidance reference point is used to measure the overall path cost of all interceptor drones performing search missions.

[0113] The penalty for overlapping search formations of intercepting drones is used to penalize the overlap of search ranges between different intercepting drones in the formation.

[0114] The coverage uniformity penalty of the search area defined by the estimation error range is used to evaluate whether the array covers the entire search area (i.e., the uncertainty sphere) uniformly.

[0115] The maneuver smoothness penalty for intercepting drones is used to ensure the smoothness of formation changes and prevent drones from making violent, energy-intensive maneuvers.

[0116] The fitness function can be constructed based on the above items. This can be expressed as a formula: .

[0117] in, The total distance between the intercepting drone and the guidance reference point; Penalty for overlapping search formations used to intercept drones; Penalty for uniformity of coverage; Penalty for smooth maneuverability; , , and The calculation weights for the above items are set with computational efficiency as the priority.

[0118] The search formation for intercepting drones can be optimized using a hybrid intelligent optimization algorithm of particle swarm genetics (PSO-GA).

[0119] First, initialize the particle swarm. The position of each particle is an array of search angles for the intercepting drone (ideally, the angles are evenly distributed across...). (with slight perturbations); combined with historical optimization perspectives for hot start-up, improving convergence speed and omitting dimensional expansion of complex particles; Secondly, hybrid iterative rapid optimization, based on inertia weights. Cognitive factors Social factors Update particle velocity and position (velocity range limited to) ); Select High-quality particles serve as parents, generating offspring through single-point crossover or uniparental inheritance. These offspring undergo dynamic mutation (with the mutation amplitude decreasing in later iterations) to replace low-quality particles. If no optimal solution is found after 20 iterations, then... The particles are subjected to random perturbations to escape local optima, ensuring efficient optimization of the search formation; After optimization, reference points for intercepting drones are generated based on the optimal angle, and historical optimal angles and historical reference points are updated to ensure that the intercepting drone swarm forms a highly efficient search formation with full coverage and no obvious overlap.

[0120] The terminal guidance method for intercepting UAVs based on ground-based visual servo guidance provided in this application constructs a multi-dimensional fitness function that integrates path, overlap, uniformity, and smoothness, and uses an efficient particle swarm genetic hybrid algorithm to solve it. This method can quickly calculate the globally optimal search formation for UAV swarms in complex and ever-changing search tasks.

[0121] In some embodiments, control commands for intercepting drones are generated based on search guidance commands, including: The search guidance command is sent to the controller of the intercepting drone. The controller determines the position information of the guidance reference point based on the search guidance command, determines the distance deviation based on the position information of the current position of the intercepting drone and the guidance reference point, and generates control commands based on the distance deviation. The controller is an enhanced robust proportional-integral-derivative controller.

[0122] Specifically, an enhanced robust proportional-integral-derivative (PID) controller can be configured for each intercepting UAV, and constraints on the base gain, maximum speed, maximum acceleration, and maximum jerk can be set (jerk = acceleration / 0.2). A sliding window of size 5 is initialized for target motion trend estimation, omitting complex target maneuver modeling.

[0123] Upon receiving the search guidance command, the controller parses the position information of the guidance reference point, determines the distance deviation based on the current position of the intercepting drone and the position information of the guidance reference point, and generates control commands.

[0124] The controller can estimate the motion trend of the target UAV based on a sliding window and second-order polynomial fitting, and smooth the data using a low-pass filter; it adjusts the PID gain based on the distance deviation between the intercepted UAV and the guidance reference point (the gain is increased when the distance deviation is greater than 50 meters). Gain decreases when the distance is less than 10 meters. ); Introduce adaptive dead zone (expected distance) (Zero deviation within the range) suppresses micro-oscillations; physical constraints are applied in the order of jerk-acceleration-velocity to output a stable guidance control speed, ensuring that the interceptor UAV maneuvers efficiently to the search area.

[0125] The terminal guidance method for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance provided in this application adopts an enhanced robust proportional-integral-derivative (PID) controller, which enables the intercepting UAV to execute guidance commands quickly, stably, accurately, and safely. It significantly improves the tracking performance of dynamic guidance reference points and is a key link in ensuring that the entire guidance method can ultimately achieve high-precision interception.

