A method and system for cooperative target tracking and assignment of a drone swarm

Abstract

The application relates to the technical field of unmanned aerial vehicle control, and discloses a method and system for cooperative target tracking and distribution of unmanned aerial vehicle groups, the method comprising the following steps: acquiring three-dimensional terrain data and current attitude data to perform trajectory extrapolation processing and obtain a target prediction position; combining the current spatial position and the gimbal angle to construct a field of view cone range and evaluate the terrain truncation degree to obtain a line-of-sight occlusion probability; performing perspective cooperation processing according to cooperative pose data to obtain a global perspective distribution matrix; performing projection calculation according to the global perspective distribution matrix and a dynamic height datum to obtain a field of view coverage union set and a field of view overlap area; performing difference set operation on a target monitoring area and the field of view coverage union set to extract a monitoring blind area boundary and solve, and obtaining an angle adjustment instruction; and generating a cooperative tracking instruction according to the angle adjustment instruction and a collision avoidance safety radius. The method can realize continuous and stable tracking of a target under complex terrain.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and system for cooperative target tracking and allocation in UAV swarms. Background Technology

[0002] Currently, swarm-based collaborative target tracking plays a crucial role in modern space exploration and security patrol missions. Conventional tracking methods often rely on each drone independently locking onto local features of the target for follow-up.

[0003] In existing technologies, visual intelligence algorithms are typically used to extract the apparent features of targets for tracking. While individual drones can achieve basic collaborative tracking through information sharing, they still rely on reactive adjustments based on local observation perspectives during replacement scheduling. Assessment of terrain occlusion depends solely on real-time video feedback, lacking occlusion probability prediction based on field-of-view geometry and 3D terrain models. In environments with complex terrain undulations, the target's attitude can drastically change with the terrain. Due to the lack of a spatial coordination mechanism based on global observation angle allocation in existing technologies, blind adjustments by individual drones often result in multiple drones crowding the same observation direction. This not only causes significant overlap and waste of field of view but also completely exposes the other side of the target to blind spots, leading to a decrease in the continuity of swarm tracking and the integrity of monitoring.

[0004] Existing technologies have technical problems such as overlapping fields of view and blind spots in collaborative observation of aircraft groups under complex terrain, making it impossible to continuously and stably track targets. Summary of the Invention

[0005] This invention provides a method and system for collaborative target tracking and allocation by a swarm of unmanned aerial vehicles (UAVs) to solve the technical problems in the prior art where collaborative observation by a swarm in complex terrain easily results in overlapping fields of view and blind spots, making it impossible to continuously and stably track targets.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs), comprising: Acquire 3D terrain data of complex terrain and current attitude data of moving target, and perform trajectory extrapolation to obtain the predicted position of target; The current spatial position and gimbal angle of each UAV in the cluster are obtained. Based on the target prediction position, the current spatial position and the gimbal angle, a field of view cone range is constructed. The degree of terrain truncation of the field of view cone range is evaluated based on the three-dimensional terrain data to obtain the probability of line of sight occlusion. If the line-of-sight occlusion probability is greater than the preset occlusion tolerance threshold, then the cooperative pose data of the cooperating UAV in the monitoring airspace is acquired, and the cooperative pose data is processed by viewpoint cooperation to obtain a global viewpoint distribution matrix. Projection calculations are performed based on the global view distribution matrix to obtain the field of view coverage union and the field of view overlap region; If the overlapping area of ​​the field of view is greater than the preset overlap threshold, then the difference operation is performed on the union of the area to be monitored and the coverage of the field of view to extract the boundary of the monitoring blind zone, and the offset compensation calculation is performed on the boundary of the monitoring blind zone to obtain the angle adjustment command. Based on the cooperative pose data, the angle adjustment command, and the preset anti-collision safety radius, obstacle avoidance trajectory planning is performed to generate cooperative tracking commands.

[0007] Secondly, the present invention provides a collaborative target tracking and allocation system for unmanned aerial vehicle (UAV) swarms, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains the predicted target position by extrapolating the trajectory from the acquired three-dimensional terrain data and the current attitude data of the moving target. It then constructs the field-of-view cone range by combining the current spatial position of the UAV with the gimbal angle, and assesses the probability of line-of-view occlusion based on the degree of terrain truncation. Thus, this invention changes the traditional UAV model of passively adjusting after blindly waiting for the target to be lost due to localized field of view. It combines the physical characteristics of terrain undulations with the depth of kinematic trajectory prediction, quantifying the degree of interference of complex environments on the observation line of sight in advance and with precision. Ultimately, it enables the proactive and smooth triggering of a swarm collaborative replenishment mechanism just before the target enters the terrain blind zone, reducing the probability of target loss in complex environments.

[0010] (2) This invention constructs a global view distribution matrix by acquiring cooperative pose data when the occlusion probability is too high, and projects it downward onto a dynamic elevation datum to calculate the union and overlapping areas of the field of view coverage. Then, after extracting the boundary of the monitoring blind spot, it performs offset compensation calculation to obtain the angle adjustment command. Thus, this invention utilizes rigorous spatial geometric projection and Boolean difference set operations to accurately analyze and lock the resource waste areas, i.e., the overlapping areas, and the perception discontinuities, i.e., the blind spots, in the swarm observation network. This completely breaks the visual limitations of individual machines operating independently, transforming the scheduling of observation angles from disordered squeezing into globally optimal geometric complementarity. Ultimately, it effectively eliminates field of view coverage redundancy, ensuring that the target is not exposed in any monitoring blind spot, and improving the coverage efficiency of the three-dimensional observation network.

