Real-time flight path planning and intelligent obstacle avoidance method and system for autonomous flight of unmanned aerial vehicle

Abstract

The invention discloses a real-time flight path planning and intelligent obstacle avoidance method and system for autonomous flight of an unmanned aerial vehicle. The method relates to the technical field of unmanned aerial vehicle obstacle avoidance, and comprises the following steps: environment perception and dynamic prediction, data reduction and macroscopic analysis, spatial modeling and pattern library construction, and hierarchical path search and fusion. According to the method, the environment is sensed in real time, the motion mode of a dynamic obstacle is analyzed, sensing data are sorted, key parts are intercepted and clustered, so that a connected region is recognized, a spatial data model and a topological relation are established, an environment mode library is constructed, path search is converted into mode matching, and an initial path is formed through preferential search based on the mode library; according to the method, the real-time flight path planning and intelligent obstacle avoidance safety and efficiency of the unmanned aerial vehicle are improved, and the problem that the real-time flight path planning and intelligent obstacle avoidance safety and efficiency of the unmanned aerial vehicle under the complex road condition are low in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of obstacle avoidance technology for unmanned aerial vehicles (UAVs), and in particular to a method and system for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of UAVs. Background Technology

[0002] First, before flight, the drone loads a pre-planned global reference track and establishes an initial environmental model based on a high-precision digital map, such as a point cloud map built using SLAM (Simultaneous Localization and Mapping) technology. After takeoff, real-time environmental data is collected through a multi-sensor fusion perception system. LiDAR provides accurate 3D point cloud data, visual cameras (usually combining monocular or binocular vision) perform dynamic target recognition and semantic segmentation, and millimeter-wave radar effectively detects obstacles at medium and long distances. The collected data is fused and processed by an embedded processor, and the drone simultaneously achieves self-localization and environmental modeling using real-time SLAM algorithms or visual inertial odometry, constructing an incremental local occupancy grid map that includes static obstacles and dynamic objects.

[0003] Subsequently, based on the real-time environment model, the planning system initiates an online replanning loop. When the perception module detects obstacles or threats (such as suddenly appearing birds, buildings, or other drones) on the preset track, an obstacle avoidance response is triggered. A hierarchical planning strategy is adopted. At the global level, an improved A algorithm or a fast random exploration tree is used to quickly generate a coarsely feasible path around static obstacles in the updated map. At the local level, optimization-based or reactive methods are used for fine-grained obstacle avoidance and trajectory generation. Meanwhile, a lightweight reactive obstacle avoidance algorithm runs in parallel for suddenly appearing close-range dynamic obstacles.

[0004] Finally, the generated smooth and safe local trajectory is analyzed by the underlying flight control system and converted into motor execution commands through a cascaded PID (Proportional-Integral-Derivative) controller or model predictive control, driving the UAV to accurately track the new trajectory.

[0005] For example, the invention patent with publication number CN118034352A discloses an obstacle avoidance method, obstacle avoidance device, and drone, which includes: using multiple sensors to monitor the surrounding environment of the drone in real time to obtain multiple sets of obstacle information, each set of obstacle information corresponding to one obstacle, and each set of obstacle information including obstacle type, distance between the drone and the obstacle, and position of the obstacle; determining the threat probability of the corresponding obstacle based on the multiple sets of obstacle information; obtaining an obstacle avoidance strategy based on the multiple sets of obstacle information and the threat probability corresponding to the obstacle; and controlling the drone to adjust its flight state to avoid the obstacle and return to the original auxiliary flight path according to the obstacle avoidance strategy, the original auxiliary flight path being the inspection route initially set for the drone.

[0006] For example, the invention patent announcement CN108062109B discloses a drone obstacle avoidance method, which includes: S1: installing a laser and a satellite positioning receiver on the drone; S2: during the drone's flight, collecting obstacle information around the drone using the laser; S3: importing a 3D map library into the drone's information processing module, performing coordinate transformation on the collected obstacle information, and searching for a 3D map of the target area in the 3D map library; S4: replanning the flight route based on the 3D map of the target area to achieve autonomous obstacle avoidance for the drone.

[0007] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems: In existing technologies, for drones used in practical applications such as logistics and delivery, firstly, existing algorithms (such as regional fast optimization methods or sampling optimization strategies like CRS-RRT) are mostly limited to path search and optimization between local two points. Although they improve local planning efficiency through constraint sampling and region extraction, they lack the ability to coordinate planning for multiple global task points (such as multiple delivery points or inspection points). In actual tasks, when delivery points are added or changed, or when charging piles are needed for range management, existing methods cannot perform dynamic and efficient global replanning. This makes it difficult for drones to optimize task sequences within their limited range, easily leading to safety issues or mission interruptions due to insufficient battery power. Essentially, they still rely on the intervention of flight control personnel, failing to achieve truly fully autonomous operation and one-person multi-drone management, thus limiting operational efficiency and large-scale application.

[0008] Secondly, the computational efficiency of existing algorithms is insufficient, with planning time typically in the order of seconds. However, the speed of drones and dynamic obstacles (such as birds and other aircraft) can reach tens of meters per second or even higher. The second-level delay cannot meet the safety requirements of real-time obstacle avoidance and poses a risk of collision due to response lag.

[0009] Finally, existing technical solutions rely on continuous and complex calculations of perceived data to generate paths, resulting in a large computational load. This makes it difficult to achieve millisecond-level real-time decision-making responses on ordinary airborne computing chips. Furthermore, the lack of an efficient data processing mechanism that balances the integrity of the overall task with the flexibility of local obstacle avoidance leads to low safety and efficiency in real-time trajectory planning and intelligent obstacle avoidance for UAVs in complex road conditions. Summary of the Invention

[0010] To address the low safety and efficiency issues in existing technologies for real-time trajectory planning and intelligent obstacle avoidance of UAVs in complex road conditions, this invention provides a method and system for real-time trajectory planning and intelligent obstacle avoidance of UAVs in autonomous flight. The technical solution is as follows: On the one hand, a method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of unmanned aerial vehicles (UAVs) is provided. This method includes: S1, acquiring real-time three-dimensional environmental perception data, including the coordinates and dimensions of static obstacles and the position and speed of dynamic obstacles, through a perception sensor mounted on the UAV; setting a planning time interval based on the movement speed of the dynamic obstacles, and performing preliminary statistical analysis on the perception data within a preset continuous time segment to determine the regularity of the dynamic obstacle movement pattern; S2, sorting the three-dimensional environmental perception data, extracting the top-ranked key data and performing cluster analysis to identify connected regions in the environment, thereby reducing the spatial scale and data volume of subsequent calculations in the early stages of global planning; S... 3. Establish a spatial geographic data model and corresponding topological relationship model based on a dual independent data structure; decompose environmental geometric features into a finite number of standard patterns, construct a pattern library, and transform the path search problem into a pattern recognition and matching problem to avoid large-scale repetitive computation; S4. Starting from the edge region of the constructed spatial data model, perform pattern recognition and matching, prioritize searching and connecting to form initial feasible paths. For areas that cannot be connected, combine the environmental data that are ranked lower, and perform supplementary searches based on the established data structure and topological relationship to calculate the minimum cost connection point with the already connected path, and add it to the global path to obtain feasible paths composed of discrete line segments.

