Unmanned aerial vehicle dynamic trajectory planning method and related device

Abstract

The invention discloses an unmanned aerial vehicle dynamic trajectory planning method and a related device. According to the method, a dynamic target is detected and positioned in real time through a target detection model, and the position of the dynamic target is obtained; based on the current position of the unmanned aerial vehicle and the dynamic target position, generating an initial feasible path from the current position of the unmanned aerial vehicle to the dynamic target position by using a JIGS algorithm; based on the initial feasible path, establishing a safe flight corridor based on ellipsoid matrix constraint through an E-SFC method, and taking the safe flight corridor as a space hard constraint condition; and with minimization of the control quantity and the flight time as optimization targets, solving the dynamic feasible trajectory of the unmanned aerial vehicle under the conditions of introducing boundary condition constraints, dynamic feasibility constraints, visibility constraints, target distance constraints and the space hard constraints by adopting the MINCO class. The safety and stability of the flight task of the unmanned aerial vehicle can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicles, and in particular to a dynamic trajectory planning method for unmanned aerial vehicles and related devices. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, its application in the fields of intelligent monitoring, disaster rescue, agricultural inspection, logistics transportation and the like is continuously expanding. The improvement of the autonomous trajectory planning and task execution capability of unmanned aerial vehicles relies on precise environmental perception, path search algorithms and trajectory optimization of unmanned aerial vehicles. Among them, the trajectory planning of unmanned aerial vehicles, as the core link for realizing autonomous trajectory planning and safe obstacle avoidance of unmanned aerial vehicles, directly determines the safety and intelligent level of unmanned aerial vehicles in task execution. The path of the trajectory planning of unmanned aerial vehicles usually faces a known and fixed target point, and assumes that the positions of the surrounding obstacles remain unchanged. However, in a real complex environment, the task target of the unmanned aerial vehicle is often in a motion state, and the surrounding obstacles may also change dynamically, and only relying on offline path planning cannot meet the dynamic trajectory planning of the unmanned aerial vehicle for dynamic targets. SUMMARY

[0003] In order to solve the above technical problems, the present application proposes a dynamic trajectory planning method for unmanned aerial vehicles and related devices, which can realize autonomous perception, real-time path generation and smooth trajectory tracking of unmanned aerial vehicles for moving targets in a complex three-dimensional dynamic environment, and improve the safety and stability of the flight task of the unmanned aerial vehicle.

[0004] In order to achieve the above purpose, the technical solution of the present application is as follows:

[0005] A dynamic trajectory planning method for unmanned aerial vehicles, comprising the following steps:

[0006] detecting and positioning a dynamic target in real time by a target detection model to obtain a dynamic target position;

[0007] generating an initial feasible path from the current position of the unmanned aerial vehicle to the dynamic target position by using a JIGS algorithm based on the current position of the unmanned aerial vehicle and the dynamic target position;

[0008] establishing a safe flight corridor based on an ellipsoid matrix constraint by using an E-SFC method based on the initial feasible path, as a spatial hard constraint condition;

[0009] solving a dynamic feasible trajectory of the unmanned aerial vehicle under the spatial hard constraint condition by using a MINCO type trajectory optimization with the minimization of control amount and flight time as the optimization target.

[0010] Preferably, detecting and positioning a dynamic target in real time by a target detection model to obtain a dynamic target position comprises the following steps:

[0011] Acquiring a camera image, inputting the camera image into a YOLO11 model, identifying a dynamic target and outputting two-dimensional coordinates of the dynamic target;

[0012] Converting the two-dimensional coordinates of the dynamic target into a three-dimensional position in the camera coordinate system by using the internal parameters of the pre-calibrated depth camera, obtaining the position of the dynamic target, and the formula is as follows:

[0013] , , Z=depth value (20)

[0014] Wherein, is the two-dimensional coordinates of the dynamic target, is the principal point position of the camera, and respectively represent the focal length of the camera in the horizontal direction and the vertical direction, and Z is the distance between the target and the camera.

[0015] 3. The dynamic trajectory planning method of the unmanned aerial vehicle according to claim 1, wherein based on the initial feasible path, a safe flight corridor based on ellipsoid matrix constraint is established by an E-SFC method as a spatial hard constraint condition, comprising the following steps:

[0016] Generating a plurality of ellipsoid regions covering the path segment with each path segment in the initial feasible path as the center;

[0017] Converting the boundary of each ellipsoid region into a convex polyhedron composed of a plurality of hyperplanes, and connecting a plurality of convex polyhedrons in sequence according to the path segment to obtain a safe flight corridor;

[0018] Taking the safe flight corridor as a constraint basis for spatial feasibility in the trajectory optimization process to ensure that the trajectory is always located within the safe flight corridor, and obtaining a spatial hard constraint condition.

[0019] 4. The dynamic trajectory planning method of the unmanned aerial vehicle according to claim 1, wherein a plurality of ellipsoid regions covering the path segment are generated with each path segment in the initial feasible path as the center, comprising the following:

[0020] In a three-dimensional space, each ellipsoid region is expressed in the following QP form:

[0021] (6)

[0022] Wherein, represents the position point of the unmanned aerial vehicle in the three-dimensional space during path planning, is the center point of the ellipsoid, and is usually selected as the center point of each path The geometric center point, The covariance matrix of the ellipsoid is in the following form:

[0023] (7)

[0024] in, Depend on unit eigenvectors The orthogonal matrices formed represent the directions of the major axis, median axis, and minor axis of the ellipsoid, respectively. It is a diagonal matrix.

[0025] 5. A dynamic trajectory planning method for unmanned aerial vehicles according to claim 1, characterized in that the boundary of each ellipsoidal region is transformed into a convex polyhedron composed of multiple hyperplanes, and several convex polyhedra are sequentially connected according to the path segments to obtain a safe flight corridor, including the following:

[0026] For the An ellipsoid can be determined based on the direction of its principal axis. Calculate the boundary points of the ellipsoid. Since the center of the ellipsoid is... Then, the two boundary points along the principal axis of the ellipsoid... It can be represented as:

[0027] (8)

[0028] in, Indicates the ellipsoid at the th Distance from the center point along each principal axis direction The two boundary points are located along the feature vector. Extending in positive and negative directions At a distance,

[0029] The mathematical expression for a hyperplane is:

[0030] (9)

[0031] in, Indicates the direction relative to the principal axis Orthogonal, passing through the boundary points respectively. The two hyperplanes,

[0032] Path segment convex polyhedron It is the intersection of all hyperplanes:

[0033] (10)

[0034] All path segments Corresponding convex polyhedron The sections are pieced together sequentially to form a complete safe flight corridor. , can be represented as:

[0035] (11)

[0036] Where N is the number of path segments.