[0011] (3) This invention plans candidate spatial coordinates by combining collaborative pose data and angle adjustment commands, and introduces a collision avoidance safety radius and a virtual repulsion force field to perform boundary obstacle avoidance verification on the predicted trajectory spacing during the movement process, and finally controls the swarm to reach the corrected position. Thus, while pursuing optimal field of view coverage compensation, this invention is closely linked to the physical safety constraints of the multi-aircraft dynamic collaboration underlying layer, and resolves the risk of trajectory crossing and aircraft interference that is very likely to occur in emergency replacement scheduling of the swarm in advance by correcting the position offset. Ultimately, it achieves both the absolute safety of the UAV swarm flight scheduling and the maintenance of a stable, high-fidelity seamless continuous tracking state in complex terrain. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the unmanned aerial vehicle (UAV) swarm cooperative target tracking method provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the probability assessment of field of view cone and terrain truncation provided in the first embodiment of the present invention; Figure 3 This is a schematic diagram of the extraction and angle adjustment of the monitoring blind zone boundary provided in the first embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides a method for cooperative target tracking by a swarm of unmanned aerial vehicles (UAVs), comprising the following steps: S1: Acquire 3D terrain data of complex terrain and current attitude data of moving target, and perform trajectory extrapolation to obtain the predicted position of target; S2, obtain the current spatial position and gimbal angle of each UAV in the cluster, construct the field of view cone range based on the target prediction position, the current spatial position and the gimbal angle, and evaluate the degree of terrain truncation of the field of view cone range based on the three-dimensional terrain data to obtain the line of sight occlusion probability. S3, if the line-of-sight occlusion probability is greater than the preset occlusion tolerance threshold, then acquire the cooperative pose data of the cooperative UAV in the monitoring airspace, and perform viewpoint cooperative processing on the cooperative pose data to obtain a global viewpoint distribution matrix. S4. Projection calculation is performed based on the global view distribution matrix to obtain the field of view coverage union and the field of view overlap region. S5. If the overlapping area of ​​the field of view is greater than the preset overlap threshold, then the difference operation is performed on the union of the monitoring area and the coverage of the field of view to extract the monitoring blind zone boundary, and the offset compensation calculation is performed on the monitoring blind zone boundary to obtain the angle adjustment command. S6. Based on the cooperative pose data, the angle adjustment command, and the preset anti-collision safety radius, an obstacle avoidance trajectory is planned, and a cooperative tracking command is generated.

[0015] In step S1, three-dimensional terrain data of complex terrain and current attitude data of moving target are acquired, and trajectory extrapolation is performed to obtain the predicted position of target.

[0016] This involves extrapolating the trajectory to obtain the predicted target location, including: The target heading angle and centroid displacement are extracted based on the three-dimensional terrain data and the current attitude data of the moving target. The target's real-time linear velocity is calculated using the centroid displacement and a preset observation time interval. The target's heading angle is differentiated according to the preset observation time interval to obtain the heading angle change rate. The trajectory curvature radius is determined by combining the heading angle change rate with the target's real-time linear velocity. The trajectory curvature radius is fitted using a pre-defined nonholonomic constrained kinematic model, and the predicted target position is generated by extrapolation.

[0017] In one implementation, an airborne lidar collects a discrete set of 3D points containing spatial coordinate information, which is then used to determine the 3D terrain data. An airborne inertial navigation unit reads the roll, pitch, and yaw state scalars of the moving target in real time, and combines them to determine the current attitude data. This embodiment extracts two consecutive frames of 3D terrain data, uses an iterative nearest-point algorithm to perform spatial geometric matching on the local point cloud enveloping the moving target in the two frames, calculates the coordinate difference of the geometric center of the target's bounding rectangle in the 3D coordinate system, and obtains the centroid displacement. Simultaneously, the yaw scalar value is extracted from the current attitude data and determined as the target heading angle.

[0018] It should be noted that the preset observation time interval is determined by the physical sampling hardware parameters of the airborne sensors. The system reads the physical scanning frequency value triggered synchronously by the airborne lidar and the airborne inertial navigation unit, divides a constant by this physical scanning frequency value, and the reciprocal of the result is directly determined as the preset observation time interval. In this embodiment, the Euclidean distance scalar of the centroid displacement is divided by the preset observation time interval to calculate the target's real-time linear velocity. The target's heading angle in the current frame and the heading angle in the previous historical frame are extracted, and the angle difference between the two is calculated. This angle difference is divided by the preset observation time interval to perform a first-order difference calculation to obtain the heading angle change rate. In this embodiment, the target's real-time linear velocity is divided by the absolute value of the heading angle change rate to calculate the trajectory curvature radius, which characterizes the degree of trajectory curvature.

[0019] It is worth noting that the preset nonholonomic constrained kinematic model is a set of constant curvature two-dimensional kinematic integral equations constructed based on the Ackermann steering geometry principle. This set of equations assumes that the target has no lateral sideslip displacement in a short time. The iterative value of its ordinate plane is the product of the previous moment's ordinate, the linear velocity, and the sine of the heading angle, multiplied by the time step; the iterative value of its abscissa plane is the product of the previous moment's abscissa, the linear velocity, and the cosine of the heading angle, multiplied by the time step. In this embodiment, the physical control cycle constant of the airborne flight control system's underlying hardware, i.e., the time interval between two consecutive control commands, is extracted and determined as the time extrapolation step. The current spatial coordinates are used as the initial reference point for integration. The trajectory curvature radius, the target's real-time linear velocity, and the target's heading angle are substituted into the set of equations to perform numerical accumulation and integration. The two-dimensional plane coordinate parameters output at the integration endpoint are extracted and determined as the target's predicted position.

[0020] For example, the airborne lidar has a physical scanning frequency of 20 Hz, and the system calculates a preset observation time interval of 0.05 seconds. After matching calculation, the centroid displacement of the moving target within 0.05 seconds is 0.4 meters, the current target heading angle is 45.5 degrees, and the historical heading angle of the previous frame is 44.5 degrees. The system divides 0.4 meters by 0.05 seconds to calculate the target's real-time linear velocity as 8.0 meters per second. The angle difference between 45.5 degrees and 44.5 degrees is converted to radians and divided by 0.05 seconds to obtain the heading angle change rate. Subsequently, the system divides 8.0 meters per second by the absolute value of this heading angle change rate to obtain the trajectory curvature radius. The system substitutes the above scalar parameters into the constant curvature two-dimensional kinematic integral equation system, iterates ten times with an integration step size of 0.1 seconds, and accurately extrapolates the target's predicted position 1.0 second ahead.

[0021] In step S2, the current spatial position and gimbal angle of each UAV in the cluster are obtained. The field of view cone range is constructed based on the target prediction position, the current spatial position and the gimbal angle. The degree of terrain truncation of the field of view cone range is evaluated based on the three-dimensional terrain data to obtain the line of sight occlusion probability.

[0022] The process of constructing the field of view cone range based on the predicted target location, the current spatial location, and the gimbal angle includes: Based on the current spatial location and the predicted target location, a spatial vector is constructed by performing spatial vector analysis; The target observation center axis is determined by combining the line-of-sight space vector with the gimbal angle; Using the target observation center axis as a reference, the spatial extension boundary is calculated to construct the field of view cone range.