[0011] On the other hand, a real-time trajectory planning and intelligent obstacle avoidance system for autonomous UAV flight is provided. This system applies a method for real-time trajectory planning and intelligent obstacle avoidance for autonomous UAV flight. The system includes: an environmental perception and dynamic prediction module, a data reduction and macroscopic analysis module, a spatial modeling and pattern library construction module, and a hierarchical path search and fusion module. The environmental perception and dynamic prediction module is used to acquire real-time 3D environmental perception data, including the coordinates and dimensions of static obstacles and the position and speed of dynamic obstacles, through perception sensors mounted on the UAV. It sets a planning time interval based on the movement speed of the dynamic obstacles and performs preliminary statistical analysis on the perception data within a preset continuous time segment to determine the regularity of the dynamic obstacle movement pattern. The data reduction and macroscopic analysis module is used to sort the 3D environmental perception data, extract the top-ranked key data, and perform cluster analysis. To identify connected regions in the environment, thereby reducing the spatial scale and data volume of subsequent calculations in the early stages of global planning; the spatial modeling and pattern library construction module is used to establish a spatial geographic data model and corresponding topological relationship model based on a dual independent data structure; it decomposes environmental geometric features into a finite number of standard patterns, constructs a pattern library, and transforms the path search problem into a pattern recognition and matching problem to avoid large-scale repetitive calculations; the hierarchical path search and fusion module is used to start from the edge region of the constructed spatial data model, perform pattern recognition and matching, prioritize searching and connecting to form initial feasible paths, and for areas that cannot be connected, combine the environmental data with lower rankings, perform supplementary searches based on the established data structure and topological relationships, calculate the minimum cost connection point with the already connected path, and add it to the global path to obtain feasible paths composed of discrete line segments.

[0012] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. Through the environmental perception and dynamic prediction module, the system acquires and analyzes the motion state and patterns of dynamic obstacles in real time, adaptively adjusting the time interval of trajectory planning to enable the system to respond to environmental changes within milliseconds. Combined with the data reduction and macro-analysis module, a ranking and dynamic interception strategy based on distance or threat level is adopted to quickly filter out key obstacle information. A macro-environment connectivity model is constructed through cluster analysis, reducing the amount of original large-scale environmental data processing. The spatial modeling and pattern library construction module further decomposes the environmental structure into a limited number of standard patterns, transforming the traditional geometric search-based path planning problem into an efficient pattern recognition and matching problem. In typical scenarios, the local path calculation time can be reduced from seconds to milliseconds, enabling UAVs to achieve real-time obstacle avoidance and path replanning even during high-speed flight (such as in urban logistics delivery scenarios). This solves the problem of insufficient dynamic obstacle avoidance capabilities and safety hazards caused by computational delays in existing technologies, further improving the safety and efficiency of real-time trajectory planning and intelligent obstacle avoidance for autonomous UAV flight.

[0013] 2. By employing a hierarchical path search and fusion module, the system first rapidly constructs an initial feasible path skeleton at the global level based on key datasets, identifying major connecting corridors and congested areas. Building upon this, for unconnected areas not covered by the skeleton, the system only calls the later-ranked detailed environmental data for localized refined searching. Based on a dual-independent data structure and a multi-factor cost model (comprehensively considering path length, smoothness, and safety), the optimal connection scheme is selected. This avoids the resource waste of uniform computation on all data in traditional methods. While ensuring network connectivity, the overall computational complexity is reduced, making it particularly suitable for multi-pickup and delivery point planning in logistics, multi-point inspection tasks in urban patrols, and drone swarm collaborative operations. This allows the system to maintain the independence, safety, and overall efficiency of each drone's path even when one person controls multiple drones, further improving the safety and efficiency of real-time trajectory planning and intelligent obstacle avoidance for autonomous drone flight.

[0014] 3. By establishing a dual-independent data structure, the environment is uniformly represented as a node-arc-polygon topological network. This not only supports efficient geometric relationship querying and dynamic updates but also provides a structured foundation for patterned path recognition. The pattern library construction module extracts typical scene features based on historical data to form reusable standard path patterns. This allows the system to directly call pre-stored solutions when facing repetitive or similar environments, eliminating the need for recalculation. Simultaneously, the system possesses a self-learning and expansion mechanism for the pattern library. When encountering a new environment that fails to match, it automatically employs traditional search algorithms for calculation and stores the results as new samples in the pattern library, enabling continuous system evolution. This allows the system to accumulate experience as tasks are executed, gradually improving the planning success rate and efficiency in various complex environments (such as dense buildings, dynamic traffic, and severe weather). This provides a solid technical foundation for the large-scale and long-term application of UAVs in the low-altitude economy, further enhancing the safety and efficiency of UAV autonomous flight real-time trajectory planning and intelligent obstacle avoidance. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of an unmanned aerial vehicle (UAV) provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the hierarchical path construction core of a real-time trajectory planning and intelligent obstacle avoidance method for autonomous flight of unmanned aerial vehicles provided in this embodiment of the invention. Figure 3 This is a flowchart of the repair and fusion sub-path of the skeleton path to the complete path of a real-time trajectory planning and intelligent obstacle avoidance method for autonomous flight of unmanned aerial vehicles provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the structure of a real-time trajectory planning and intelligent obstacle avoidance system for autonomous flight of unmanned aerial vehicles provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] like Figure 1 The diagram shows a flowchart of a real-time trajectory planning and intelligent obstacle avoidance method for autonomous flight of an unmanned aerial vehicle (UAV) according to an embodiment of this application. The method includes the following steps: S1, acquiring real-time three-dimensional environmental perception data, including the coordinates and dimensions of static obstacles and the position and speed of dynamic obstacles, using a perception sensor mounted on the UAV; setting a planning time interval based on the movement speed of the dynamic obstacles, and performing preliminary statistical analysis on the perception data within a preset continuous time segment to determine the regularity of the dynamic obstacle movement pattern; S2, sorting the three-dimensional environmental perception data, extracting the top-ranked key data and performing cluster analysis to identify connected regions in the environment, thereby reducing the spatial scale of subsequent calculations in the early stages of global planning. With data volume; S3, establish a spatial geographic data model and corresponding topological relationship model based on a dual independent data structure; decompose environmental geometric features into a finite number of standard patterns, construct a pattern library, and transform the path search problem into a pattern recognition and matching problem to avoid large-scale repetitive computation; S4, starting from the edge region of the constructed spatial data model, perform pattern recognition and matching, prioritize searching and connecting to form initial feasible paths, for areas that cannot be connected, combine the environmental data with lower rankings, perform supplementary searches based on the established data structure and topological relationship, calculate the minimum cost connection point with the already connected path, and add it to the global path to obtain feasible paths composed of discrete line segments.