[0126] The method in this application relies on a ground-based dual-vision lightweight platform and conventional commercial drones. It does not require additional high-precision positioning and communication hardware. It improves search guidance efficiency through pure algorithm optimization, significantly reducing system deployment costs.

[0127] The drone's position can be updated based on the control speed, and the status of the intercepted drone and the estimated position of the target drone can be recorded at each time step. The formation optimization-guided output-position update process is repeated until the drone completes full coverage search, encirclement and interception of the search area, and the final search trajectory and mission result judgment information are output.

[0128] Figure 2 This is an architecture diagram of the ground-based visual servo-guided terminal guidance method for intercepting unmanned aerial vehicles provided in this application, as shown below. Figure 2 As shown, the method includes: (1) State estimation of target UAVs is performed by ground dual vision, including multi-target observation modeling, coplanar adaptive determination, multi-constraint fast matching, uncertainty sphere representation and occlusion virtual guidance mechanism, which solves the technical problems of continuous detection and fast estimation of target UAVs; (2) By switching between regional search and close-range detection modes, lightweight cost modeling, multi-target adaptive allocation, simplified air-ground information interaction and dynamic closed-loop correction mechanism, the technical problems of information interaction efficiency and real-time command conversion are solved; (3) By using intelligent hybrid optimization of search formation, enhanced robust PID control, online estimation of target UAV, physical constraint level control and dynamic evolution of search formation, the technical problems of adaptability and interception accuracy of highly maneuverable targets are solved.

[0129] The method provided in this application will be explained below using a low-altitude multi-target interception scenario as an example.

[0130] In this scenario, relying on a ground-based dual-vision platform (including camera 1 and camera 2), six intercepting drones (red team drones) and four harassing drones (blue team targets) are deployed. Combined with pre-defined parameter configurations and algorithm flows, high-precision terminal-guided interception of dynamic targets is achieved. The algorithm flow can be implemented using Python code.

[0131] I. Set up the implementation scenario and configure the core parameters.

[0132] 1. Define the implementation scenario.

[0133] Implementation scenario: Open low-altitude area.

[0134] Interception targets: 4 intrusive drones, whose initial positions are defined by the initial deployment position parameters and move along random variable acceleration maneuver trajectories.

[0135] Operational situation: The intruding drones start from the initial area and maneuver in random directions. The ground dual-vision platform observes the target in real time and guides 6 intercepting drones (forming a drone swarm) to work together to approach, surround and intercept 4 intruding drones.

[0136] Task cycle: Iterate 300 times according to the system control cycle of 0.1s (seconds). During the iteration, the core logic execution and parameter update are completed through the core function of target allocation and guidance control by multiple interceptors.

[0137] 2. Configure core parameters.

[0138] (1) General control parameters (global initialization logic): The minimum distance threshold for coordinated encirclement is 10m, and the maximum flight speed of the intercepting UAV is 20m / s; the dual-mode guidance mode switching distance threshold is 50m (default close-range detection mode); the spacing between refined guidance sub-reference points is 1m, and the algorithm iteration step size is 1; the system control cycle is 0.1s; PID controller parameters (proportional gain coefficient is 0.8, derivative gain coefficient is 0.5, integral gain coefficient is 0); the maximum allowable displacement deviation of the target maneuver (x / y / z axis) is [1.0, 1.0, 1.0]m (maximum displacement deviation within the control cycle); the speed constraint of the intercepting UAV (x / y / z axis) is [20.0, 20.0, 5.0]m / s, and the acceleration constraint of the intercepting UAV (x / y / z axis) is [10.0, 10.0, 1.5]m / s.2 (meters per second squared); the dual-vision target matching threshold is 10.0m; the number of iterations for single-round task allocation optimization is 1.

[0139] (2) Task allocation configuration parameters (task allocation parameter configuration class): path cost weight coefficient is 0.6, angle cost weight coefficient is 0.15, load balancing cost weight factor is 2.0; UAV cooperative collision avoidance safety distance is 5.0m (used to avoid path conflicts between UAVs); the maximum allowable deviation of load balancing is 1, the minimum target load threshold is 1; the post-task allocation optimization mechanism is enabled, and the number of post-optimization iterations is 10.