[0023] In one implementation, the absolute three-dimensional Cartesian coordinates of each UAV are read by an onboard high-precision real-time dynamic differential positioning system, and these coordinates are used to determine the current spatial position. The yaw and pitch angles relative to the fuselage are extracted by reading the encoder feedback values ​​of the brushless gimbal motors, and combined to determine the gimbal angle. This embodiment utilizes the three-dimensional vector subtraction rule to subtract the coordinates of the current spatial position from the coordinates of the predicted target position, thus constructing a line-of-sight spatial vector pointing from the UAV to the target.

[0024] It should be noted that the projection direction of the line-of-sight space vector onto the local horizontal coordinate plane is taken as the reference zero degree, and the gimbal angle is subjected to a spatial coordinate system alignment transformation. The transformed yaw scalar and pitch scalar are extracted and synthesized into a three-dimensional vector, and the ray vector representing the actual optical acquisition center of the camera is calculated and determined as the target observation center axis.

[0025] It is worth noting that the field of view cone range is determined by the factory physical optical parameters of the airborne camera. The inherent values ​​of the horizontal and vertical field of view angles are extracted from the camera's hardware registers. If the absolute value of the difference between the two is less than or equal to 10 degrees, the arithmetic mean is calculated as the apex angle of the field of view cone. If the absolute value of the difference is greater than 10 degrees, half of each of the horizontal and vertical field of view angles is used as the two directional half-cone angles of the pyramid, constructing the pyramid as the field of view cone range. Using the target observation center axis as the spatial central axis of symmetry of the cone, spatial ray extension calculations are performed along the three-dimensional direction of this axis, using values ​​equal to half the apex angle of the cone, to construct and generate a geometric envelope surface with closed lateral boundaries, which is then determined as the field of view cone range.

[0026] The evaluation of the terrain truncation degree within the field of view cone range based on the three-dimensional terrain data to obtain the line-of-sight occlusion probability includes: Based on the three-dimensional terrain data, terrain elevation features are extracted, and spatial Boolean intersection is performed between the field of view cone range and the terrain elevation features to identify terrain protrusions. The ratio of the orthogonal projected area of ​​the terrain protrusion on the bottom surface of the field of view cone to the total cross-sectional area of ​​the cone is calculated to obtain the truncation ratio. Extract the current pitch angle contained in the gimbal angle, and combine it with the cosine factor of the current pitch angle to perform weight adjustment calculation on the cutoff ratio, and output the line-of-sight occlusion probability.

[0027] In one implementation, the three-dimensional terrain data is horizontally sliced ​​along the gravity axis at preset discrete intervals. The closed outer contours of two-dimensional polygons and their corresponding absolute elevation values ​​are extracted from each slice and combined to determine the terrain elevation features. A three-dimensional Boolean intersection is then performed between the geometric space enclosed by the field of view cone and all the closed outer contours of the two-dimensional polygons. If a non-empty geometrically connected region exists in the solution, this non-empty geometrically connected region is identified as a terrain protrusion.

[0028] It should be noted that the preset discrete spacing is determined based on the hardware scanning parameters of the airborne lidar. The spatial average resolution constant of adjacent point clouds in the 3D terrain data is extracted, for example, 0.5 meters in this embodiment, and this spatial average resolution constant is directly assigned a value to determine the preset discrete spacing. This effectively avoids the unnecessary computational overhead caused by oversampling without distorting the original terrain elevation features.

[0029] It should be noted that in this embodiment, along the normal direction of the target observation center axis, the two-dimensional area of ​​the terrain protrusion projected onto the largest base surface of the field of view cone is calculated using a polygon area integration algorithm, and this area is determined as the orthogonal projected area. Simultaneously, the total physical area of ​​this base surface within the field of view cone is calculated as the total cross-sectional area of ​​the cone. The orthogonal projected area is divided by the total cross-sectional area of ​​the cone, and the resulting real-valued scalar ratio is determined as the truncation ratio.

[0030] It is worth noting that the weight adjustment operation is performed through a pre-trained line-of-sight occlusion assessment model. This pre-trained model is a logistic regression-based probabilistic assessment model with a single-layer network structure without hidden layers, specifically containing two input nodes and one probability output node. The model is constructed and trained by building a digital terrain and UAV swarm model covering various terrain types such as mountains, hills, and urban canyons in a 3D physical simulation software. The simulated UAVs are then controlled to perform 100,000 tracking traversals of ground targets at different altitudes and viewpoints. The actual occlusion situation is accurately calculated using an offline ray intersection method, marking complete occlusion as a value of one and clearly visible areas as a value of zero, thus constructing a label set. Simultaneously, the truncation ratio and the corresponding pitch angle cosine value are extracted during the simulation process and combined to form a two-dimensional feature vector input set. The model uses binary cross-entropy as the loss function and employs maximum likelihood estimation combined with adaptive moment estimation algorithm for iterative updating and optimization of the weight parameters. The initial learning rate was set to 0.01. After 5,000 iterations, training stopped when the loss function on the validation set decreased by less than 0.001 for ten consecutive rounds, thus solidifying the two feature weight parameters and one bias parameter contained in the model. In this embodiment, the current pitch angle is extracted from the gimbal angle, and a trigonometric cosine function is performed to obtain the pitch angle cosine factor. This pitch angle cosine factor and the cutoff ratio are respectively input into the two input nodes of the evaluation model. The pre-trained and solidified feature weight parameters are linearly weighted and summed, and then the bias parameter is added. The result is then substituted into the Sigmoid activation function to perform a nonlinear mapping, outputting a continuous floating-point value between zero and one, which is determined as the line-of-sight occlusion probability.

[0031] For example, with the UAV hovering at a height of 120.0 meters, after the system constructs the target observation center axis and the field of view cone range, it identifies a terrain protrusion with a relative elevation of 8.5 meters in front of the line of sight through spatial Boolean intersection operation. After integration calculation, the orthogonal projected area of ​​this terrain protrusion on the bottom surface of the field of view is 320.0 square meters, while the total cross-sectional area of ​​the cone is 1000.0 square meters. The system divides 320.0 by 1000.0, obtaining a truncation ratio of 32.0%. The system extracts the current pitch angle of the gimbal as -45.0 degrees and calculates its corresponding pitch angle cosine factor to be approximately 0.707. The system inputs the 32.0% truncation ratio and the 0.707 cosine factor into a pre-trained logistic regression model. After parameter weighting and nonlinear activation, the model outputs a line-of-sight occlusion probability of 38.5%.