[0023] It should be understood that, firstly, in the environmental perception and dynamic prediction stage of step S1, the UAV uses sensors such as vision and lidar to collect real-time three-dimensional environmental data, including the coordinates and dimensions of static obstacles and the position and speed of dynamic obstacles. The planning time interval is adaptively set according to the movement speed of the dynamic obstacles, and the perception data within continuous time periods is statistically analyzed to identify their movement patterns (regular or random). Through real-time behavior modeling and motion prediction of dynamic obstacles, the system can identify threats in advance and adjust the planning rhythm, achieving millisecond-level response in high-speed dynamic environments, fundamentally improving the timeliness of obstacle avoidance and flight safety.

[0024] Next, in step S2, the data reduction and macroscopic analysis stage, the acquired 3D environmental perception data is sorted according to preset rules (such as the distance of obstacles from the drone, their size, or threat level), and the top-ranked key data is dynamically extracted (the proportion can be adjusted between 10% and 30% depending on the environmental complexity) to form a key dataset. Subsequently, cluster analysis is performed on this dataset to identify spatially clustered obstacle groups and abstract them into composite obstacle regions. Then, the spatial relationships between regions are analyzed to preliminarily determine the "connecting corridors" and "blocking areas" in the environment. Through intelligent data filtering and macroscopic clustering, the system compresses environmental information in the early planning stage, significantly reducing the data scale and spatial complexity of subsequent calculations, while quickly constructing the topological skeleton of the environment, laying a structural foundation for efficient global planning.

[0025] Then, in step S3, spatial modeling and pattern library construction, a dual-independent data structure is used to encode the space, representing the environment as a topological network composed of nodes, arcs, and polygons, and automatically maintaining its connections and adjacency relationships. Furthermore, based on historical planning data, geometric and topological features of typical local environments are extracted and categorized into a limited number of standard patterns (such as "corridor passage," "obstacle detour," and "narrow passage crossing"), constructing a queryable pattern library. Through topological modeling and pattern representation, the system partially transforms the traditional geometry-based path planning problem into an efficient pattern recognition and matching problem. When facing repetitive or similar scenarios, pre-stored optimal path solutions can be directly invoked, avoiding a large amount of repetitive computation and reducing local planning time from seconds to milliseconds, greatly improving the algorithm's real-time performance.

[0026] Finally, in step S4, the hierarchical path search and fusion stage, the system starts from the edge regions of the spatial data model (such as the mission start point, target point, and flyable airspace boundaries), and uses a pattern library for rapid matching and connection to prioritize the formation of a connected initial feasible path skeleton. For unconnected areas not covered by the skeleton, the system performs a local refined search by combining detailed environmental data ranked lower, calculates the minimum cost connection point with the skeleton path based on the established topological relationships, and selects the optimal branch for fusion through a multi-factor cost model (combining path length, smoothness, and safety). Through the hierarchical strategy of "global skeleton first, then local patching," the system ensures the connectivity and safety of the entire path network while avoiding the uniform calculation of all data in traditional methods, improving overall computational efficiency by approximately 50%. It is particularly suitable for real-time trajectory planning and dynamic replanning in multi-task-point and complex urban environments, providing reliable technical support for large-scale, high-efficiency autonomous operation in scenarios such as UAV logistics delivery and urban patrol.

[0027] Furthermore, the specific steps for setting the planning time interval based on the movement speed of the dynamic obstacle are as follows: the setting of the planning time interval is inversely proportional to the movement speed of the dynamic obstacle; the perception data of the dynamic obstacle within a preset continuous time segment is calculated, the perception data including the mean displacement, the variance of velocity, and the rate of change of the direction angle; the mean displacement, the variance of velocity, and the rate of change of the direction angle are compared with their preset thresholds; if the variance of velocity is lower than the first velocity threshold and the rate of change of direction angle is lower than the second rate of change threshold, then the movement mode of the dynamic obstacle is determined to be regular movement; otherwise, the movement mode of the dynamic obstacle is determined to be random movement.

[0028] It should be understood that, as Figure 2 The diagram shows the core flowchart of a hierarchical path construction method for real-time trajectory planning and intelligent obstacle avoidance of unmanned aerial vehicles (UAVs) in accordance with an embodiment of the present invention. The specific process is as follows: Starting with "perception data input," static and dynamic obstacle information is acquired. Next, the "data reduction and macroscopic analysis" stage is performed, including distance-based sorting and dynamic extraction, clustering of key data, identification of complex obstacle regions, and determination of connected corridors and blocked areas in the environment. Following this, "spatial modeling and pattern library construction" is carried out by establishing a DIME data structure, constructing a topological relationship network, and extracting and defining standard patterns to build a pattern library. Then, the "hierarchical path search and fusion" stage begins, starting the search from edge nodes. By matching local topology with the pattern library, it determines whether to use pre-stored rules for rapid expansion or to enable traditional search algorithms, forming a global initial path skeleton. Then, unconnected areas are identified, supplemented with detailed data, and optimal fusion is achieved through multi-objective cost evaluation. Finally, a globally discrete feasible path is output. This process embodies a systematic path planning approach of "perception-modeling-hierarchical planning-fusion output."

[0029] It is important to understand that, in order to effectively cope with the uncertainty brought about by dynamic obstacles during the real-time trajectory planning process of UAVs, this invention proposes an adaptive time interval setting mechanism. The core of this mechanism is to intelligently adjust the planning frequency according to the motion state of the dynamic obstacles in order to achieve the optimal balance between computational efficiency and safety.