[0140] (3) Deployment parameters (position initialization logic): The initial deployment positions of the 4 intrusion drones are [[100.0,50.0,125.0],[120.0,52.0,125.0],[140.0,54.0,125.0],[160.0,56.0,125.0]]m; the initial deployment positions of the 6 interception drones are [[100.0,3.0,10.0],[ [110.0,3.0,10.0],[120.0,3.0,10.0],[130.0,3.0,10.0],[140.0,3.0,10.0],[150.0,3.0,10.0]]m; Ground dual-camera deployment positions: The main dual-camera deployment position is (0.0,0.0,0.0)m, and the auxiliary dual-camera deployment position is (100.0,0.0,0.0)m.

[0141] (4) Initial state parameters (state variable initialization logic): The initial pitch angle of all 6 interceptor UAVs is 0.2 rad (radians), and the initial azimuth angle is 0.2 rad (radians). rad; Historical mission allocation angle records and historical target-interceptor mapping relationship records are initially empty sets; Historical virtual guidance reference points are initialized as an array of 6 null objects (corresponding to 6 interceptor drones, with no historical reference point data initially).

[0142] II. Implementation process of dual-vision observation and target tracking module.

[0143] Step 1: Dual-vision observation modeling and data preprocessing.

[0144] (1) Observation data acquisition. Ground dual cameras simultaneously acquire observation angles (pitch angles) of four targets. Azimuth Output the angle array, ensuring the array dimensions are 1. (2 corresponds to dual cameras, 4 corresponds to the number of targets); the initial observation angles of the 6 interceptor drones are preset according to the preconditions: initial pitch angle Initial azimuth angle In subsequent iterations, the angle value is updated through a multi-interceptor collaborative target allocation function.

[0145] (2) Direction Vector Calculation. Based on the mapping relationship between spherical coordinates and Cartesian coordinates, the observation direction vector of each target is calculated through vectorization operations. The calculation formula is expressed as: .

[0146] Taking the initial angle of intercepting a drone as an example, substituting... , calculate: , , , The initial direction vector is obtained as follows: .

[0147] (3) Normalization and validity verification. The direction vector is normalized using the L2 norm. The normalization formula is: .

[0148] in, , and These represent the three-dimensional coordinates of the direction vector.

[0149] like Then directly order (To avoid a denominator of 0); In the embodiments of this application No truncation is required. The angle range is also verified simultaneously. , The initial angles are all compliant, and the direction vector after noise removal is passed to the subsequent stages.

[0150] Step 2: Vectorization calculation of dual-vision geometric parameters.

[0151] (1) Calculate the baseline vector based on the deployment location of the dual cameras. Substitute , ,have to Baseline length .

[0152] (2) Extracting the dual-camera orientation vectors , The following matrices are computed in vector form: Direction vector dot product matrix : .

[0153] in, Representing camera 1 Target and Camera 2 The angle of view of each target.

[0154] dot product vector , : .

[0155] Ray parameter matrix , : .

[0156] against Parallel scenes, making To avoid calculation failures.

[0157] based on , Calculate the nearest point on the ray , Solve for the spacing matrix: .

[0158] The initial calculations in the embodiments of this application All less than (Matching threshold), enter the candidate matching set.

[0159] Step 3: Multi-constraint target matching and location estimation.

[0160] 1. Utilize the azimuth sorting conservation property to generate a rank vector and guide unidentified (ID) matching.

[0161] On camera 1 side: right Sort ,generate ( ); For quantity; On both sides of the camera: right Sort ,generate ; Calculate the rank difference matrix: , Correct match .

[0162] 2. Multi-constraint filtering.

[0163] Reverse ray hard constraint: Preserved and The effective intersection points are set to have an infinite cost, while the ineffective intersection points are set to have an infinite cost. The safe distance from the target to the camera.

[0164] Spatial distance constraints: The candidate pairs are set to infinity; Depth Consistency Constraint: Calculate the depth ratio: .

[0165] in, ,set up , Generate penalty items.

[0166] .

[0167] 3. Cost matrix and greedy matching.