[0032] like Figure 2As shown in the figure, this diagram illustrates the geometric relationship between the UAV's field of view and complex terrain, as well as the occlusion risk assessment. The green irregular area represents the actual topographical undulations of the environment, and its height variation directly affects the unobstructed view of the UAV. The gray shaded area represents the portion of the UAV's field of view cone obstructed by terrain protrusions; the shape and size of this area intuitively reflect the physical occlusion of the view by the terrain. The blue triangle and red dot represent the spatial coordinates of the observation platform and the target, respectively, and the line connecting them forms the baseline direction of the line of sight. The red dashed boundary represents the lateral boundary of the field of view cone; when this boundary line crosses the gray truncated area, it means that the view in the corresponding direction is obstructed. The occlusion probability of 38.5% is a quantitative indicator calculated based on the proportion of the truncated area of ​​the field of view, comprehensively reflecting the probability of target visibility under the current terrain environment, and is a key judgment basis for triggering subsequent collaborative repositioning or perspective adjustment.

[0033] In step S3, if the line-of-sight occlusion probability is greater than a preset occlusion tolerance threshold, then the cooperative pose data of the cooperating UAVs in the monitored airspace is acquired, and the cooperative pose data is subjected to viewpoint cooperative processing to obtain a global viewpoint distribution matrix, including: If the line-of-sight occlusion probability is greater than the preset occlusion tolerance threshold, a cooperative replacement request is generated, and the cooperative pose data of the cooperative UAV in the monitoring airspace is obtained according to the cooperative replacement request. The current planar coordinates, azimuth angle allocation value, and pitch angle offset of the cooperative UAV are extracted based on the cooperative pose data. The terrain undulation gradient is calculated based on the three-dimensional terrain data, and the hovering altitude of the cooperative UAV is divided into steps based on the terrain undulation gradient to obtain altitude layer identifiers. The camera optical parameters of each of the cooperative drones are obtained, and the inter-drone distance between adjacent cooperative drones is calculated by combining the current planar coordinates. The inter-drone distance, the azimuth angle allocation value, the pitch angle offset, the altitude layer identifier and the camera optical parameters are then combined into a multi-dimensional tensor to construct a global view distribution matrix.

[0034] In one implementation, the line-of-sight occlusion probability is input into a numerical comparator and evaluated against a preset occlusion tolerance threshold. It should be noted that the preset occlusion tolerance threshold is obtained through a measured limit recognition boundary calibration experiment. In an open test area, a drone is controlled to dynamically photograph a standard calibration board with a physical size of one square meter at different heights and angles. The captured images are input into an airborne visual target detection algorithm, such as the YOLO series algorithm. The minimum percentage threshold value occupied by the calibration board in the total pixel area of ​​the image is recorded when the recall rate of the target detection algorithm first drops below 90% due to the introduction of an occlusion. In this embodiment, a constant is subtracted from this minimum percentage threshold value, and the difference is assigned as the preset occlusion tolerance threshold. If the evaluation result shows that the line-of-sight occlusion probability is greater than the preset occlusion tolerance threshold, a blocking Boolean flag is triggered in the underlying controller. The system then calls a data encapsulation protocol to stitch together and package the current trigger timestamp, the current device code, and the calculated blind zone 3D coordinates to generate a fixed-length cooperative filling request. The airborne wireless bridge module broadcasts the cooperative positioning request to the central control node and simultaneously receives real-time position streams and gimbal status streams from the node, which contain the working status of other UAVs in the airspace. These messages are then unpacked and identified as cooperative pose data.

[0035] It should be noted that the azimuth angle allocation value is pre-calculated by the central control node based on a global perspective optimization strategy and distributed to each cooperative UAV, and is included in the cooperative pose data. Bitmap field parsing is performed on the cooperative pose data to extract the longitude and latitude floating-point numbers of other cooperative UAVs in the geodetic coordinate system. These are then converted into rectangular coordinate parameters in a two-dimensional plane coordinate system using a general transverse Mercator projection algorithm to obtain the current plane coordinates. Simultaneously, the yaw angle relative to the aircraft's heading and the vertical pitch angle relative to the horizontal gravity plane, fed back by the airborne gimbal in the data packet, are extracted and directly determined as the azimuth angle allocation value and pitch angle offset, respectively.

[0036] It is worth noting that the terrain undulation gradient is calculated based on the aforementioned three-dimensional terrain data. Specifically, the maximum and minimum absolute elevation values ​​within the monitored area are retrieved from the three-dimensional terrain data, and the absolute height difference is calculated by arithmetic subtraction. This absolute height difference is then divided by the maximum two-dimensional horizontal diagonal physical distance of the monitored area, and the resulting real constant is determined as the terrain undulation gradient. Based on this preset terrain undulation gradient, a reference elevation zero point is established along the gravity direction. The vertical airspace is then discretized into mathematical intervals according to a preset arithmetic step size for each height layer, forming multiple height profile intervals. The real-time altitude value of each collaborative UAV is placed into these profile intervals for matching and determination. The natural number index label corresponding to the successfully matched interval is extracted and determined as the height layer identifier.

[0037] It should be noted that the preset height layer arithmetic step size is determined through the linkage calibration of the airborne camera optical parameters. The vertical physical field of view angle constant of each cooperating UAV airborne camera is extracted, and the effective observation distance limit value derived based on the limit resolution of the visual target detection algorithm is obtained. The vertical projection height of the field of view cone at this distance is solved using trigonometric functions. Half of this vertical projection height is rounded down to an integer multiple of ten, and this is determined as the preset height layer arithmetic step size, which is calculated to be 20 meters in this embodiment. This calibration method can geometrically ensure that the field of view between adjacent height layers can be seamlessly connected without generating excessive vertical redundancy.

[0038] The current planar coordinates of each drone are extracted, and the lengths of the straight line segments between the geometric center points of the drones in the coordinate system are calculated pairwise using the two-dimensional planar Euclidean distance formula, thus obtaining the planar inter-drone distance representing the proximity relationship. It should be noted that, to eliminate data discontinuity issues in the subsequent spatial projection calculation process, this embodiment retrieves the pre-written intrinsic optical parameters of each cooperative UAV's onboard camera from the system's local configuration library based on the device code, specifically including the camera's horizontal physical field of view constant and optical focal length constant. The system constructs a multidimensional tensor data structure with the cooperative UAV device code as the index axis. In this embodiment, the planar inter-drone distance representing spatial topological relationships, the azimuth angle allocation value and the pitch angle offset representing the observation orientation, the height layer identifier representing the vertical level, and the camera's horizontal physical field of view constant and optical focal length constant used to calculate the projection section are sequentially used as independent numerical dimension vectors. Memory-level data stitching and matrix alignment operations are performed along the high-dimensional feature stitching axis of this multidimensional data structure to generate a high-dimensional data array that combines spatial geometric position information and underlying optical hardware parameters, and its output is determined as a global viewpoint distribution matrix.