[0030] First, the system acquires real-time perception data of dynamic obstacles over a continuous time period, including their mean displacement, velocity variance, and rate of change of the direction angle. Then, by analyzing the velocity variance and the rate of change of the direction angle, the system quantifies the predictability of the dynamic obstacle's motion: if the velocity variance is below a preset first velocity threshold and the rate of change of the direction angle is below a second rate of change threshold, the obstacle's motion pattern is determined to be regular (e.g., uniform straight-line flight, stable hovering), indicating that its trajectory is highly predictable in the short term. Conversely, if either indicator exceeds the threshold, it is determined to be random (e.g., irregular flight of birds, maneuvering to avoid obstacles by other drones), indicating that its behavior is difficult to predict accurately. Based on this determination, the system dynamically sets the planning time interval: for obstacles with regular motion, due to the high predictability of their trajectory, the system can use a longer time interval for planning, thereby reducing the computational frequency and saving computing resources; for obstacles with random motion, to respond promptly to unpredictable motion changes, the system will use a shorter time interval to achieve high-frequency replanning to ensure the real-time performance and reliability of obstacle avoidance.

[0031] In this embodiment, by introducing an adaptive time control strategy based on motion pattern recognition, the system no longer employs a fixed-frequency planning approach but flexibly adjusts the calculation rhythm according to the dynamic characteristics of the environment. In scenarios where obstacle movement is regular and the environment is relatively stable, the system can reduce the planning frequency, significantly saving onboard computing resources and extending the effective control cycle. In highly dynamic and highly random environments, the system can shorten the planning interval to the millisecond level, enabling rapid response to sudden threats and avoiding collision risks caused by planning delays. This mechanism not only enhances the UAV's autonomous obstacle avoidance capabilities in complex airspaces but also provides an intelligent decision-making basis that balances efficiency and safety when facing mixed dynamic obstacles (such as other UAVs, birds, and moving vehicles) in tasks such as logistics delivery and urban patrol.

[0032] Furthermore, the specific steps for sorting the 3D environment perception data are as follows: The acquired 3D environment perception data is sorted according to preset rules to obtain an ordered dataset; for each obstacle data point, one or more feature attributes for sorting are assigned. The sorting process is performed according to one or a combination of the following two modes: The 3D Euclidean distance between each obstacle data point and the current position of the UAV is calculated; all obstacle data points are sorted in ascending order based on the 3D Euclidean distance, so that the obstacle point closest to the UAV is placed at the beginning of the sequence; the number of key data points is calculated based on the total number of actual obstacle data points, and the number of key data points is selected from the ordered dataset to form a key dataset.

[0033] It is important to understand the obstacle data points: Current location of the drone: ; Three-dimensional Euclidean distance: ; N represents the number of key data points, M represents the total number of actual obstacle data points, and K represents the preset interception ratio.

[0034] In the initial stage of trajectory planning, to address the computational burden caused by the massive amount of perception data in the 3D environment, this invention employs an efficient data sorting and filtering mechanism. This mechanism aims to quickly extract key obstacle data from global information, significantly reducing the data scale for subsequent processing and laying the foundation for real-time planning. The specific implementation steps are as follows: The system first performs structured preprocessing on the raw 3D environmental perception data acquired through sensors, including the coordinates, dimensions, and velocity information of all static and dynamic obstacles. Each obstacle data point is assigned one or more feature attributes for sorting. In practice, the system prioritizes calculating the 3D Euclidean distance between each obstacle data point and the UAV's current location, and then sorts all obstacle points in ascending order based on this distance value, thereby constructing an ordered dataset centered on the UAV and arranged from near to far. This sorting strategy ensures that nearby obstacles posing the most direct and urgent threat to UAV flight safety are identified and processed first. Based on this, the system further performs intelligent data interception: according to the total number of actual obstacle data points, combined with the preset interception ratio (usually dynamically adjusted between 10% and 30% depending on the complexity of the environment), the number of key data points that need to be retained is calculated, and then the number of key data points is selected from the front end of the sorted ordered dataset to form a key dataset for subsequent macro-environment modeling.

[0035] In this embodiment, by combining a strategy of "distance-based priority ranking" with "proportional dynamic truncation," the system can reduce the amount of raw environmental data processed in the early stages of planning, thereby significantly reducing the computational complexity of subsequent clustering analysis, topology modeling, and path search. This not only greatly improves the overall efficiency of trajectory planning, enabling the system to quickly construct the environmental skeleton within milliseconds, but also ensures that the priority obstacle set covers the nearest threats that have the greatest impact on the current flight, without sacrificing critical safety information while reducing data. This method is particularly suitable for dense environments or large-scale UAV swarm scheduling scenarios, providing efficient and reliable data preprocessing support for real-time, multi-task parallel trajectory planning.

[0036] Further, the specific steps for identifying connected regions in the environment are as follows: Cluster analysis is performed on the key dataset to identify spatially clustered obstacle groups, and each identified cluster is considered a composite obstacle region; the spatial relationships between composite obstacle regions are analyzed to preliminarily determine connected corridors and blocked regions in the environment. Specifically, based on the geometric boundaries of each composite obstacle region, an environmental topology map is constructed, abstracting each composite obstacle region as a node in the environmental topology map, and calculating the minimum Euclidean distance between the boundaries of any two composite obstacle regions; if the minimum distance between two composite obstacle regions is greater than a preset width threshold required for safe passage of the UAV, then a connected corridor is determined to exist between these two regions. The geometric features of the connected corridor are defined by the line segment connecting the nearest points of the two composite obstacle regions and its vertical extension surface; if multiple composite obstacle regions are spatially closely adjacent and together enclose a closed region with a maximum inscribed circle diameter smaller than the preset width threshold required for safe passage of the UAV, then the closed region is determined to be a blocked region.

[0037] It should be explained that after completing the screening and sorting of key obstacle data, this invention enters the crucial stage of structured environmental understanding: identifying connected regions in the environment. This step aims to aggregate discrete obstacle information into a macroscopic spatial structure representation, providing high-level semantic information for global path planning. The specific implementation steps are as follows: The system first inputs the aforementioned key dataset into a clustering algorithm (such as DBSCAN or K-Means). By analyzing the distribution density or geometric proximity of obstacles in three-dimensional space, it identifies spatially clustered obstacle groups and uniformly treats each identified cluster as a composite obstacle region. This operation aggregates a large number of discrete obstacle point clouds into a few macroscopic obstacle bodies with clear boundaries, achieving the first abstraction and simplification of the environment.