[0168] The final cost matrix formula is: .

[0169] The valid values ​​are calculated, sorted from smallest to largest, and a greedy strategy is used to select the optimal matching pair. In this embodiment, all candidate pairs satisfy the constraints, and finally 4 sets of valid matching pairs are obtained, completing the IDless target matching.

[0170] 4. Location and uncertainty estimation are included.

[0171] Target location: Take and The midpoint is taken as the center of the uncertain sphere. That is, the estimated location of the target; Uncertainty radius: This factor combines ray gap error and physical error, and is expressed by the following formula: .

[0172] in, (Angle error converted to radians) The initial embodiment of this application. All ranges are within 0.8-1.2m, providing a range reference for drone searches; The location estimation results are used as input to the cooperative allocation function and are iteratively updated at control cycles of 0.01s.

[0173] Step 4: Ensure target tracking is protected by occlusion.

[0174] If target occlusion occurs during the iteration, causing matching to fail, a virtual guide point is generated: .

[0175] in, ;Will Set as the center of the sphere, and set the radius as... Mark the monocular mode and update to the historical virtual guidance reference point (initially 6 null arrays, corresponding to 6 interceptors). Combine this with angle calibration to guide the drone search and ensure uninterrupted tracking.

[0176] III. Implementation of dual-mode guidance for air-ground collaboration and UAV mission allocation.

[0177] Based on preset task allocation parameters, the system completes the collaborative allocation between the UAV and the target area and generates guidance reference points. All calculation processes are consistent with the task allocation logic within the core function of multi-interceptor collaborative target allocation and guidance control. The specific implementation process is as follows: Step 1: Dual-mode guidance determination and reference point generation.

[0178] 1. Mode Switching Determination. The minimum distance between the UAV and the uncertainty sphere of the target position estimation is calculated and compared with a preset dual-mode guidance mode switching distance threshold of 50.0m. In the initial state of this embodiment, after the UAV reaches the distance threshold with the uncertainty sphere, it switches to close-range detection mode and outputs refined sub-reference points.

[0179] 2. Based on the target position estimate and a preset fine-grained guidance sub-reference point spacing of 10m, uniformly distributed sub-reference points are generated within the uncertainty sphere. Taking an initial target position (100.0, 50.0, 25.0)m as an example, multiple sub-reference points are generated around this position at 10m intervals to guide the UAV to focus on a local area for fine-grained scanning. The generated sub-reference points are updated to historical virtual guidance reference points.

[0180] Step 2: Construction of the lightweight cost matrix.

[0181] 1. Path Cost Calculation. Based on the preset initial deployment position of the interceptor drone and the coordinates of the sub-reference point, the path cost is calculated using the spatial distance formula (Euler's formula), which serves as the core element of the path cost matrix.

[0182] 2. Angle Cost Calculation. Calculate the deviation between the UAV's current observation angle and the corresponding observation angle of the sub-reference point, and substitute it into the formula (absolute value of azimuth deviation + absolute value of pitch deviation) × normalization factor. The angle cost is calculated. In the embodiments of this application, the initial observation angle is uniform, and the angle cost is within a small range.

[0183] 3. Load penalty cost calculation. Based on the ratio of the number of drones to the number of target areas, the load of each area is 0 in the initial state, and the penalty value is uniformly set to 1; if the number of drones allocated to a certain area is greater than or equal to 3, the penalty value is doubled.

[0184] 4. Global Cost Fusion. Based on the preset weight configuration (path cost weight coefficient is 0.6, angle cost weight coefficient is 0.15, and load balancing cost weight factor is 0.25), the cost components are normalized and then weighted and summed to obtain the global cost matrix, which provides the core basis for subsequent initial allocation.

[0185] Step 3: Implementation of UAV mission allocation.

[0186] 1. Initial Allocation. Initialize the allocation flag (0 indicates unassigned, 1 indicates assigned), and filter unassigned drones by target area. Calculate the distance from the drone to the midpoint of the line connecting the drone to the sub-reference point according to the code logic, and compare it with the 5.0m safe distance for drone cooperative collision avoidance to determine path conflicts. If the midpoint distance is greater than or equal to 5.0m, it is determined to be conflict-free. Select the drone with the lowest global cost and assign it to the corresponding area, update the allocation flag and load information, and generate the initial allocation result.