[0039] For example, calibration experiments showed that when occlusion caused the pixel percentage of the calibration board to fall below 35%, the target detection recall rate dropped below 90%. The system then subtracted 35% from the constant to calculate a preset occlusion tolerance threshold of 65.0%. When the calculated 38.5% line-of-sight occlusion probability did not exceed this threshold, the drone fleet maintained its current formation. If severe occlusion caused the probability to surge to 70.0% and exceed the threshold, the system immediately broadcast a cooperative replacement request. After acquiring the cooperative pose data, the system calculated the planar coordinates of a cooperative drone as (150.0, 200.0) meters, with an azimuth angle of 45.0 degrees and a pitch angle offset of -30.0 degrees. The system used the maximum altitude of 450 meters, the minimum altitude of 50 meters, and the distance diagonal to calculate the preset terrain undulation gradient. Finally, the system will combine the calculated inter-machine distance, various angle labels, and the camera's 60-degree field of view constant extracted from the configuration library in memory to construct a global view distribution matrix containing complete observation state parameters.

[0040] In step S4, projection calculation is performed based on the global view distribution matrix to obtain the field of view coverage union and the overlapping region of the field of view, including: The three-dimensional field-of-view parameters and gimbal optical axis pointing of the collaborative UAV are extracted based on the global view distribution matrix. A dynamic elevation datum is constructed based on the absolute elevation corresponding to the predicted target location. Using the gimbal optical axis of each of the cooperative UAVs as the projection direction, the three-dimensional field of view parameters are intersected with the dynamic elevation reference plane by ray intersection calculation to generate a field of view projection polygon. Boolean union operation is performed on all the field-of-view projection polygons to obtain the field-of-view coverage union, and the intersection region covered by at least two field-of-view projection polygons is extracted and pixel grid integral summation operation is performed to obtain the field-of-view overlap region.

[0041] In one implementation, the global view distribution matrix constructed in the preceding steps is read. The physical field-of-view angle constant and optical focal length constant of the cooperative UAV's onboard camera, pre-bound according to the device index, are extracted from the matrix and combined to determine the three-dimensional field-of-view parameters. Simultaneously, the azimuth angle allocation value and pitch angle offset stored in the matrix are extracted, and a three-dimensional spatial vector rotation matrix calculation is performed to map the zero-degree pointing vector of the local coordinate system to the current observation orientation. The three-dimensional spatial unit vector representing the direction of the camera lens's central symmetry axis is calculated and determined as the gimbal's optical axis pointing direction.

[0042] It should be noted that the dynamic elevation datum is constructed by dynamically extracting the real-time spatial elevation of the target. The absolute elevation corresponding to the predicted target position in the preceding steps is extracted as a reference elevation constant. A two-dimensional horizontal spatial plane equation, whose absolute height is always equal to this reference elevation constant, is established in the local coordinate system and defined as the dynamic elevation datum. This datum does not fluctuate with the undulations of the original offline terrain but remains locked on the current physical profile where the target is located, used to uniformly quantify the true effective field-of-view overlap of the drone swarm around the target. Using the three-dimensional spatial coordinates of each cooperative UAV extracted from the matrix as the starting point of the geometric rays, along the corresponding gimbal optical axis, the boundary of the truncated quadrangular spatial beam containing the three-dimensional field-of-view parameters is converted into four discrete spatial straight-line ray equations. The spatial straight-line ray equations are simultaneously solved analytically with the two-dimensional horizontal spatial plane equation of the dynamic elevation datum. The four intersecting coordinate point sequences calculated are extracted and closed in a clockwise direction to generate a two-dimensional region with absolute coordinate boundaries, which is defined as the field-of-view projection polygon.

[0043] It is worth noting that the Boolean union operation and the Boolean intersection pixel grid integral summation operation are performed using a two-dimensional graphic polygon grid discretization statistical algorithm. In this embodiment, a global two-dimensional blank grid matrix of the same size as the target monitoring area is constructed. All the projected polygons of the field of view are traversed, and a polygon interior point determination algorithm is used to mark the grid cells within the polygon envelope as valid coverage states. The set of all grid cells marked as valid coverage states in the global two-dimensional grid matrix is ​​extracted to obtain the field of view coverage union. For calculating the overlapping region of the field of view, this embodiment constructs an independent global overlap Boolean mask matrix; iterates through each grid cell in the global two-dimensional grid matrix and counts the number of field-view projected polygons enclosing the grid cell; if the number is greater than or equal to 2, the grid cell is extracted as the intersection region covered by at least two field-view projected polygons, and the grid cells inside the Boolean intersection are marked as overlapping in the global overlap Boolean mask matrix; after the traversal, the total number of grid cells in the overlapping state in the global overlap Boolean mask matrix is ​​summed, and the sum is multiplied by the physical real area of ​​a single grid cell to calculate the numerical result, which is then determined as the overlapping region of the field of view.

[0044] For example, the system extracts the 60-degree horizontal field-of-view constant of UAV No. 3 from the global view distribution matrix as a three-dimensional field-of-view parameter, and extracts its corresponding gimbal optical axis pointing vector. The system reads the plane equation with a reference altitude of 5.0 meters as a dynamic elevation reference surface, and uses ray intersection operation to calculate the coordinates of the four intersection points between the beam ray and the ground, generating a trapezoidal field-of-view projection polygon. The system performs a mesh union operation on the field-of-view projection polygons of 12 UAVs to obtain the field-of-view coverage union. Subsequently, the system performs pairwise intersection calculations on adjacent UAV pairs, and projects all intersection meshes uniformly onto a global overlap Boolean mask matrix for state marking. After global traversal, the system counts that the total number of meshes marked as overlapping in the mask matrix is ​​2060, multiplied by the area of ​​a single mesh of 2.0 square meters, and calculates that the total field-of-view overlap area is 4120.0 square meters.

[0045] In step S5, if the overlapping area of ​​the field of view is greater than a preset overlap threshold, the difference operation is performed on the union of the monitoring area and the coverage of the field of view to extract the monitoring blind zone boundary, and the offset compensation calculation is performed on the monitoring blind zone boundary to obtain the angle adjustment command.