[0037] 6. A dynamic trajectory planning method for unmanned aerial vehicles according to claim 1, characterized in that the spatial hard constraints include the following:

[0038] Trajectory points at any time All are within the safe flight corridor Within, that is:

[0039] (12)

[0040] in, Indicates the drone at a certain time Spatial location,

[0041] The spatial hard constraint condition is expressed as:

[0042] (13)

[0043] in, It is a path segment The coefficient matrix consists of the normal vectors of each hyperplane in the polyhedron, where each row corresponds to the normal vector of a hyperplane. ,vector It consists of the offset terms of each hyperplane, and its first term is... Each element is calculated as .

[0044] 7. A method for planning the dynamic trajectory of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that solving the dynamic feasible trajectory of the UAV includes the following:

[0045] The objective function for MINCO-type trajectory optimization is defined as follows:

[0046] (14)

[0047] in, The Jerk energy of the drone trajectory over the entire time interval. Total flight time The adjustment parameters are used to balance smoothness and time efficiency.

[0048] Boundary condition constraints for drone trajectory:

[0049] (15)

[0050] in, These respectively represent the drones in and state, and These represent the start and end states of the drone, respectively.

[0051] The dynamic feasibility constraints for drones are defined as follows:

[0052] (16)

[0053] in, Indicates the time of the drone speed, Indicates the time of the drone acceleration, and These are the maximum speed and acceleration allowed for the drone, respectively.

[0054] Introduce visibility constraints :

[0055] (17)

[0056] in, Represents a discrete set of time. This means that the drone's trajectory points must be within the visible area of ​​the dynamic target, thereby preventing the drone from losing sight of the dynamic target points.

[0057] The relative distance between the drone and the dynamic target must meet the target distance constraint:

[0058] (18)

[0059] in, Represents the dynamic target at discrete time. Location, and These represent the minimum and maximum safe distances, respectively.

[0060] Based on the above, the present invention also discloses a dynamic trajectory planning system for unmanned aerial vehicles (UAVs), comprising:

[0061] The dynamic target detection module is used to detect and locate dynamic targets in real time through a target detection model, and obtain the position of the dynamic targets.

[0062] The path generation module is used to generate an initial feasible path from the current position of the UAV to the position of the dynamic target based on the current position of the UAV and the position of the dynamic target using the JIGS algorithm.

[0063] The constraint generation module is used to establish a safe flight corridor based on ellipsoid matrix constraints using the E-SFC method based on the initial feasible path, as a hard constraint condition in space.

[0064] The trajectory optimization module is used to solve the dynamic feasible trajectory of the UAV with the goal of minimizing the control input and flight time. It adopts the MINCO class to solve the dynamic feasible trajectory of the UAV under the conditions of introducing boundary condition constraints, dynamic feasibility constraints, visibility constraints, target distance constraints and the aforementioned spatial hard constraints.

[0065] Based on the above, the present invention also discloses a drone, comprising:

[0066] The fuselage itself;

[0067] A control system that communicates with the fuselage body, the control system including a memory and a processor, the memory for storing computer programs; the processor for executing the computer programs to implement any of the methods described above.

[0068] Based on the above, the present invention also discloses a readable storage medium storing a computer program, which, when executed by a processor, implements any of the methods described above.

[0069] Based on the above technical solution, the beneficial effects of this invention are as follows: This invention provides a dynamic trajectory planning method and related apparatus for unmanned aerial vehicles (UAVs) that integrates YOLO11 model dynamic target detection, JIGS algorithm, E-SFC method, and MINCO-type trajectory optimization. This method achieves real-time closed-loop fusion of perception, path search algorithms, and trajectory optimization at the UAV architecture level, enabling dynamic target detection, path search algorithms, and obstacle avoidance for UAVs in complex dynamic environments. After obtaining the spatial position of the dynamic target in real time using the YOLO11 model, the UAV system uses the JIGS algorithm to generate an initial feasible path from the UAV's current position to the target position. Then, the E-SFC method is used to construct a safe corridor based on ellipsoidal constraints, ensuring that the UAV's trajectory always remains within a collision-free space. Finally, MINCO-type trajectory optimization is used to solve the trajectory under hard constraints, balancing trajectory smoothness, time efficiency, and dynamic feasibility. This method achieves real-time dynamic response from target detection to trajectory generation, effectively addressing issues of target motion, obstacle changes, and environmental uncertainties. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of a dynamic trajectory planning method for a drone in one embodiment;

[0071] Figure 2 This is a diagram of the overall network architecture of YOLO11 in one embodiment;

[0072] Figure 3 This is a schematic diagram of the E-SFC method in a UAV dynamic trajectory planning method in one embodiment;

[0073] Figure 4 This is a comparison of different methods for constructing flight corridors, among which... Figure 4 (a) showcases a traditional safe flight corridor. Figure 4 (b) Demonstrates a segmented method for constructing safe flight corridors based on ellipsoidal function reconstruction;

[0074] Figure 5 This is a schematic diagram of the generation of a hyperplane from an ellipsoid in one embodiment;

[0075] Figure 6 This is a flowchart of the dynamic trajectory planning process for a drone in one embodiment;

[0076] Figure 7 This is an example of a drone hardware system architecture;

[0077] Figure 8 This is a simulation experiment of dynamic target detection by a UAV in one embodiment, wherein, Figure 8 (a) shows that the drone's camera detected a dynamic target in Gazebo. Figure 8 (b) Demonstrates the localization and trajectory display of dynamic targets for UAVs in RViz;

[0078] Figure 9 This is a sample embodiment of the velocity response curves for each of the three axes during the dynamic trajectory planning process of a UAV.

[0079] Figure 10 This is an outdoor experiment on dynamic target detection using a drone in one embodiment, wherein, Figure 10 (a) shows a dynamic target detected by a camera in an outdoor environment by a drone. Figure 10 (b) Demonstrates the visualization of dynamic target localization and trajectory generation in RViz. Figure 10 (c) shows the dynamic target displacement detected by the UAV in an outdoor environment. Figure 10 (d) Shows real-time obstacle avoidance and dynamic trajectory generation of the drone as displayed in RViz;

[0080] Figure 11 This is an example of dynamic trajectory planning for the speed of each axis of a drone. Detailed Implementation

[0081] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0082] like Figure 1 As shown, this embodiment provides a method for dynamic trajectory planning of unmanned aerial vehicles (UAVs), including the following steps:

[0083] Step 100: Detect dynamic targets in real time using a target detection model, and fuse the detection results of the dynamic targets with the data from the depth camera to obtain the location of the dynamic targets.