[0038] Based on this, the spatial relationships between these complex obstacle regions are further analyzed to determine the passable and impassable areas in the entire environment. Specifically, the system constructs an environmental topology map based on the geometric boundaries of each complex obstacle region, where each complex obstacle region is abstracted as a node in the graph. Subsequently, the system calculates the minimum Euclidean distance between the boundaries of any two complex obstacle regions. If this distance is greater than a preset width threshold required for safe passage of the UAV (this threshold can be dynamically set according to the UAV size, maneuverability, and safety margin), it is determined that there is a connecting corridor between the two regions. The geometric characteristics of this corridor are defined by the line segment connecting the nearest points of the two regions and its vertical extension surface, representing a passage that the UAV can safely traverse. Conversely, if multiple complex obstacle regions are spatially close to each other and they together enclose a closed area, and the maximum inscribed circle diameter of this area is less than the safe passage width threshold for the UAV, then this area is determined to be a blocked area, indicating that the space is narrow and the UAV cannot pass safely.

[0039] In this embodiment, by combining cluster analysis and topology modeling, the system can quickly extract high-level environmental structures with navigational significance from massive, low-level obstacle point cloud data: passable "corridors" and impassable "blockage areas." This progressive understanding from "points" to "regions" and then to "structures" not only reduces the complexity of environmental information but also provides clear guidance for subsequent path search: the planner can prioritize building a global path skeleton in the identified connected corridors while actively avoiding blockage areas, thereby avoiding local deadlocks or meaningless searches. This greatly improves the directionality, efficiency, and success rate of global path planning, especially in complex scenarios such as urban canyons with towering buildings and industrial parks with dense obstacles, ensuring that the UAV can quickly find safe and continuous flightable channels, laying a solid environmental cognitive foundation for real-time autonomous navigation.

[0040] It should be understood that, as Figure 3The diagram shows a sub-flowchart of the skeleton path to complete path repair and fusion process in a real-time trajectory planning and intelligent obstacle avoidance method for autonomous UAV flight provided by an embodiment of the present invention. The specific process is as follows: Starting with the "initial feasible path skeleton", the system first identifies those "unconnected regions" that are surrounded by obstacles and not covered by the skeleton; then, for each unconnected region, a refined search is enabled and combined with the environmental detail data ranked later, to systematically calculate all possible "reachable connection paths and their corresponding costs" from the region to the skeleton path; next, the system enters the "multi-objective cost evaluation" stage, comprehensively considering multiple factors such as "path geometric length", "path smoothness index" and "minimum safe distance from obstacles", to calculate the weighted total cost of each path; by comparison, the system "selects the path with the minimum total cost as the optimal connection scheme" and inserts it as a new arc segment into the global path network, while updating the relevant topology; after repairing a region, the system determines "are there still unconnected regions?", if so, the above process is repeated, if not, the final "global discrete feasible path is generated". This process embodies a refined fusion mechanism that achieves global path integrity through locally optimal connections while ensuring path security and smoothness.

[0041] Furthermore, the specific steps for establishing a spatial geographic data model based on a dual independent data structure are as follows: A node represents an independent geographic location, and its data unit includes at least a node ID and three-dimensional coordinates; an arc segment represents a directed line segment defined by a start node and an end node, and its data unit includes at least an arc segment ID, a start node ID, an end node ID, an arc segment length, and the IDs of the polygons on the left and right sides of the arc segment; a polygon represents a closed area enclosed by a series of connected arc segments, and its data unit includes at least a polygon ID and a sequence of arc segment IDs that make up the polygon; the polygon is used to identify obstacle areas or flyable airspace; the topological relationships between nodes, arc segments, and polygons are automatically established and maintained through the left and right polygon IDs and start and end node IDs recorded in the arc segment data unit.

[0042] It should be noted that, in order to efficiently represent complex 3D environments and support rapid topological relationship queries and path reasoning, this invention constructs a spatial geographic data model based on a dual-independent data structure. This model achieves unified management and efficient computation of environmental geometry and topological information through hierarchical encoding of nodes, arcs, and polygons. The specific modeling steps are as follows: First, the system defines a node as the most basic geographic entity, used to represent an independent 3D spatial location point. Its data unit contains at least a unique node ID and precise 3D coordinates (x, y, z). Based on this, an arc is defined as a spatial line segment with direction, uniquely determined by a start node and an end node. Its data unit, in addition to the arc ID, also records the start and end node IDs, the arc's geometric length, and the IDs of the polygons on the left and right sides of the arc. It explicitly encodes the topological relationship between the arc and its adjacent spatial regions. Furthermore, a polygon is defined to represent a closed region enclosed by a series of connected arcs. Its data unit contains at least a polygon ID and an ordered sequence of arc IDs that make up the polygon. Semantically, polygons are used to identify different attribute regions in the environment, such as obstacle regions, flyable airspace, or specific mission regions. By recording the left and right polygon IDs and start and end node IDs in the arc segment data unit, the system can automatically establish and dynamically maintain a complete topological relationship network among nodes, arc segments, and polygons, including the connection relationship between nodes and arc segments (i.e., which arc segments a node connects to), the adjacency relationship between arc segments and polygons (i.e., which two polygons are to the left and right of each arc segment), and the adjacency relationship between polygons (i.e., two polygons sharing the same arc segment are neighbors).

[0043] In this embodiment, the dual-independent data structure upgrades the traditional environment representation based on raw point clouds or lattices to a highly structured and topological vector model. This representation method not only has high storage efficiency, but more importantly, it enables the system to query the connections, adjacencies, and containment relationships between any spatial entities with constant or logarithmic time complexity. For example, when performing path search, the system can directly determine whether an arc segment belongs to an obstacle boundary or a passable passage based on the left and right polygon information recorded, without performing complex geometric intersection calculations. When the environment changes dynamically (such as obstacle movement), only local updates to the attributes and connections of relevant nodes, arc segments, and polygons are needed to achieve rapid incremental updates of the model. This efficient and flexible data expression and maintenance mechanism provides a solid data foundation for subsequent millisecond-level pattern recognition, real-time path search, and dynamic replanning, and is a key support for the present invention to achieve efficient autonomous navigation in complex environments.