[0046] Specifically, the process involves performing a difference operation on the union of the area to be monitored and the field of view coverage to extract the boundary of the monitoring blind zone, and then performing offset compensation calculation on the boundary of the monitoring blind zone to obtain an angle adjustment command, including: The boundary features of the uncovered area are extracted by performing a difference operation on the union of the area to be monitored and the field of view coverage, and are used as the boundary of the monitoring blind zone. Geometric feature extraction is performed on the union of the field of view coverages to obtain the coordinates of the coverage center; The geometric centroid coordinates are obtained by calculating the geometric centroid of the monitoring blind zone boundary, and the centroid displacement is obtained by vector subtraction between the geometric centroid coordinates and the coverage center coordinates. The gimbal yaw rotation parameters and pitch tilt parameters are calculated based on the centroid displacement and the preset proportional-integral-derivative control strategy, and an angle adjustment command is generated based on the gimbal yaw rotation parameters and the pitch tilt parameters.

[0047] In one implementation, the overlapping area of ​​the field of view is input to a numerical comparator and compared with a preset overlap threshold. It should be noted that the preset overlap threshold is calibrated through offline tracking simulation experiments in a three-dimensional dynamic environment. A digital elevation model consistent with the measured terrain is constructed in the simulation environment, simulating a target object moving along a random path at a speed of 15 meters per second. Test overlap area thresholds ranging from 0 square meters to 5000 square meters are dynamically set in the simulation system. A swarm cooperative algorithm is run, and the miss rate of the target leaving the global field of view is statistically analyzed. The minimum allowable overlap area value corresponding to the first drop in miss rate below one percent is extracted and set as the preset overlap threshold. If the overlapping area of ​​the field of view is larger than the preset overlap threshold, the absolute spatial closed polygon boundary coordinate sequence imported through the offline task planning file is extracted and identified as the area to be monitored. A two-dimensional Boolean difference operation is performed on the union of the area to be monitored and the field of view coverage, and the outer contour features of the polygons not covered by the current field of view in the result are extracted and identified as the monitoring blind zone boundary.

[0048] It is worth noting that the coordinates of the coverage center are obtained by geometric feature extraction of the union of the field of view coverage. In this embodiment, the coordinate sequences of all discrete vertices on the outer contour of the union of the field of view coverage are extracted, and the arithmetic mean of the horizontal and vertical coordinates of all vertices is performed. The calculated two-dimensional plane average coordinate point is determined as the coordinates of the coverage center. Similarly, the integral formula of the area of ​​the centroid of a closed polygon is used to solve the contour coordinate sequence of the monitoring blind zone boundary to calculate the geometric centroid coordinates representing the physical distribution center of the blind zone. In this embodiment, the horizontal and vertical components of the coverage center coordinates are subtracted from the horizontal and vertical components of the geometric centroid coordinates to obtain a two-dimensional displacement vector with direction and magnitude attributes, which is determined as the centroid displacement containing horizontal and vertical components.

[0049] It should be noted that the gimbal yaw rotation parameters and pitch tilt parameters are calculated based on the centroid displacement and the preset proportional-integral-derivative control strategy. This embodiment performs a geometric transformation mapping from position to angle to obtain the horizontal straight-line physical distance between the current two-dimensional coordinates of the target cooperative UAV and the coordinates of the coverage center. The lateral component of the centroid displacement is divided by this horizontal straight-line physical distance, and the quotient is used to solve for the arctangent function. The resulting angle value is determined as the gimbal yaw angle correction input. Similarly, the longitudinal component of the centroid displacement is divided by this horizontal straight-line physical distance to solve for the arctangent function, and the resulting angle value is determined as the gimbal pitch angle correction input.

[0050] Subsequently, the two angle correction inputs are substituted into the preset proportional-integral-derivative (PID) control strategy. The proportional, integral, and derivative coefficients in the preset PID control strategy are calibrated using the Ziegler-Nichols closed-loop tuning rule. Through frequency response testing, the airborne gimbal angle servo motor is identified as a first-order inertial physical model with pure hysteresis, for example, with a transfer function set to a gain of 1.5, a time constant of 0.2 seconds, and a hysteresis of 0.05 seconds. Based on this transfer function, the critical proportional gain rule is applied, and the proportional coefficient is calculated to be 12.5, the integral coefficient to be 0.8, and the derivative coefficient to be 0.0. Combining the above calibration coefficients, a linear weighted summation operation is performed on the angle correction input and the historical error integral term to calculate the gimbal yaw rotation parameters and pitch tilt parameters that drive the gimbal motor. Finally, the calculated parameters and the device identification code are encapsulated into a low-level communication protocol data packet to generate an angle adjustment command.

[0051] For example, the system reads that the current field of view overlap area is 4120.0 square meters, which is greater than the preset overlap threshold of 3000.0 square meters calibrated in the simulation. The system extracts the monitoring blind zone boundary with an area of ​​856.0 square meters. The system solves for the geometric centroid coordinates of the blind zone and subtracts the coverage center coordinates from them, calculating that the lateral component of the two-dimensional centroid displacement is 21.5 meters and the longitudinal component is 8.4 meters. It is known that the horizontal straight-line physical distance between the cooperative UAV and the coverage center is 100.0 meters. The system uses the arctangent function to convert the displacement into angular input quantities, with a yaw angle deviation of 12.1 degrees and a pitch angle deviation of 4.8 degrees. The system uses these two deviation quantities as inputs, and after the proportional-integral-differential algorithm loop operation, it calculates the actual compensation parameters for the gimbal to yaw by 12.5 degrees and tilt by 4.9 degrees, and packages and issues the angle adjustment command.

[0052] like Figure 3 As shown in the figure, this diagram illustrates the process of blind spot identification and gimbal control calculation based on the field of view coverage results. The black rectangle represents the total spatial range required for mission coverage. The light blue filled area represents the total area observable after the current viewpoints of all UAVs are superimposed. As can be seen in the figure, this area does not completely fill the black monitoring frame, resulting in uncovered blank areas. The red closed line delineates the boundary between the light blue coverage area and the black monitoring area, representing the visual blind spot identified by the system. The blue cross marks the geometric centroid of the light blue coverage area, representing the current concentration trend of the UAV group's field of view. The red cross marks the geometric centroid of the red blind spot boundary, representing the core location of the blind spot. The displacement (21.5, 8.4)m is the vector connecting the coverage center and the blind spot centroid. The projection components of this vector on the X and Y axes are 21.5 meters and 8.4 meters, respectively, quantifying the degree and direction of the blind spot's deviation from the current field of view center. This displacement is the direct input parameter for calculating the gimbal adjustment angle.