[0084] In this embodiment, considering that acquiring dynamic target points is a crucial step in achieving dynamic trajectory planning during autonomous UAV flight missions, the YOLO11 model is employed for real-time dynamic target detection. The YOLO11 model architecture consists of three parts: a backbone network, a neck network, and a target detection head. These components work together to achieve efficient and accurate dynamic target detection. The YOLO11 model employs lightweight convolution and efficient feature fusion strategies in its design, thereby improving computational efficiency and optimizing detection performance. Figure 2 The overall architecture of the YOLO11 model is shown.

[0085] The YOLO11 model's backbone is responsible for extracting multi-level features from the input image and transforming them into feature maps suitable for subsequent object detection. Compared to YOLOv8, the YOLO11 model has improved its backbone design, mainly consisting of the Conv module, C3k2 module, SPPF module, and C2PSA module. The structure of the YOLO11 model effectively reduces the computational cost and parameter size of the network while maintaining detection accuracy. The Conv module, composed of convolutional layers, batch normalization (BN) layers, and the SiLU activation function, serves as the most basic feature extraction unit. The C3k2 module further optimizes the C2f structure of YOLOv8, improving feature extraction capability and receptive field through a more efficient convolution stacking method while reducing computational complexity. The Spatial Pyramid Pooling-Fast (SPPF) module expands the receptive field through multi-scale max pooling operations and concatenates multi-scale features before fusion via convolution, thereby enhancing the network's ability to represent targets at different scales. The C2PSA module introduces channel and spatial attention mechanisms, which can enhance feature selectivity in multiple dimensions and further improve the network's adaptability to complex scenarios.

[0086] Following the Backbone, the YOLO11 model's Neck employs an efficient cross-scale feature fusion structure, including upsampling, downsampling, and cross-layer connections, to integrate feature maps of different scales. The main function of the Neck module is to fuse feature maps from different levels of the Backbone to obtain feature representations containing multi-scale information, thereby better detecting targets with large size differences. The YOLO11 model's Neck design optimizes this fusion method while maintaining a lightweight footprint, improving detection performance for both small and large targets. The Head module performs final class prediction and bounding box regression on the multi-scale feature maps from the Neck. Unlike traditional Heads that use anchor boxes, the YOLO11 model's Head uses an anchor-free structure, decoupling the classification and regression branches. This design simplifies the training process and helps improve detection accuracy and the model's generalization ability.

[0087] In summary, the YOLO11 model, through optimizations in modules such as Conv, C3k2, SPPF, and C2PSA, achieves more efficient feature extraction and cross-scale fusion, significantly improving detection accuracy and inference speed, and enhancing its adaptability to complex dynamic scenes and multi-scale targets. Its lightweight design and anchor-free head structure enable the model to maintain real-time performance while possessing stronger robustness and generalization ability. With these advantages, the YOLO11 model provides reliable support for target detection and localization of UAVs in dynamic environments, laying a solid foundation for subsequent dynamic trajectory planning and dynamic target detection.

[0088] Step 200: Based on the current position of the UAV and the position of the dynamic target, generate an initial feasible path from the current position of the UAV to the position of the dynamic target using the JIGS algorithm.

[0089] In this embodiment, to overcome the dense turning points in the graph search path, after obtaining the spatial location of the dynamic target, the UAV system uses the Jump-Interpolated Graph Search (JIGS) algorithm for path searching. This algorithm combines graph search with heuristic optimization mechanisms to quickly generate an initial feasible path from the UAV's current position to the target position in the environmental map. Through a local jump search strategy, the JIGS algorithm can achieve incremental path updates in dynamic environments. Simultaneously, the JIGS algorithm can combine the distribution of environmental obstacles and the target's movement characteristics to make real-time corrections to path nodes, ensuring the feasibility and connectivity of the initial path. Specific details are as follows:

[0090] Quickly generate a sequence of key points from the start to the end point based on jump point rules. This ensures the heuristics and global connectivity of the search:

[0091] (1)

[0092] in, Indicates starting from the origin To the finish line Jump point sequence generated by JPS algorithm This indicates the first in the path One key point, which satisfies and All hops have feasible connectivity, which facilitates local path interpolation optimization in subsequent learning.

[0093] The JPS algorithm boasts strong global connectivity and heuristic search advantages, but its jumping behavior may ignore some local obstacle features, leading to a lack of obstacle avoidance redundancy or failure to meet dynamic constraints. In local regions, for adjacent jump points... The present invention constructs a local search region on its connecting segments. This area is defined by path segments. The search radius is adaptively expanded based on the distribution of surrounding obstacles, with the central axis as the axis. Let the path segment length be: And define the midpoint of the path segment as: .set up For path segment The minimum Euclidean distance to the nearest obstacle. For safety margin, the local search radius is... Defined as:

[0094] (2)

[0095] in, This is the maximum expansion radius. Therefore, the local search region... for:

[0096] (3)

[0097] in, Point The minimum Euclidean distance to the line segment is strictly limited by the A* interpolation search process. It is carried out internally. In each... Within the system, the globally optimal path is obtained using the heuristic A* algorithm. Its objective function is:

[0098] (4)

[0099] in, This represents the Euclidean distance between adjacent path points. This refers to the index variable within the path segment. The final interpolated path is constructed by concatenating the local paths of each segment.

[0100] (5)

[0101] in, This represents the final generated interpolation path, which consists of all locally optimal path segments. It is pieced together. Specifically, these local path segments The splicing method in formula (5) obtained by using the A* algorithm for local interpolation optimization between each pair of adjacent jump points ensures the continuity of the path in the global scope. At the same time, the optimized path has significantly improved in terms of smoothness and dynamic feasibility.

[0102] Step 300: Based on the initial feasible path, establish a safe flight corridor based on ellipsoid matrix constraints using the E-SFC method as a hard constraint condition in space.

[0103] In this embodiment, to further improve the geometric adaptability and safety space compactness of SFC in complex 3D environments, an ellipsoid-reconstructed segmental safe flight corridor (E-SFC) method is proposed. The E-SFC method uses each segment of the initial feasible path generated by the JIGS algorithm. As a reference path, E-SFC constructs a covered path segment. The three-dimensional ellipsoid is adaptively characterized to depict the local free space distribution, thereby achieving a continuous, collision-free, and compact safe flight corridor.