[0187] 2. Post-optimization adjustment: Enable the post-optimization mechanism according to the configuration in the code, and optimize the initial allocation result according to the preset number of 10 post-optimization iterations. If there is a situation where the regional load difference is greater than the maximum allowable deviation of load balancing, adjust the drone allocation scheme to ensure load balancing in each region (maximum load difference less than or equal to 1), and finally update to the optimal task allocation result.

[0188] Step 4: Air-to-ground information exchange and closed-loop correction.

[0189] 1. Lightweight information transmission. The ground transmits three types of core data to the UAV (target position estimate, radius of the uncertainty sphere, sub-reference point and mode marker); the UAV feeds back two types of status data to the ground (current position, mission completion status), avoiding redundant data occupying the communication link and ensuring real-time iteration.

[0190] 2. Closed-loop correction implementation. If a UAV reports that it has not completed its mission and its position deviates from the reference point by more than 20m, only the guidance reference point of that UAV is corrected (the historical virtual guidance reference point is updated), without recalculating the global allocation scheme; if a UAV reports that it has detected a target, it immediately sends a focusing command to all UAVs, stops searching other areas, and forms a cooperative detection posture.

[0191] IV. Encirclement Formation Optimization and Robust Guidance Control Implementation.

[0192] Strictly following the loop execution logic of controller initialization-velocity calculation-position update in the code, and based on preset PID parameters and motion constraints, a 3D robust PID controller group completes the precise guidance of six interceptor drones. All calculation processes are consistent with the control logic in the core function of multi-interceptor collaborative target allocation and guidance control. The specific implementation process is as follows: Step 1: Initialize the enhanced 3D robust PID controller.

[0193] Following the code initialization logic, enhanced 3D robust PID controllers are configured for each of the six interceptor drones, forming a 3D robust PID controller group. The initialization parameters strictly match the code configuration: the basic proportional gain coefficient is 0.8, the basic derivative gain coefficient is 0.5, and the integral gain coefficient is 0; the interceptor drone velocity constraints (x / y / z axes) are [20.0, 20.0, 5.0] m / s, and the interceptor drone acceleration constraints (x / y / z axes) are [10.0, 10.0, 1.5] m / s. 2 The maximum jerk is calculated by dividing the acceleration value by 0.2, i.e., [50.0, 50.0, 7.5] m / s². 3 (meters per cubic second); Set a sliding window of size 5 to store historical target location data to estimate the target's movement trend.

[0194] Step 2: Target motion trend estimation and dynamic gain adjustment.

[0195] 1. Motion trend estimation. Five consecutive frames of target position data are stored through a sliding window. A second-order polynomial fitting method is used to fit the position data to obtain the target velocity and acceleration. This result serves as the input to the controller, providing data support for the guidance velocity calculation.

[0196] 2. Dynamic Gain Adjustment. The 3D robust PID controller group dynamically adjusts the PID gain based on the distance between the UAV and the sub-reference point. According to the code logic, if the distance is greater than 50m, the proportional gain coefficient is adjusted to 1.2 times (0.96) of the base gain and the derivative gain coefficient is adjusted to 1.1 times (0.55) of the base gain to improve the response speed; if the distance is less than 10m, the proportional gain coefficient is adjusted to 0.8 times (0.64) of the base gain and the derivative gain coefficient is adjusted to 0.9 times (0.45) of the base gain to avoid maneuver overshoot.

[0197] Step 3: Robustly guided velocity calculation and physical constraint application.

[0198] 1. The 3D robust PID controller group sets the dead zone range at 90% of the desired distance. The desired distance = fine-tuned guide sub-reference point spacing + target radius. If the distance between the UAV and the sub-reference point exceeds the dead zone, the distance deviation is calculated as the PID input; if it is within the dead zone, the deviation is set to 0 to avoid frequent adjustments.

[0199] 2. The 3D robust PID controller group calculates the desired control speed based on the input deviation using the PID control formula. Taking a certain UAV as an example, the desired speed is calculated by PID as: proportional gain coefficient × deviation + derivative gain coefficient × deviation rate of change. This speed result is output as a component of the interceptor UAV's guidance speed set.