[0053] In step S6, obstacle avoidance trajectory planning is performed based on the cooperative pose data, the angle adjustment command, and the preset anti-collision safety radius to generate cooperative tracking commands, including: By combining the cooperative pose data with the angle adjustment command, spatial coordinate mapping calculations are performed to generate candidate spatial coordinates for each of the cooperative UAVs; Calculate the predicted trajectory spacing between adjacent cooperative UAVs as they move toward the candidate spatial coordinates; If it is determined that the predicted trajectory spacing is less than the preset anti-collision safety radius, a repulsive force field is generated at the collision boundary, and the body position offset is calculated. The candidate spatial coordinates are corrected by combining the fuselage position offset to generate a cooperative tracking command.

[0054] In one implementation, the initial spatial parameters of the UAVs are extracted from the cooperative pose data. Based on the required observation viewpoint position specified in the angle adjustment command, a spatial analytical geometric translation transformation is performed to map and calculate the expected target hovering three-dimensional coordinates of each cooperative UAV, which are then determined as candidate spatial coordinates. Linear spatial discretization interpolation is performed between the initial position coordinates and the candidate spatial coordinates to generate corresponding flight prediction trajectory segments. Using the three-dimensional skew-plane straight-line spacing formula, the minimum physical straight-line distance between any two adjacent cooperative UAV flight prediction trajectory segments is calculated and determined as the prediction trajectory spacing.

[0055] It should be noted that the predicted trajectory spacing is numerically compared with a preset collision avoidance safety radius. The preset collision avoidance safety radius is determined by extracting the maximum wingspan diameter constant of the UAV's physical rotor and the maximum three-dimensional positioning error range constant of the airborne real-time dynamic differential positioning system, and then summing the two arithmetically. If it is determined that the predicted trajectory spacing is less than the preset collision avoidance safety radius, an artificial potential field method is introduced to generate a repulsive force field at the boundary where the predicted trajectory spatial interference occurs.

[0056] It is worth noting that the logic for generating the repulsive force field and converting it into the fuselage position offset is as follows: Define the repulsive force field function of the artificial potential field. When the predicted trajectory spacing is less than the safe radius, the virtual repulsive force magnitude is equal to the preset repulsive force gain coefficient multiplied by the difference between the reciprocal of the safe radius and the reciprocal of the trajectory spacing, then divided by the square of the trajectory spacing; wherein the virtual repulsive force direction radiates outwards along the line connecting the physical geometric centers of adjacent fuselages. In this embodiment, after calculating the virtual repulsive force magnitude, it is divided by the fuselage physical mass constant according to Newton's second law to obtain a virtual acceleration scalar. Combining this with the discrete time step of the flight control system, the virtual acceleration scalar is multiplied by the square of the time step, then multiplied by half the constant to calculate the single-step absolute distance scalar for interference avoidance. This absolute distance scalar is mapped along the repulsive force direction to a three-dimensional spatial vector, which is then determined as the fuselage position offset.

[0057] It should be noted that the preset repulsive force gain coefficient is determined through dynamic physical constraint calibration. The maximum physically available thrust constant of the UAV brushless motor, calibrated at the factory, is obtained, and then multiplied by the square of the preset collision avoidance safety radius. The product result is determined as the preset repulsive force gain coefficient. This calibration method ensures that the output virtual repulsive force is strictly constrained within the physical limits of the airframe's flight control system, avoiding the calculation of displacements exceeding the motor's performance capabilities.

[0058] In this embodiment, the three-dimensional vector of the fuselage position offset is added to the corresponding candidate spatial coordinates to achieve iterative correction of the position parameters. The safe coordinate parameters confirmed to be free from collision interference, together with the gimbal attitude parameters, are encoded into machine code recognizable by the underlying ESC to generate a cooperative tracking instruction; the execution steps are returned to S1 until the tracking task ends.

[0059] For example, given that the UAV's rotor wingspan is 1.2 meters and its GPS positioning error is 0.5 meters, the system calculates a preset collision avoidance safety radius of 1.7 meters. The spatial topology analysis module detects that the predicted trajectory distance between adjacent UAVs 4 and 5 has decreased to 1.2 meters, triggering the collision avoidance logic. The system substitutes a repulsive gain coefficient of 1000 for calculation, resulting in a virtual repulsive modulus of 170.0 Newtons. Given that the UAV's mass is 5.0 kg and the control cycle time step is 0.1 seconds, the system uses the dynamic displacement formula to calculate a single-step UAV position offset of 0.17 meters. After three iterative cumulative corrections, the total offset reaches 0.51 meters, successfully expanding the distance to 1.71 meters and removing it from the collision risk boundary. The system then packages the fine-tuned safety coordinates into a cooperative tracking command and issues it.

[0060] In summary, this invention extrapolates kinematic trajectories by acquiring 3D terrain and the target's current posture, and performs Boolean intersection operations on the field of view cone constructed by combining the aircraft's pose with terrain elevation features to quantify the probability of line-of-sight occlusion. When occlusion exceeds the limit, it deeply fuses the collaborative pose of the aircraft swarm and camera optical parameters to construct a global view distribution matrix, and uses rigorous analytical geometric projection and a global deduplication grid integration algorithm to accurately extract field-of-sight overlap redundancy and monitoring blind zone boundaries. Furthermore, it calculates the gimbal compensation angle through arctangent mapping of geometric centroid displacement and proportional-integral-differential closed-loop control, while also introducing... The artificial potential field method and dynamic displacement formula are used to perform rigorous anti-collision displacement correction on the predicted trajectory, and finally control the swarm to reach the corrected position. This realizes the leap from passive local blind tracking to global active collaborative perception based on three-dimensional terrain and motion prediction. It breaks the limitations of mutual compression of the swarm's field of vision and frequent blind spots in complex environments, effectively avoids the risk of trajectory collision in multi-aircraft emergency replacement, and finally achieves global optimal coordination and seamless full-space coverage of the swarm's observation perspective. This improves the robustness, coverage efficiency and security of the underlying physical scheduling of continuous target tracking in complex terrain.

[0061] The second embodiment of the present invention provides a collaborative target tracking and allocation system for unmanned aerial vehicle swarms, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the method described above.

[0062] It should be noted that the UAV swarm cooperative target tracking and allocation system provided in this embodiment of the invention is used to execute all the process steps of the UAV swarm cooperative target tracking and allocation method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0063] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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.