[0104] like Figure 3 This diagram illustrates the compactness construction process among SFC methods. The E-SFC method first uses the path generated by the JIGS algorithm from the starting point. To the finish line Divide the path into several segments in sequence, each segment of the path The corresponding local space is constructed in the form of a three-dimensional ellipsoid, where the yellow ellipsoid represents the reachable region of the local free space. Each ellipsoid is constructed around the center of the path segment. By optimizing its covariance matrix parameters, the principal axis direction and scale of the ellipsoid are adaptively adjusted to ensure it completely encloses the path segment and avoids surrounding obstacles to the greatest extent possible, thereby improving space utilization and reducing redundant envelopes. To use the constructed ellipsoidal region for trajectory optimization modeling, this invention further generates a supporting hyperplane on its boundary based on an ellipsoidal function, and converts each ellipsoid into a convex polyhedron structure composed of a finite number of faces through a supporting hyperplane clipping operation. The polyhedra are sequentially assembled according to the path segment order, ultimately forming a continuous, collision-free SFC, which serves as the spatial constraint domain in the UAV trajectory optimization process. The E-SFC method combines the accuracy of geometric representation with the compactness of the trajectory feasible region, providing higher degrees of freedom and solution space for UAV trajectory optimization. Compared to traditional SFC methods that use fixed volume elements or regular polyhedra, the E-SFC method can more flexibly adapt to non-convex spatial structures and obstacle distribution variations, significantly improving the solvability of UAV trajectories, path safety, and optimization efficiency in complex scenarios.

[0105] To further illustrate the advantages of the proposed E-SFC method in spatial modeling capabilities, Figure 4 The results show a comparison between the traditional SFC method and the E-SFC method under the same obstacle environment. Figure 4 The colored voxels in the diagram represent the spatial distribution of obstacles, the purple lines represent the UAV flight path obtained by the path search algorithm, the light blue area represents the flight corridor composed of polyhedra, and the black borders represent the constraint boundaries used for trajectory optimization. Figure 4 (a) shows the construction effect of the traditional SFC method. The path of the traditional SFC method is generated by the A* algorithm. Because the traditional SFC method shows obvious spatial adaptability when encountering dense obstacle areas, some safe flight corridors of the UAV overlap with the obstacle boundaries, and there is even a risk that the constraints are not feasible. Figure 4 (b) This paper presents the flight corridor construction results after combining the JIGS algorithm proposed in this invention with the E-SFC construction method. The E-SFC method can more accurately capture the shape of the surrounding free space by constructing local adaptive ellipsoids between path segments and generating convex polyhedra by combining hyperplane clipping. The polyhedra form a more compact, continuous and collision-free safe passage. Figure 4As can be seen, the flight corridor generated by the E-SFC method better conforms to the actual environmental boundary, maintaining good connectivity and space utilization even in areas with dense obstacles. Comprehensive comparison shows that the E-SFC method significantly outperforms traditional safe flight corridor generation methods in terms of spatial geometric representation, corridor construction compactness, and feasible region integrity. The E-SFC method effectively improves the modeling accuracy of free space through ellipsoidal adaptive fitting and convex polyhedron trimming. It not only maximizes the utilization of passable space while ensuring path obstacle avoidance safety but also significantly expands the solution space of the trajectory optimizer. In scenarios with dense obstacles or complex environmental topology, the corridor constructed by the E-SFC method exhibits stronger continuity and local adaptability, significantly enhancing the robustness to environmental changes and the accessibility of flight tasks during trajectory optimization, thus providing a more robust structural guarantee for the efficient autonomous flight of UAVs in complex 3D environments. To accurately define and optimize the modeling of each ellipsoidal region in the safe flight corridor of the UAV, this invention uses a quadratic programming (QP) problem to formally describe the ellipsoid. Specifically, for each segment in the path Construct an ellipsoidal region covering the path segment, and use this as the basis for constructing a constrained polyhedron. In three-dimensional space, each ellipsoidal region... It can be represented in the following QP form:

[0106] (6)

[0107] in, This indicates the UAV in three-dimensional space during path planning. The location point in the middle, The center point of the ellipsoid is usually selected for each path segment. The geometric center point, Let be the covariance matrix of the ellipsoid. It can be used to control the shape and ductility of an ellipsoid in different spatial directions. Specifically, The eigenvalues ​​determine the extent of the ellipsoid's extension along each principal axis, while the eigenvectors determine its directionality. To further clarify the ellipsoid's geometry, especially its dimensions and directionality along each principal axis, this invention addresses the covariance matrix of the ellipsoid. Eigenvalue decomposition is performed, in the following form:

[0108] (7)

[0109] in, Depend on unit eigenvectors The orthogonal matrices formed represent the directions of the major axis, median axis, and minor axis of the ellipsoid, respectively. For a diagonal matrix, its diagonal elements The eigenvalues ​​determine the extent of the ellipsoid's extension along each principal axis. During the construction of the safe flight corridor, to transform the ellipsoidal region into a polyhedral constraint usable for path optimization, the ellipsoidal boundary needs to be further fitted with a set of hyperplanes. These hyperplanes, determined by the normal vector and boundary position, can be used to construct polyhedral boundaries, thus forming a convex polyhedral corridor tightly adhering to the ellipsoid's edge. Specifically, for the ... An ellipsoid can be determined based on the direction of its principal axis. Calculate the boundary points of the ellipsoid. Since the center of the ellipsoid is... Then, the two boundary points along the principal axis of the ellipsoid... It can be represented as:

[0110] (8)

[0111] in, Indicates the ellipsoid at the th Distance from the center point along each principal axis direction The two boundary points are located along the feature vector. Extending in positive and negative directions At a distance of [distance]. These boundary points precisely characterize the geometric extent of the ellipsoid along the principal axis, which is crucial for constructing a hyperplane that fits the ellipsoid's boundary. In the actual construction process, points can be generated around each set of boundary points along the principal axis. Orthogonal hyperplanes approximate the curved boundary of the ellipsoid into multiple planar segments. These hyperplanes collectively form a convex polyhedral shell that closely adheres to the ellipsoid's contour, providing computationally achievable linear spatial constraints for the trajectory optimization problem. Specifically, Figure 5 The black squares shown represent the obstacle locations. The ellipsoid is first expanded to the obstacle locations, and then the covariance matrix of the ellipsoid is adjusted. The ellipsoid shrinks into the blue area, in which arrive Location Each principal axis direction For the two sets of hyperplanes, the two sets of hyperplanes are perpendicular to each other. and through the boundary points .

[0112] These two hyperplanes are tangent to the boundary of the ellipsoid in the positive and negative directions of the principal axis, respectively. This hyperplane construction forms the boundary of the ellipsoid in the direction of its principal axis. Figure 4(b) Clearly demonstrates the tangency of the UAV-generated E-SFC in RViz after ellipsoidal reconstruction, highlighting the geometry of the ellipsoid defined by hyperplanes along each principal axis. To describe the mathematical characteristics of these hyperplanes, their mathematical expression is as follows:

[0113] (9)

[0114] in, Indicates the direction relative to the principal axis Orthogonal, passing through the boundary points respectively. The two hyperplanes. This mathematical expression defines a set of half-spaces, which can serve as fundamental constraints in the construction of polyhedra.