[0044] Furthermore, the topological relationships include: the connection relationship between nodes and arcs (i.e., which arcs intersect at a certain node), the adjacency relationship between arcs and polygons (i.e., which two polygons are to the left and right of each arc), and the adjacency relationship between polygons (i.e., two polygons sharing the same arc are adjacent polygons). The specific steps for decomposing environmental geometric features into a finite number of standard patterns and constructing a pattern library are as follows: Based on historical flight path planning data, local environmental geometric features are extracted. These local geometric features are composed of a predefined number of nodes, arcs, and polygons combined according to the topological relationships defined by the spatial geographic data model, and are used to characterize a typical path planning scenario. The extracted geometric features are classified and defined as a finite number of standard patterns. Each standard pattern includes at least a pattern ID, a topological structure description, path generation rules, and a pattern library establishment. The pattern ID is used to uniquely identify the pattern. The topological structure description is used to describe the connection and adjacency relationships of nodes, arcs, and polygons under the pattern. The path generation rules are used to predefine one or more optimal feasible paths for the pattern. The pattern library establishment is used to store all predefined standard patterns and their associated information as a queryable pattern library.

[0045] It is important to understand that after constructing a spatial geographic data model based on a dual-independent data structure, this invention further utilizes its inherent topological relationship network to push environmental understanding and path planning to a new stage of intelligence and efficiency. Its core is to decompose complex environmental geometric features into a finite number of standard patterns and construct a reusable pattern library. The specific steps are as follows: The system first extracts representative local environmental geometric features from data accumulated from a large number of historical flight path planning tasks. These features are not raw coordinate point clouds, but rather structured fragments composed of nodes, arcs, and polygons as basic elements, strictly following the topological relationships defined in the aforementioned data model (including the connection relationship between nodes and arcs, the adjacency relationship between arcs and polygons, and the adjacency relationship between polygons). Each fragment represents a typical scenario that repeatedly appears in path planning, such as a "T-junction," a "narrow passage entrance," an "obstacle-surrounded area," or an "open space crossing area."

[0046] Next, the system categorizes and abstracts these extracted topological fragments, defining them as a finite set of standard patterns. Each standard pattern is a well-encapsulated knowledge unit, whose data structure includes at least: a pattern ID (for unique identification), a topological description (precisely describing the specific connections and adjacencies of nodes, arcs, and polygons within the pattern using a formal language or data structure), and, most importantly, path generation rules (predefined optimal feasible paths or their generation algorithms for a specific topology). Finally, the system persistently stores all defined standard patterns and their associated information, constructing a pattern knowledge base that can be quickly indexed and queried.

[0047] In this embodiment, by elevating environmental cognition from the "geometric calculation" level to the "topological pattern recognition" level, the path planning problem of the system in complex environments is partially transformed into an efficient pattern matching and rule invocation process. When the UAV enters a new local environment, the system can quickly match the node-arc-polygon topology structure it perceives and models with standard patterns in the pattern library in real time. Once a match is successful, the system can directly invoke the pre-stored, optimized path generation rules of that pattern to instantly obtain a locally optimal path, completely avoiding the large amount of geometric search, cost evaluation, and iterative optimization calculations required by traditional methods in that local environment. This allows the system to reduce the local planning time from tens to hundreds of milliseconds to milliseconds or even sub-milliseconds when dealing with repetitive or structurally similar scenarios, improving the overall planning efficiency by more than an order of magnitude. At the same time, the establishment of the pattern library endows the system with the ability to continuously learn and evolve, providing an accumulative and generalizable intelligent navigation core for the long-term, large-scale autonomous operation of UAVs.

[0048] Furthermore, the specific steps for transforming the path search problem into a pattern recognition and matching problem are as follows: The node-arc-polygon topology of the current local environment is matched with standard patterns in the pattern library. If a match is successful, the pre-stored path generation rules in the standard pattern are directly invoked to obtain feasible paths for the local area, thus avoiding real-time, large-scale geometric search and calculation for the local environment. If a match is unsuccessful, the path search algorithm is activated for calculation, and the features of the new environment and the finally calculated optimal path are used as new samples, which are then learned and added to the pattern library to achieve self-expansion of the pattern library.

[0049] It is important to understand that, in the real-time execution phase of path planning, the core innovation of this invention lies in intelligently transforming the traditional path planning problem, which relies entirely on online geometric search, into a hybrid solution process of "pattern recognition and matching as the primary method, and geometric search as a secondary method," thereby achieving a qualitative leap in planning speed. The specific operational steps are as follows: When the UAV flies to a certain local environment, the system first uses real-time perception data and the aforementioned dual-independent data structure to quickly construct the node-arc-polygon topology of the current local area. Subsequently, the system uses this real-time generated local topology as a query request and quickly matches it with all pre-stored standard patterns in the offline pattern library. The matching process not only compares geometric shapes but, more importantly, compares the isomorphism of topological relationships, such as the number of node connections and the polygon attributes of adjacent arcs.

[0050] The matching result triggers two efficient decision paths: If the match is successful, meaning the current local environment's topology is highly consistent with a standard pattern in the pattern library (such as "Right-Angle Corridor" or "Circular Obstacle Array"), the system immediately suspends any online geometric search algorithm and directly calls the pre-stored optimal path generation rule for that pattern. This rule may be a parameterized curve equation or a sequence of arc segments connecting key nodes, allowing the system to instantly synthesize a feasible or even optimal path for the local area. If the match fails, indicating the encounter of a new pattern or an exceptionally complex structure, the system seamlessly switches to using traditional path search algorithms (such as A and RRT) for online computation. After this computation, the system not only uses the obtained path for current navigation, but more importantly, it automatically treats the topological features of the new environment and its calculated optimal path as a new sample. After necessary generalization and abstraction, it learns and stores this sample in the pattern library as a new standard pattern or a supplement to existing patterns.

[0051] In this embodiment, the real-time computational load of the planner is significantly optimized by introducing a "memory-prediction" mechanism. In a large number of typical scenarios with repetitive structural features (such as urban grid streets, standardized warehouses, and common terrains), the system can achieve "instant lookup and use" through pattern matching, reducing the local path solving time from tens of milliseconds or more to almost negligible microseconds. This improves the overall planning frequency and response speed by an order of magnitude, truly meeting the stringent requirements of high-speed UAVs for millisecond-level real-time obstacle avoidance. Simultaneously, the self-expanding capability of the pattern library endows the system with powerful lifelong learning and adaptability. As the number of tasks executed increases, its pattern library becomes increasingly rich, the proportion of scenarios it can directly handle continues to rise, and planning efficiency continuously improves, forming a virtuous cycle of becoming smarter with use. This not only reduces the dependence on the continuous high computing power of onboard chips but also provides a sustainably evolving intelligent core for the long-term autonomous and reliable operation of UAVs in unknown or dynamically changing environments.