[0064] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs), characterized in that, include: Acquire 3D terrain data of complex terrain and current attitude data of moving target, and perform trajectory extrapolation to obtain the predicted position of target; The current spatial position and gimbal angle of each UAV in the cluster are obtained. Based on the target prediction position, the current spatial position and the gimbal angle, a field of view cone range is constructed. The degree of terrain truncation of the field of view cone range is evaluated based on the three-dimensional terrain data to obtain the probability of line of sight occlusion. If the line-of-sight occlusion probability is greater than the preset occlusion tolerance threshold, then the cooperative pose data of the cooperating UAV in the monitoring airspace is acquired, and the cooperative pose data is processed by viewpoint cooperation to obtain a global viewpoint distribution matrix. Projection calculations are performed based on the global view distribution matrix to obtain the field of view coverage union and the field of view overlap region; If the overlapping area of ​​the field of view is greater than the preset overlap threshold, then the difference operation is performed on the union of the area to be monitored and the coverage of the field of view to extract the boundary of the monitoring blind zone, and the offset compensation calculation is performed on the boundary of the monitoring blind zone to obtain the angle adjustment command. Based on the cooperative pose data, the angle adjustment command, and the preset anti-collision safety radius, obstacle avoidance trajectory planning is performed to generate cooperative tracking commands.

2. The method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The process of extrapolating the trajectory to obtain the predicted target position includes: The target heading angle and centroid displacement are extracted based on the three-dimensional terrain data and the current attitude data of the moving target. The target's real-time linear velocity is calculated using the centroid displacement and a preset observation time interval. The target's heading angle is differentiated according to the preset observation time interval to obtain the heading angle change rate. The trajectory curvature radius is determined by combining the heading angle change rate with the target's real-time linear velocity. The trajectory curvature radius is fitted using a pre-defined nonholonomic constrained kinematic model, and the predicted target position is generated by extrapolation.

3. The method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The step of constructing the field of view cone range based on the predicted target position, the current spatial position, and the gimbal angle includes: Based on the current spatial location and the predicted target location, a spatial vector is constructed by performing spatial vector analysis; The target observation center axis is determined by combining the line-of-sight space vector with the gimbal angle; Using the target observation center axis as a reference, the spatial extension boundary is calculated to construct the field of view cone range.

4. The method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The step of evaluating the degree of terrain truncation within the field of view cone range based on the three-dimensional terrain data to obtain the probability of line-of-sight occlusion includes: Based on the three-dimensional terrain data, terrain elevation features are extracted, and spatial Boolean intersection is performed between the field of view cone range and the terrain elevation features to identify terrain protrusions. The ratio of the orthogonal projected area of ​​the terrain protrusion on the bottom surface of the field of view cone to the total cross-sectional area of ​​the cone is calculated to obtain the truncation ratio. Extract the current pitch angle contained in the gimbal angle, and combine it with the cosine factor of the current pitch angle to perform weight adjustment calculation on the cutoff ratio, and output the line-of-sight occlusion probability.

5. The method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, If the line-of-sight occlusion probability is greater than a preset occlusion tolerance threshold, then the cooperative pose data of the cooperating UAVs in the monitored airspace is acquired, and the cooperative pose data is subjected to viewpoint cooperative processing to obtain a global viewpoint distribution matrix, including: If the line-of-sight occlusion probability is greater than the preset occlusion tolerance threshold, a cooperative replacement request is generated, and the cooperative pose data of the cooperative UAV in the monitoring airspace is obtained according to the cooperative replacement request. The current planar coordinates, azimuth angle allocation value, and pitch angle offset of the cooperative UAV are extracted based on the cooperative pose data. The terrain undulation gradient is calculated based on the three-dimensional terrain data, and the hovering altitude of the cooperative UAV is divided into steps based on the terrain undulation gradient to obtain altitude layer identifiers. The camera optical parameters of each of the cooperative drones are obtained, and the inter-drone distance between adjacent cooperative drones is calculated by combining the current planar coordinates. The inter-drone distance, the azimuth angle allocation value, the pitch angle offset, the altitude layer identifier and the camera optical parameters are then combined into a multi-dimensional tensor to construct a global view distribution matrix.

6. The method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The projection calculation based on the global view distribution matrix to obtain the field of view coverage union and the overlapping region of the field of view includes: The three-dimensional field-of-view parameters and gimbal optical axis pointing of the collaborative UAV are extracted based on the global view distribution matrix. A dynamic elevation datum is constructed based on the absolute elevation corresponding to the predicted target location. Using the gimbal optical axis of each of the cooperative UAVs as the projection direction, the three-dimensional field of view parameters are intersected with the dynamic elevation reference plane by ray intersection calculation to generate a field of view projection polygon. Boolean union operation is performed on all the field-of-view projection polygons to obtain the field-of-view coverage union, and the intersection region covered by at least two field-of-view projection polygons is extracted and pixel grid integral summation operation is performed to obtain the field-of-view overlap region.

7. The method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The process involves performing a difference operation on the union of the area to be monitored and the field of view coverage to extract the boundary of the monitoring blind zone, and then performing offset compensation calculation on the boundary of the monitoring blind zone to obtain an angle adjustment command, including: The boundary features of the uncovered area are extracted by performing a difference operation on the union of the area to be monitored and the field of view coverage, and are used as the boundary of the monitoring blind zone. Geometric feature extraction is performed on the union of the field of view coverages to obtain the coordinates of the coverage center; The geometric centroid coordinates are obtained by calculating the geometric centroid of the monitoring blind zone boundary, and the centroid displacement is obtained by vector subtraction between the geometric centroid coordinates and the coverage center coordinates. The gimbal yaw rotation parameters and pitch tilt parameters are calculated based on the centroid displacement and the preset proportional-integral-derivative control strategy, and an angle adjustment command is generated based on the gimbal yaw rotation parameters and the pitch tilt parameters.

8. The method for cooperative target tracking and allocation in a swarm of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The step of planning an obstacle avoidance trajectory based on the cooperative pose data, the angle adjustment command, and the preset anti-collision safety radius, and generating a cooperative tracking command, includes: By combining the cooperative pose data with the angle adjustment command, spatial coordinate mapping calculations are performed to generate candidate spatial coordinates for each of the cooperative UAVs; Calculate the predicted trajectory spacing between adjacent cooperative UAVs as they move toward the candidate spatial coordinates; If it is determined that the predicted trajectory spacing is less than the preset anti-collision safety radius, a repulsive force field is generated at the collision boundary, and the body position offset is calculated. The candidate spatial coordinates are corrected by combining the fuselage position offset to generate a cooperative tracking command.

9. A collaborative target tracking and allocation system for unmanned aerial vehicle (UAV) swarms, 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 program, it implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method described in any one of claims 1 to 8.

Images