[0115] Construct two positive and two negative boundary points along the three principal axes: the major axis, the middle axis, and the minor axis. Based on these boundary points, multiple hyperplanes can be defined, thus forming a hexahedral structure in space, corresponding to the approximate convex hull boundary of an ellipsoid. Finally, the path segment... polyhedron It is the intersection of all hyperplanes:

[0116] (10)

[0117] All path segments Corresponding polyhedron The sections are pieced together sequentially to form a complete safe flight corridor. , can be represented as:

[0118] (11)

[0119] Through ellipsoidal matrix reconstruction and polyhedral approximation, a complete safe flight corridor was constructed in the aforementioned section. This constraint is used to limit the feasible range of the path in space. Each path segment is encapsulated within a clearly defined geometric constraint region, providing the necessary safe space constraints for the trajectory optimization problem and ensuring the safety and feasibility of the generated path. The polyhedrons corresponding to all path segments... A safe flight corridor pieced together This defines the spatial feasibility constraints for UAVs during mission execution, ensuring that the UAV's trajectory lies within a passable area. To satisfy the constraints of formula (6) in trajectory optimization, it is necessary to guarantee that the trajectory points at any given time... All are within the safe flight corridor Within, that is:

[0120] (12)

[0121] in, Indicates the drone at a certain time Based on the spatial location, formula (7) indicates that the UAV trajectory needs to remain within a safe flight corridor. Within the defined safe area.

[0122] The trajectory constraints of the UAV will be transformed into linear inequality constraints to facilitate efficient solution in the UAV trajectory optimization problem. Specifically, for each path segment... Corresponding polyhedron The set of hyperplanes formed by these hyperplanes can be transformed into a set of linear inequalities to constrain the spatial extent of the trajectory within this segment, ensuring that the path does not exceed the boundary of the safe flight corridor. Therefore, the spatial constraints in the trajectory optimization problem can be expressed as:

[0123] (13)

[0124] in, It is a path segment The coefficient matrix consists of the normal vectors of each hyperplane in the polyhedron, where each row corresponds to the normal vector of a hyperplane. .vector It consists of the offset terms of each hyperplane, and its first term is... Each element is calculated as ,and Let be any point on the hyperplane.

[0125] In summary, by constructing a safe flight corridor for UAVs, this invention introduces a well-defined and analytical spatial constraint structure into the UAV trajectory optimization problem, strictly limiting the feasible region of the UAV trajectory. To facilitate efficient solution during the optimization process, each spatial region constructed from an ellipsoid within the safe flight corridor is ultimately transformed into a set of linear inequalities, thereby constraining the UAV trajectory optimization problem within a safe corridor composed of multiple convex polyhedra. The safe flight corridor not only simplifies the complexity of constraint modeling but also enhances the adaptability and robustness of the UAV trajectory generation algorithm to spatial boundaries.

[0126] Step 400: With minimizing control input and flight time as the optimization objective, the MINCO class is used to solve the dynamic feasible trajectory of the UAV under the conditions of introducing boundary condition constraints, dynamic feasibility constraints, visibility constraints, target distance constraints, and the aforementioned spatial hard constraints.

[0127] In this embodiment, under the hard spatial constraints provided by the E-SFC method, the MINCO class is used to generate the UAV flight trajectory. This trajectory aims to minimize control input and flight time, balancing trajectory smoothness and time efficiency. During optimization, boundary condition constraints, dynamic feasibility constraints, hard spatial constraints, visibility constraints, and target distance constraints are introduced simultaneously to ensure trajectory continuity at start and end, velocity and acceleration satisfying dynamic feasibility, trajectory points within the E-SFC safe space, and maintaining effective observation distance to dynamic targets throughout the UAV's flight. The objective function of the UAV trajectory optimization problem can be defined as:

[0128] (14)

[0129] in, The Jerk energy of the drone trajectory over the entire time interval. This represents the total flight time. The adjustment parameters are used to balance smoothness and time efficiency. This objective function... This ensures the smoothness of the drone trajectory while also considering time optimization. Secondly, the boundary condition constraints of the drone trajectory must be satisfied:

[0130] (15)

[0131] in, These respectively represent the drones in and state, and These represent the initial and final states of the UAV, including information such as position, velocity, and acceleration. By constraining these boundary conditions, the consistency between the trajectory and mission requirements can be ensured, enabling the UAV to take off smoothly from the designated starting point and ultimately reach the target location in the desired state. The dynamic feasibility constraint of the UAV is defined as follows:

[0132] (16)

[0133] in, Indicates the time of the drone speed, Indicates the time of the drone acceleration, and These are the maximum speed and acceleration allowed for the UAV, respectively. The constraints of formula (11) ensure that the UAV's motion state always meets the dynamic feasibility requirements during the entire UAV trajectory execution process, thereby avoiding uncontrollable or unstable flight caused by overspeed or overacceleration.

[0134] Regarding path safety, this step uses the E-SFC hard constraint method as shown in formula (7), which requires that the UAV's trajectory is within the safe flight corridor at any time. This ensures the collision-free nature of the UAV trajectory. Similar to Minimum Jerk modeling, spatial constraints are also used to ensure the feasibility of the trajectory. However, the difference is that Minimum Jerk is only used as a trajectory smoothness modeling method, while the core of the MINCO class lies in the objective function of formula (9), which is to optimize both trajectory smoothness and time efficiency within the framework of minimum control. Based on this, formula (7) formed by E-SFC is combined to ensure that the UAV trajectory satisfies the smoothness objective while further guaranteeing spatial safety. To maintain the observation of dynamic targets by the UAV during flight, visibility constraints need to be introduced. :

[0135] (17)

[0136] in, Represents a discrete set of time. This means that the drone's trajectory points must be within the visible area of ​​the dynamic target to prevent the drone from losing sight of the target. Simultaneously, the relative distance between the drone and the dynamic target must meet the following constraints:

[0137] (18)

[0138] in, Represents the dynamic target at discrete time. Location, and These represent the minimum and maximum safe distances, respectively. Formula (13) ensures that the UAV maintains an appropriate relative distance from dynamic targets during flight, avoiding the risk of collision when too close or the loss of observation and tracking effectiveness when too far.