[0052] Furthermore, the specific steps for forming a connected initial feasible path are as follows: the mission starting point, the mission target point, and the nodes located on the boundary of the flyable airspace in the spatial data model are defined as edge nodes, and a path search is initiated from the edge nodes; starting from each edge node, the arc segments and adjacent polygons connected to it are queried, and the current local topology is matched in real time with the pattern library constructed in step S3; if a standard pattern is matched, the optimal connection rules or path segments pre-stored in that pattern are used first to expand into the interior of the space and quickly connect to the next key node; repeat the above matching process, prioritizing the connection of path branches extended from different edge nodes, until one or more initial feasible paths consisting of key nodes and arc segments are formed that run through the main flyable area.

[0053] It is important to understand that after completing environmental topology modeling and pattern library construction, this invention enters the rapid generation stage of the global path skeleton. Its goal is to efficiently "grow" initial feasible paths connecting major regions from the environmental boundaries. First, the task starting point, task target point, and nodes located on the boundaries of flyable airspace polygons, identified based on a dual-independent data structure, are uniformly defined as edge nodes. These edge nodes constitute the anchor points for path search. Subsequently, the path search process, with these edge nodes as roots, synchronously advances into the environment: starting from each edge node, the system instantly queries all connected arcs and adjacent polygon information of that node in the topology model, thus forming a snapshot of the local topology centered on that node. The system then uses this local topology as input and performs real-time matching with the pattern library constructed in step S3.

[0054] If a standard pattern is successfully matched (e.g., "entrance passage," "boundary turning point," etc.), the system will immediately prioritize the pre-stored, optimized optimal connection rules or preset path segments for that pattern. This is typically a sequence of arc segments or motion parameters pointing to the next critical node within the space. Applying this rule, the system can quickly and accurately extend the current path branch to the next critical node without computation. If no pattern is matched, the system uses a lightweight heuristic search in that local area to determine the extension direction. This process is performed in parallel or iteratively on all active edge nodes. The system will prioritize attempting to connect path branches growing from different edge nodes (especially the starting and target points), and once the critical nodes of two branches are topologically connected, they are immediately joined together.

[0055] In this embodiment, by employing a "growing from the edge inwards" search strategy combined with a "pattern-first" approach, the system can efficiently construct an initial feasible path skeleton that traverses the main flyable areas of the environment. This method abandons the inefficient approach often used by traditional global planners, which involves blindly searching from the starting point to the end point or uniform sampling. Instead, it utilizes the topological boundary information of the environment and patterned prior knowledge for directional intelligent expansion. Practice shows that this strategy can shorten the generation time of the global skeleton path in complex urban environments, and the generated skeleton path naturally conforms to the macroscopic structure of the flyable airspace, avoiding paths getting stuck in local dead zones or generating a large number of ineffective detours. This high-quality initial skeleton provides a clear and reliable backbone for subsequent local fine-tuning (handling disconnected areas), greatly reducing the complexity and uncertainty of overall path planning. It is a core technological guarantee for ensuring that UAVs can quickly find a safe and coherent global passage in a vast and complex three-dimensional space.

[0056] Further, the specific steps for obtaining a feasible path composed of discrete line segments are as follows: S100, identifying regions not covered by the initial feasible path skeleton and surrounded by obstacle polygons as disconnected regions; S101, combining the environmental data processed later in step S2, within the disconnected regions, based on the established dual-independent data structure, calculating all reachable nodes from the region to the initial feasible path and their corresponding costs; S102, evaluating the costs of all calculated reachable node paths, where the path cost is a multi-objective weighted function, and its factors include at least: path geometric length, path smoothness index, and [other factors]. S103: Minimum safe distance of obstacle boundary; By comparing the weighted total cost of all reachable node paths, select the path with the minimum cost as the optimal connection scheme, and determine its connection nodes with the skeleton path; Insert the optimal path as a new arc segment into the global path network, and update the topological relationship between relevant nodes and polygons to achieve structural integration with the initial skeleton path; Repeat steps S100-S103 until all unconnected areas are processed to generate a globally discrete feasible path that integrates the initial skeleton path and all supplementary branch paths, wherein the globally discrete feasible path consists of an ordered sequence of nodes and their connecting arc segments.

[0057] It should be understood that after obtaining the initial feasible path skeleton traversing the main areas of the environment, this invention enters the final refinement stage of path planning: generating globally complete discrete feasible paths. The core task of this stage is to identify and repair "island" areas not covered by the skeleton, ensuring the global connectivity of the navigation network. The specific steps are as follows: First, the system analyzes the constructed initial feasible path skeleton and the polygonal regions it traverses, precisely defining those spatial regions not covered by the skeleton and completely surrounded by polygons identified as obstacles as disconnected regions. These regions are usually not traversed by the skeleton path due to their small size, remote location, or being merged in macro-clustering. Next, for each disconnected region, the system does not reprocess all environmental data, but intelligently recalls detailed environmental data that was not selected into the key dataset in the sorting process in step S2. Using this local data containing fine-grained obstacle information, the system, based on the established dual-independent data structure, employs an efficient graph search algorithm within the disconnected region to systematically calculate every possible path from any starting point within the region to all reachable nodes on the initial skeleton path, and simultaneously calculates the corresponding cost of each path.

[0058] The cost evaluation of a path is not based solely on length, but rather on a comprehensive quantification using a multi-objective weighted cost function. This function integrates at least three key factors: path geometric length (directly affecting flight time and energy consumption), path smoothness index (assessing turning rate to ensure compliance with UAV dynamics constraints), and minimum safe distance from obstacle boundaries (a core safety constraint). The system compares the weighted total cost of all candidate paths, selecting the path with the lowest total cost as the optimal connection scheme for the disconnected region, and precisely determines its optimal connection node with the skeleton path. Subsequently, the system inserts this optimal path as a new arc segment into the global path network, automatically updating the topological relationships of all relevant nodes, arc segments, and polygons to achieve seamless structural integration of the new branch with the original skeleton path. This process of identification, search, evaluation, and fusion is repeated until all identified disconnected regions have been processed.

[0059] In this embodiment, a closed-loop path construction process from "macro-skeleton" to "micro-completeness" is achieved, ensuring that the planned trajectory network covers all necessary navigable areas. Through a strategy combining "local fine-grained data retrieval" and "global topology fusion," the system avoids the need for global recalculation when repairing paths, thus avoiding the enormous overhead of global replanning triggered by local modifications in traditional methods, reducing the computational load of path repair by approximately 70%. Simultaneously, a multi-objective cost model is introduced for connection point decisions, ensuring that supplementary branches are not only geometrically connected but also optimal in terms of flight safety, smoothness, and efficiency, effectively preventing the introduction of sharp turns or dangerous close-proximity flight in pursuit of connectivity. The final output global discrete feasible path is a precise topology network composed of a series of ordered nodes and connecting arcs, providing a directly executable and reliable geometric foundation for subsequent path smoothing and motion control. This marks the realization of fully autonomous real-time planning from environmental perception to a complete navigable trajectory.