[0200] 3. The 3D robust PID controller group corrects the desired speed according to the motion constraints preset in the code. First, it limits the acceleration change according to the maximum jerk, and then limits the final acceleration output according to the acceleration constraint of the intercepting drone, so as to ensure that the speed output meets the drone's motion capability and finally obtain a safe and stable guidance speed.

[0201] Step 4: Update the drone's location and iterate and optimize its formation.

[0202] 1. Position Update Calculation. Based on the final output guidance speed and the system control cycle of 0.01s, the UAV displacement increment is calculated as guidance speed × system control cycle, and then superimposed on the current position to obtain the new position. Taking a UAV guidance speed of [4.0, 3.0, 0.0] m / s as an example, the displacement increment is [4.0 × 0.1, 3.0 × 0.1, 0.0 × 0.1] = [0.4, 0.3, 0.0] m. If the current position is the initial deployment position of the intercepting UAV [100.0, 3.0, 10.0] m, the updated new position is [100.4, 3.3, 10.0] m.

[0203] 2. After each iteration, the logic updates according to the code parameters, assigning the updated pitch angle and azimuth angle output by the multi-interceptor cooperative target allocation and guidance control core function to the current pitch angle and azimuth angle of the interceptor drone, respectively. The historical task allocation angle record is updated to the current task allocation result, the historical target-interceptor mapping relationship record is updated to the current target-interceptor mapping relationship, and the historical virtual guidance reference point is updated. At the same time, the updated drone position is copied and assigned as the initial position for the next iteration, repeating the "trend estimation-velocity calculation-position update" process to achieve dynamic optimization of the encirclement formation.

[0204] V. Iterative Operation and Implementation Instructions

[0205] For iterative execution, see: Figure 3 This is a schematic diagram of the hypothetical initial moment provided in this application (indicating a static scene that is not coplanar). Figure 4 This is one of the schematic diagrams of the group trajectory at the final state provided in this application (indicating that the static scene is not coplanar). Figure 5 This is the second schematic diagram of the group trajectory at the final state provided in this application (indicating that the dynamic scene is not coplanar). Figure 6 This is the third schematic diagram of the group trajectory at the final state provided in this application (representing coplanar static scenes). Figure 7This is the fourth schematic diagram of the group trajectory at the final state provided in this application (representing coplanar static scenes).

[0206] like Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, the implementation effect of the method of this application in a dynamic scene is clearly demonstrated, which adopts the target to move at a 45° angle upward (three-dimensional space) with uniform acceleration.

[0207] like Figure 7 As shown, to clearly demonstrate the implementation effect of the method of this application in a static scene with the target coplanar, the initial deployment positions of the four intruding drones are set as [[100.0,50.0,125.0],[120.0,50.0,125.0],[140.0,50.0,125.0],[160.0,50.0,125.0]]m, that is, the line connecting the target is parallel to the line connecting the two cameras, so that the guide line from the camera to the target is coplanar and displayed at a specific angle.

[0208] This application strictly follows the iterative execution logic of the code, iterating 1000 times according to a system control cycle of 0.01s. During the iteration process, the core function of multi-interceptor collaborative target allocation and guidance control completes target position estimation (target position update), task allocation optimization (current task allocation result update), guidance control (guidance speed calculation), and position update (new UAV position generation) in real time. Ultimately, it achieves the coordinated encirclement and capture of 6 interceptor UAVs and 4 intrusion UAVs. The UAV trajectory and target trajectory are recorded synchronously during the iteration process.

[0209] This application's embodiments fully reproduce the core logic of the Python code. From parameter initialization and iterative execution to result output, all implementation steps closely match the function calls, variable passing, and calculation processes in the code. The implementation process does not introduce any additional complex procedures, precisely ensuring the lightweight and efficient nature of the technical solution. The presented implementation details can be directly verified line by line against the code, making it fully adaptable to actual engineering deployments.

[0210] The apparatus provided in the embodiments of this application is described below. The apparatus described below can be referred to in correspondence with the method described above.