[0139] In summary, the MINCO class, through its optimization framework of minimum control variables, organically combines trajectory smoothness, time efficiency, dynamic feasibility, and spatial safety. It balances Jerk energy and flight time in the objective function, resulting in a smooth and efficient trajectory. The constraints not only satisfy the start and end boundaries and dynamic requirements such as velocity and acceleration, but also introduce hard spatial constraints provided by the E-SFC method to ensure collision-free trajectory execution. Furthermore, it incorporates visibility and target distance maintenance constraints to achieve stable tracking of dynamic targets. Compared to the Minimum Jerk method, the MINCO class not only maintains the continuity and analytical nature of trajectory optimization but also exhibits stronger real-time performance and robustness in complex dynamic environments, providing solid theoretical and methodological support for autonomous dynamic trajectory planning for UAVs.

[0140] In the dynamic trajectory planning of UAVs, tracking dynamic target points is a crucial task. To achieve real-time tracking of dynamic target points, UAVs typically need to acquire the position coordinates of the dynamic target point in the image and the depth information of the dynamic target point. This invention combines YOLO11 target detection and depth camera data, enabling the UAV to effectively calculate the distance between the dynamic target point and the UAV, and maintain the relative position between the dynamic target point and the UAV in real time through formula (13). The physical position information between the UAV and the dynamic target point is published in real time through depth camera detection. The UAV system first detects the target position in the camera through YOLO11 and generates a bounding box. This bounding box contains the two-dimensional coordinates of the target position, where the coordinates of the four vertices of the bounding box are respectively represented as: , and YOLO11 further calculates the coordinates of the target's center point. The calculation formula is as follows:

[0141] , (19)

[0142] Formula (14) can be used to effectively calculate the two-dimensional position of the dynamic target in the camera image. The two-dimensional position of the dynamic target provides the basis for calculating the distance between the target and the UAV.

[0143] After acquiring the two-dimensional image coordinates of the dynamic target, the UAV system uses the depth information provided by the depth camera, taking the distance between the pixel and the camera as the Z-coordinate input. Combining the X and Y coordinates from the image with the camera's intrinsic parameters, the target's pixel coordinates can be converted into a three-dimensional position in the camera coordinate system. Thus, the UAV can calculate the target's absolute position in three-dimensional space in real time and further obtain the relative positional relationship between the target and itself. The target's three-dimensional position in the camera coordinate system can be calculated using the following formula:

[0144] , Z = depth value (20)

[0145] in, It is the principal point position of the camera, that is, the principal point coordinates in the camera's intrinsic parameters, which is usually the center of the image. and , representing the camera's focal length in the horizontal and vertical directions, respectively. Z is the target depth value obtained from the depth image, representing the distance between the target and the camera. Using formula (15), the three-dimensional position of the target in the camera coordinate system can be accurately calculated, thus providing support for the dynamic trajectory planning of the UAV.

[0146] Figure 6 This document presents a complete flowchart of a drone's dynamic trajectory planning using YOLO11. The drone process comprises multiple steps, from drone detection of dynamic targets to dynamic trajectory planning, each closely linked to ensure the drone can effectively maintain a dynamic trajectory and avoid obstacles. First, the drone acquires environmental information via its camera, providing the spatial relationship between the drone and its environment. By obtaining the coordinates of the surrounding environment, the drone system can determine the relative positions of dynamic targets and the distribution of obstacles. Then, the YOLO11 detection module performs real-time analysis of the image, detecting and generating bounding boxes within the dynamic targets, such as... Figure 6 As shown, after detecting a dynamic target, the system returns the target's category and confidence level. Next, the UAV system combines the detection results with environmental information to generate dynamic data on the target's movement. During this process, the YOLO11 model updates the relative position between the UAV and the dynamic target in real time and calculates the target's direction of movement, providing a basis for the UAV's dynamic trajectory planning.

[0147] experiment

[0148] 1. Platform Settings

[0149] The simulation experiments were conducted using an Ubuntu 20.04 system environment, building a simulation platform integrating ROS Noetic and Gazebo9. The host machine used was equipped with an Intel Core i7-14650HX processor with a clock speed of 2.20 GHz, 16 GB of memory, and an NVIDIA GeForce RTX 4060 dedicated graphics card. All simulation tasks were executed in single-threaded mode on this platform.

[0150] In terms of practical verification, an autonomous UAV platform integrating perception, control, and computing was built, such as... Figure 7As shown in the diagram, the UAV hardware architecture incorporates a perception module using an Intel RealSense D455f depth camera, capable of real-time acquisition and processing of environmental depth information for high-precision obstacle recognition and map construction. The UAV's flight controller system utilizes an NxtPX4V2 controller with Stable Release v1.16.0 firmware, ensuring stable and reliable flight attitude control. The computing module employs an Allspark 2-X86 onboard computing platform, featuring an Intel Core i7-1165G7 processor with a clock speed of up to 4.7 GHz and 16 GB of LPDDR4x memory. This allows for real-time execution of YOLO11 object detection, JIGS path planning, E-SFC methods, and MINCO-type trajectory optimization under Linux. For communication, a Wi-Fi module enables low-latency information exchange between the UAV and the ground station.

[0151] The entire UAV system is based on Ubuntu 20.04 and ROS Noetic. All the core functions of the UAV run independently on the onboard computing platform without relying on external servers or remote computing support, which fully verifies the deployment capability and engineering application value of the method proposed in this invention in a real system.

[0152] 2. Simulation Experiment of Dynamic Trajectory Planning for Unmanned Aerial Vehicles

[0153] To verify the effectiveness of the YOLO11-based dynamic target detection module in dynamic trajectory planning and dynamic target tracking of UAVs, this invention conducted dynamic target detection and UAV localization experiments in the Gazebo9 simulation environment and the RViz visualization platform. Static obstacles and dynamically moving targets were set up in the experimental scenario to simulate the perception and localization process of a UAV performing dynamic trajectory planning in a real-world mission.

[0154] like Figure 8 As shown, Figure 8 (a) This is a first-person view captured by the drone's onboard camera in the Gazebo environment. The YOLO11 model can accurately identify and select dynamic targets in front of the drone in the self-built scene. The YOLO11 detection module outputs the target's position in 3D space in real time and performs differential calculations with the drone's current position to obtain the relative position between the drone and the dynamic target, providing a basis for maintaining a safe distance between the drone and the dynamic target. The red dashed boxes in the image mark the location of obstacles. During the drone's dynamic trajectory planning process, the drone can actively avoid obstacles by combining with the path planning module, ensuring the safety and continuity of the drone's flight mission.

[0155] Figure 8(b) Demonstrates the visualization results of dynamic target localization and trajectory generation in the RViz platform within this constructed scenario. The E-SFC method and MINCO class take the target spatial coordinates output by YOLO11 as input, and combine the UAV's current position with the environmental map to generate a safe and feasible dynamic trajectory from the current position to the dynamic target position. Figure 8 In (b), the green dots represent the current position of the UAV, the blue circles represent the detected target positions, the red area is the 3D obstacle occupancy grid, and the black ellipsoid is the safe flight corridor constructed by the E-SFC method. Figure 8 (b) It can be seen that the UAV system can quickly complete the construction of a safe flight corridor and the generation of MINCO-type trajectories while sensing the position of dynamic targets in real time, thus realizing a closed-loop processing flow from dynamic target detection to dynamic trajectory planning. Figure 9 The three-axis velocity variation curves of the UAV during dynamic target tracking are presented. Statistical results show that the UAV's longitudinal velocity... The maximum value was 1.61 m / s, and the average value was 1.13 m / s. This was the largest amplitude during the entire drone tracking process, indicating that the drone adjusted its speed most frequently when longitudinally following a dynamic target. (Drone lateral velocity) The maximum value is 0.69 m / s, and the average value is 0.062 m / s, exhibiting moderate fluctuations. This reflects the drone's fine-tuning process in the lateral position to maintain horizontal alignment with the target. Vertical velocity The maximum value is 0.29 m / s, and the average value is only 0.0019 m / s. The changes are stable and the amplitude is small, indicating that the UAV makes few altitude adjustments during flight, which helps maintain a stable flight attitude. Overall, the dynamic trajectory generated by combining the E-SFC method and MINCO class with YOLO11 detection results ensures the obstacle avoidance safety of the UAV while enabling the UAV to maintain stable speed changes during dynamic trajectory planning, avoiding trajectory jitter caused by frequent acceleration and deceleration, thereby improving the stability of the UAV's dynamic trajectory planning. Experimental results show that the dynamic trajectory planning based on trajectory optimization using YOLO11, E-SFC method, and MINCO class maintains stable detection accuracy and trajectory feasibility even in complex environments with static obstacles and dynamic targets.

[0156] 3. Dynamic trajectory planning experiment of UAVs

[0157] This invention presents a dynamic trajectory planning experiment for a drone in a forest environment. During the experiment, the drone uses the YOLO11 model to detect dynamic targets in real time and combines it with a MINCO-type trajectory optimization method to perform dynamic trajectory planning and dynamic target tracking in the environment. Figure 10 The experiment demonstrates the process of dynamic target detection and trajectory planning in an outdoor environment.

[0158] Figure 10 In (a), the UAV detected the dynamic target using the D455f depth camera and optimized the path using the JIGS algorithm and the E-SFC method, successfully bypassing the obstacle. Simultaneously, the UAV trajectory was further optimized using MINCO-like methods, ensuring the smoothness and dynamic feasibility of the trajectory. Figure 10 (b) Demonstrates the visualization of real-time dynamic target localization and UAV dynamic trajectory planning in RViz, clearly showing the relative position of the dynamic target and the UAV, as well as the dynamic generation of the trajectory. Figure 10 (c) demonstrates how the UAV ensures the safety of its flight trajectory and achieves real-time obstacle avoidance by using E-SFC hard constraints when performing dynamic target tracking.

[0159] To ensure the validity and stability of the experimental results, multiple experiments were conducted. The experimental data show that by combining YOLO11 target detection, the JIGS algorithm, and the MINCO-type trajectory optimization method, the UAV's trajectory generation in outdoor environments exhibits strong adaptability and flexibility, enabling it to respond to changes in dynamic targets in real time and maintain dynamic trajectory planning and obstacle avoidance capabilities. Figure 11 This experiment showcases the velocity response data of a UAV along the X, Y, and Z axes during dynamic trajectory planning. In this experiment, the UAV's velocity changes significantly with the movement of the dynamic target, particularly in the Y and Z directions. These fluctuations demonstrate that the UAV adjusts its flight speed based on the dynamic position of the target. Figure 11 As can be seen, when the UAV begins dynamic trajectory planning, its X-axis velocity increases sharply, reaching a maximum of 1.6 m / s, and remains constant during the subsequent stable flight phase, reflecting the UAV's rapid response to dynamic targets and adjustment of its trajectory. Meanwhile, the Y-axis velocity exhibits smaller fluctuations, with a maximum velocity of 0.3 m / s, indicating that the UAV can adapt to changes in the position of dynamic targets and make necessary adjustments. In the Z-axis direction, the velocity change is relatively small, with a maximum velocity of approximately 0.2 m / s, showing that the UAV primarily makes dynamic adjustments in the horizontal plane, with less longitudinal variation.

[0160] Experimental results show that the UAV dynamic trajectory planning method proposed in this invention, which combines YOLO11 target detection, the JIGS algorithm, the E-SFC method, and MINCO-type trajectory optimization technology, exhibits robustness and adaptability in complex environments. By detecting dynamic targets in real time, optimizing the path with hard spatial constraints, and combining MINCO-type trajectory optimization, the UAV can respond to changes in dynamic targets in real time, smoothly adjust its own trajectory, and effectively avoid obstacles, demonstrating efficient and stable dynamic trajectory planning capabilities. Specifically, the YOLO11 model can detect and locate dynamic targets with high accuracy, the JIGS algorithm effectively realizes path search, and the E-SFC method provides hard constraints for path optimization, ensuring the safety of the UAV's flight trajectory. Based on this, the MINCO-type method further optimizes the UAV's trajectory, ensuring trajectory smoothness, time efficiency, and dynamic feasibility.

[0161] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0162] Based on the same inventive concept, this application also provides a system for implementing the above-described method for dynamic trajectory planning of unmanned aerial vehicles. The solution provided by this system is similar to the solution described in the above method, and therefore will not be repeated here.

[0163] In one embodiment, a method system for planning the dynamic trajectory of an unmanned aerial vehicle (UAV) is also provided, comprising:

[0164] The dynamic target detection module is used to detect and locate dynamic targets in real time through a target detection model, and obtain the position of the dynamic targets.

[0165] The path generation module is used to generate an initial feasible path from the current position of the UAV to the position of the dynamic target based on the current position of the UAV and the position of the dynamic target using the JIGS algorithm.

[0166] The constraint generation module is used to establish a safe flight corridor based on ellipsoid matrix constraints using the E-SFC method based on the initial feasible path, as a hard constraint condition in space.

[0167] The trajectory optimization module is used to solve the dynamic feasible trajectory of the UAV with the goal of minimizing the control input and flight time. It adopts the MINCO class to solve the dynamic feasible trajectory of the UAV under the conditions of introducing boundary condition constraints, dynamic feasibility constraints, visibility constraints, target distance constraints and the aforementioned spatial hard constraints.

[0168] In the above embodiments, each module of the UAV dynamic trajectory planning system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0169] In one embodiment, a drone is also provided, comprising: a fuselage; and a control system communicating with the fuselage, the control system including a memory and a processor, the memory for storing a computer program; and the processor for executing the computer program to implement the steps as described in all the above method embodiments.