[0211] Figure 8 This is a schematic diagram of the terminal guidance device for intercepting unmanned aerial vehicles based on ground-based visual servo guidance provided in this application, as shown below. Figure 8 As shown, the device includes: The position estimation module 810 is used to simultaneously acquire observation data of at least one target UAV based on at least two ground-deployed visual sensors, calculate the position estimate of each target UAV and the estimation error range characterizing the uncertainty of the position estimate; The air-ground coordination module 820 is used to generate guidance reference points for at least one interceptor drone to search for at least one target drone based on the position estimate and the estimation error range, and to generate search guidance instructions for each interceptor drone based on the guidance reference points. The search guidance module 830 is used to optimize the search formation of the intercepting drone based on the estimated error range, generate control commands for the intercepting drone based on the search guidance commands, and drive the intercepting drone to search for the target drone.

[0212] The terminal guidance device for intercepting unmanned aerial vehicles (UAVs) based on ground-based visual servo guidance provided in this application achieves precise tracking of the target UAV by collecting observation data from a low-cost visual sensor deployed on the ground. It generates guidance reference points based on position estimates and estimation error ranges, and further generates search guidance commands for the intercepting UAV, achieving efficient conversion of ground detection information into airborne interception commands. By optimizing the search formation of the intercepting UAV through estimation error ranges, and generating control commands for the intercepting UAV through search guidance commands, it drives the intercepting UAV to search for the target UAV, achieving lightweight air-ground coordination. Overall, by constructing an air-ground coordinated architecture combining ground-based visual servo guidance and airborne autonomous search, it successfully solves the dual dilemmas of difficult detection and tracking and low cost-effectiveness of guidance in traditional anti-UAV methods. It significantly improves the response speed and interception success rate of the interception system against highly maneuverable targets in complex electromagnetic and low-altitude environments, while maintaining controllable overall system cost, demonstrating extremely high practical value and economic efficiency.

[0213] Figure 9 This is a schematic diagram of the structure of the electronic device provided in this application, such as... Figure 9 As shown, the electronic device may include: a processor 910, a communications interface 920, a memory 930, and a communications bus 940, wherein the processor, communications interface, and memory communicate with each other via the communications bus. The processor can invoke logical commands stored in the memory to execute the methods described in the above embodiments, for example: Based on the simultaneous acquisition of observation data from at least one target UAV by at least two ground-deployed visual sensors, the system calculates the position estimates of each target UAV and the estimation error range characterizing the uncertainty of the position estimates. Based on the position estimates and the estimation error range, it generates guidance reference points to guide at least one interceptor UAV to search for at least one target UAV, and generates search guidance commands for each interceptor UAV based on the guidance reference points. Based on the estimation error range, it optimizes the search formation of the interceptor UAVs, and generates control commands for the interceptor UAVs based on the search guidance commands, driving the interceptor UAVs to search for the target UAVs.

[0214] Furthermore, the logical commands in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several commands to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0215] The processor in the electronic device provided in this application embodiment can call logical instructions in the memory to implement the above method. Its specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effect, which will not be repeated here.

[0216] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments.

[0217] The specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, so it will not be repeated here.

[0218] This application provides a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0219] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0220] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0221] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance, characterized in that, include: Based on the simultaneous acquisition of observation data of at least one target UAV by at least two ground-deployed visual sensors, the position estimates of each target UAV and the estimation error range characterizing the uncertainty of the position estimates are calculated. Based on the estimated position and the estimated error range, a guidance reference point is generated to guide at least one interceptor drone to search for the at least one target drone, and search guidance instructions for each interceptor drone are generated based on the guidance reference point. The search formation of the intercepting drone is optimized based on the estimated error range, and control commands for the intercepting drone are generated based on the search guidance commands to drive the intercepting drone to search for the target drone.