[0170] In one embodiment, a readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps as described in all the above method embodiments.

[0171] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0172] The above are merely preferred embodiments of the present application and are not intended to limit the embodiments of the present application. For those skilled in the art, the embodiments of the present application can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of the present application should be included within the protection scope of the embodiments of the present application.

Claims

1. A method for dynamic trajectory planning of unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: The target detection model is used to detect and locate dynamic targets in real time, thereby obtaining the position of the dynamic targets. Based on the current position of the UAV and the position of the dynamic target, the JIGS algorithm is used to generate an initial feasible path from the current position of the UAV to the position of the dynamic target. Based on the initial feasible path, a safe flight corridor based on ellipsoid matrix constraints is established using the E-SFC method as a hard constraint condition in space. With the goal of minimizing control input and flight time, the MINCO class is used to solve the dynamic feasible trajectory of the UAV under the conditions of introducing boundary condition constraints, dynamic feasibility constraints, visibility constraints, target distance constraints, and the aforementioned spatial hard constraints.

2. The method for dynamic trajectory planning of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The method of using a target detection model to detect and locate dynamic targets in real time, thereby obtaining the position of the dynamic targets, includes the following steps: Acquire camera images, input the camera images into the YOLO11 model, identify dynamic targets, and output the two-dimensional coordinates of the dynamic targets; Using the pre-calibrated intrinsic parameters of the depth camera, the two-dimensional coordinates of the dynamic target are converted into its three-dimensional position in the camera coordinate system, thus obtaining the position of the dynamic target. The formula is as follows: , , Z=depth value (20) in, It is the two-dimensional coordinates of a dynamic target. This is the principal position of the camera. and These represent the camera's focal length in the horizontal and vertical directions, respectively, and Z is the distance between the target and the camera.

3. The method for dynamic trajectory planning of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, Based on the initial feasible path, a safe flight corridor constrained by an ellipsoid matrix is ​​established using the E-SFC method as a hard spatial constraint, including the following steps: Using each path segment in the initial feasible path as the center, generate several ellipsoidal regions covering the path segments; The boundary of each ellipsoidal region is transformed into a convex polyhedron composed of multiple hyperplanes. Several convex polyhedra are connected in sequence according to the path segments to obtain a safe flight corridor. During trajectory optimization, the safe flight corridor is used as the basis for spatial feasibility constraints to ensure that the trajectory always lies within the safe flight corridor, thus obtaining hard spatial constraints.

4. The method for dynamic trajectory planning of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, Centered on each path segment in the initial feasible path, several ellipsoidal regions covering the path segments are generated, including the following: In three-dimensional space, each ellipsoidal region This can be represented in the following QP form: (6) in, This indicates the UAV in three-dimensional space during path planning. The location point in the middle, The center point of the ellipsoid is usually selected for each path segment. The geometric center point, The covariance matrix of the ellipsoid is in the following form: (7) in, Depend on unit eigenvectors The orthogonal matrices formed represent the directions of the major axis, median axis, and minor axis of the ellipsoid, respectively. It is a diagonal matrix.

5. The method for dynamic trajectory planning of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The boundary of each ellipsoidal region is transformed into a convex polyhedron composed of multiple hyperplanes. Several convex polyhedra are then connected in sequence according to the path segments to obtain a safe flight corridor, including the following: For the An ellipsoid can be determined based on the direction of its principal axis. Calculate the boundary points of the ellipsoid. Since the center of the ellipsoid is... Then, the two boundary points along the principal axis of the ellipsoid... It can be represented as: (8) in, Indicates the ellipsoid at the th Distance from the center point along each principal axis direction The two boundary points are located along the feature vector. Extending in positive and negative directions At a distance, The mathematical expression for a hyperplane is: (9) in, Indicates the direction relative to the principal axis Orthogonal, passing through the boundary points respectively. The two hyperplanes, Path segment convex polyhedron It is the intersection of all hyperplanes: (10) All path segments Corresponding convex polyhedron The sections are pieced together sequentially to form a complete safe flight corridor. , can be represented as: (11) Where N is the number of path segments.

6. The method for dynamic trajectory planning of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The spatial hard constraints include the following: Trajectory points at any time All are within the safe flight corridor Within, that is: (12) in, Indicates the drone at a certain time Spatial location, Spatial hard constraints are expressed as: (13) in, It is a path segment The coefficient matrix consists of the normal vectors of each hyperplane in the polyhedron, where each row corresponds to the normal vector of a hyperplane. ,vector It consists of the offset terms of each hyperplane, and its first term is... Each element is calculated as .

7. The method for dynamic trajectory planning of an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The process of solving for the dynamic feasible trajectory of the UAV includes the following: The objective function for MINCO-type trajectory optimization is defined as follows: (14) in, The Jerk energy of the drone trajectory over the entire time interval. Total flight time The adjustment parameters are used to balance smoothness and time efficiency. Boundary condition constraints for drone trajectory: (15) in, These respectively represent the drones in and state, and These represent the start and end states of the drone, respectively. The dynamic feasibility constraints for drones are defined as follows: (16) in, Indicates the time of the drone speed, Indicates the time of the drone acceleration, and These are the maximum speed and acceleration allowed for the drone, respectively. Introduce visibility constraints : (17) in, Represents a discrete set of time. This means that the drone's trajectory points must be within the visible area of ​​the dynamic target, thereby preventing the drone from losing sight of the dynamic target points. The relative distance between the drone and the dynamic target must meet the target distance constraint: (18) in, Represents the dynamic target at discrete time. Location, and These represent the minimum and maximum safe distances, respectively.

8. A dynamic trajectory planning system for unmanned aerial vehicles (UAVs), characterized in that, include: The dynamic target detection module is used to detect and locate dynamic targets in real time through a target detection model, and obtain the position of the dynamic targets. The path generation module is used to generate an initial feasible path from the current position of the UAV to the position of the dynamic target based on the current position of the UAV and the position of the dynamic target using the JIGS algorithm. The constraint generation module is used to establish a safe flight corridor based on ellipsoid matrix constraints using the E-SFC method based on the initial feasible path, as a hard constraint condition in space. The trajectory optimization module is used to solve the dynamic feasible trajectory of the UAV with the goal of minimizing the control input and flight time. It adopts the MINCO class to solve the dynamic feasible trajectory of the UAV under the conditions of introducing boundary condition constraints, dynamic feasibility constraints, visibility constraints, target distance constraints and the aforementioned spatial hard constraints.

9. A drone, characterized in that, include: The fuselage itself; A control system that communicates with the fuselage body, the control system including a memory and a processor, the memory for storing computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images