2. The terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance according to claim 1, characterized in that, The process involves simultaneously acquiring observation data from at least one target UAV using at least two ground-deployed visual sensors, calculating the position estimates of each target UAV, and determining the estimation error range characterizing the uncertainty of the position estimates. This includes: Construct a matching cost matrix by integrating at least two of the following: geometric topology sorting constraints, back ray hard constraints, depth consistency constraints, and spatial distance threshold constraints; The line-of-sight relationships of the target UAV observed by the at least two visual sensors are matched based on the matching cost matrix to determine the line-of-sight correspondence; the line of sight is generated based on the observation data. Based on the line-of-sight correspondence, calculate the point pair with the closest spatial distance on each pair of matching lines of sight, and use the midpoint of the line connecting the point pairs as the estimated position of the target UAV. The estimated error range is determined based on the gap error of the matched line of sight and the physical measurement error of the vision sensor.

3. The terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance according to claim 2, characterized in that, The method further includes: In the event that matching fails due to occlusion of any target UAV, a primary visual sensor is determined among the at least two visual sensors, and a virtual guide point for the target UAV is generated along the observation direction of the primary visual sensor toward the target UAV.

4. The terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance according to claim 1, characterized in that, The step of generating guidance reference points based on the estimated position and the estimation error range to guide at least one interceptor drone to search for the at least one target drone includes: An uncertainty sphere is generated with the estimated position as the center and the estimated error range as the radius. If the minimum distance between the intercepting drone and the uncertainty sphere is greater than a preset mode switching distance threshold, a close-range search mode is adopted, and the center of the uncertainty sphere is used as the guiding reference point. When the minimum distance between the intercepting drone and the uncertainty sphere is less than or equal to a preset mode switching distance threshold, a close-range detection mode is adopted to generate multiple uniformly distributed sub-reference points inside the uncertainty sphere, and the sub-reference points are used as guiding reference points.

5. The terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance according to claim 1, characterized in that, The process of generating search guidance instructions for each intercepting drone based on the guidance reference point includes: Construct a task allocation cost matrix that includes at least one of path cost, angle cost, and load penalty cost; the path cost is determined based on the distance between the current position of the intercepting UAV and the guidance reference point; the angle cost is determined based on the deviation between the current observation angle of the intercepting UAV and the observation angle corresponding to the guidance reference point; the load penalty cost is determined based on the number of intercepting UAVs allocated in the search area defined by the estimation error range. Based on the task allocation cost matrix, the guidance reference point is assigned to each intercepting drone, and the position information of the guidance reference point is used as the search guidance instruction for each intercepting drone.

6. The terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance according to claim 1, characterized in that, Optimizing the search formation of the intercepting drone based on the estimated error range includes: A fitness function is constructed based on at least one of the following: the total distance between the intercepting UAV and the guiding reference point, the penalty for the number of times the search formation of the intercepting UAV overlaps, the penalty for the coverage uniformity of the search area defined by the estimation error range, and the penalty for the maneuver smoothness of the intercepting UAV. Based on the fitness function, a particle swarm genetic hybrid intelligent optimization algorithm is used to optimize the search formation of the intercepting drone.

7. The terminal guidance method for intercepting unmanned aerial vehicles based on ground-based visual servo guidance according to claim 1, characterized in that, The control commands for generating the intercepting drone based on the search guidance commands include: The search guidance command is sent to the controller of the intercepting drone. The controller determines the position information of the guidance reference point based on the search guidance command, determines the distance deviation based on the current position of the intercepting drone and the position information of the guidance reference point, and generates the control command based on the distance deviation. The controller is an enhanced robust proportional-integral-derivative controller.

8. A terminal guidance device for intercepting unmanned aerial vehicles based on ground-based visual servo guidance, characterized in that, include: The position estimation module is used to simultaneously acquire observation data of at least one target UAV based on at least two ground-deployed visual sensors, calculate the position estimate of each target UAV and the estimation error range characterizing the uncertainty of the position estimate; The air-ground coordination module is used to generate guidance reference points for at least one interceptor drone to search for at least one target drone based on the position estimate and the estimation error range, and to generate search guidance instructions for each interceptor drone based on the guidance reference points. The search guidance module is used to optimize the search formation of the intercepting drone based on the estimated error range, generate control commands for the intercepting drone based on the search guidance commands, and drive the intercepting drone to search for the target drone.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the terminal guidance method for intercepting unmanned aerial vehicles based on ground visual servo guidance as described in any one of claims 1 to 7.

Images