[0060] like Figure 4 The diagram shown is a structural schematic of a real-time trajectory planning and intelligent obstacle avoidance system for autonomous flight of a UAV provided in this application embodiment. This system includes: an environmental perception and dynamic prediction module, a data reduction and macroscopic analysis module, a spatial modeling and pattern library construction module, and a hierarchical path search and fusion module. The environmental perception and dynamic prediction module is used to acquire three-dimensional environmental perception data in real time, including the coordinates and dimensions of static obstacles and the position and speed of dynamic obstacles, through perception sensors mounted on the UAV; it sets a planning time interval based on the movement speed of the dynamic obstacles and performs preliminary statistical analysis on the perception data within a preset continuous time segment to determine the regularity of the dynamic obstacle movement pattern. The data reduction and macroscopic analysis module is used to sort the three-dimensional environmental perception data, extract the top-ranked key data, and further process it. Clustering analysis is performed to identify connected regions in the environment, thereby reducing the spatial scale and data volume of subsequent calculations in the early stages of global planning. The spatial modeling and pattern library construction module is used to establish a spatial geographic data model and corresponding topological relationship model based on a dual independent data structure. The environmental geometric features are decomposed into a finite number of standard patterns to build a pattern library, transforming the path search problem into a pattern recognition and matching problem to avoid large-scale repetitive calculations. The hierarchical path search and fusion module is used to perform pattern recognition and matching starting from the edge regions of the constructed spatial data model, prioritizing the search and connection to form initial feasible paths. For regions that are not connected, combined with the environmental data ranked lower, supplementary searches are performed based on the established data structure and topological relationships to calculate the minimum cost connection point with the already connected path and add it to the global path, resulting in feasible paths composed of discrete line segments.

[0061] In this embodiment, through the coordinated operation of four core modules, a closed-loop technology for real-time, safe, and efficient autonomous trajectory planning of UAVs in complex dynamic environments is achieved. First, the environmental perception and dynamic prediction module intelligently analyzes the movement patterns of dynamic obstacles and adaptively adjusts the planning frequency, reducing system response latency to milliseconds, fundamentally solving the real-time and safety challenges of high-speed dynamic obstacle avoidance. Second, the data reduction and macroscopic analysis module adopts a priority-based and dynamic extraction strategy to compress the amount of raw environmental data at the initial planning stage and quickly construct a macroscopic environmental skeleton representing connecting corridors and congested areas, providing clear structured guidance for global planning and greatly reducing subsequent costs. The computational complexity is reduced; furthermore, the spatial modeling and pattern library construction module achieves efficient topological representation of the environment through a dual independent data structure, and builds an scalable pattern library by combining historical data. This transforms a large number of local path search problems into real-time pattern matching, reducing the planning time in typical scenarios from seconds to milliseconds, significantly improving the algorithm's real-time performance and reducing computational power requirements. Finally, the hierarchical path search and fusion module adopts a hierarchical strategy of "growing a skeleton from the edge and patching it locally as needed." It prioritizes the rapid generation of global backbone paths using the pattern library, and then recalls detailed data for unconnected areas for refined connection and multi-objective optimization fusion. While ensuring global path connectivity and safety smoothness, it improves the overall planning efficiency. The entire system, through progressive processing of intelligent perception, lightweight data, topological modeling, pattern-based search, and refined fusion, ultimately outputs a high-quality discrete track that can be directly executed by UAVs, providing a feasible, scalable, high-performance autonomous navigation solution for low-altitude economic applications such as urban logistics and large-scale patrols.

[0062] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0063] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0064] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0065] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0066] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0067] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0068] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0069] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0070] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0071] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0072] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: S1, real-time acquisition of three-dimensional environmental perception data including the coordinates and size of static obstacles and the position and speed of dynamic obstacles through the perception sensors mounted on the drone; setting a planned time interval according to the movement speed of the dynamic obstacles, and performing preliminary statistical analysis on the perception data within the preset continuous time segment to determine the regularity of the movement pattern of the dynamic obstacles; S2 sorts the 3D environment perception data, extracts the key data at the top of the sort and performs cluster analysis to identify connected regions in the environment, thereby reducing the spatial scale and data volume of subsequent calculations in the early stage of global planning. S3, Establish a spatial geographic data model and corresponding topological relationship model based on a dual independent data structure; The environmental geometric features are decomposed into a finite number of standard patterns, and a pattern library is constructed. The path search problem is partially transformed into a pattern recognition and matching problem to avoid large-scale repetitive computation. S4. Starting from the edge region of the constructed spatial data model, pattern recognition and matching are performed. Priority is given to searching and connecting to form an initial feasible path. For regions that cannot be connected, combined with the environmental data that is ranked lower, supplementary searches are performed based on the established data structure and topological relationships. The minimum cost connection point between the region and the already connected path is calculated and added to the global path, resulting in a feasible path composed of discrete line segments.

2. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of unmanned aerial vehicles as described in claim 1, characterized in that, The specific steps for setting the planning time interval based on the movement speed of the dynamic obstacle are as follows: The planning time interval is inversely proportional to the speed of the dynamic obstacle. The perception data of the dynamic obstacle within a preset continuous time segment is calculated. The perception data includes the mean displacement, the variance of velocity, and the rate of change of the direction angle of motion. The mean displacement, velocity variance, and direction angle change rate are compared with their preset thresholds. If the velocity variance is lower than the first velocity threshold and the direction angle change rate is lower than the second change rate threshold, the dynamic obstacle movement pattern is determined to be regular movement. Otherwise, the movement pattern of the dynamic obstacle is determined to be random movement.

3. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of unmanned aerial vehicles as described in claim 1, characterized in that, The specific steps for sorting the 3D environment perception data are as follows: The acquired 3D environmental perception data is sorted according to preset rules to obtain an ordered dataset; For each obstacle data point, one or more feature attributes are assigned for sorting. The sorting process is performed according to one or a combination of the following two modes, and the specific steps are as follows: Calculate the three-dimensional Euclidean distance between each obstacle data point and the current position of the drone. Sort all obstacle data points in ascending order according to the three-dimensional Euclidean distance, so that the obstacle point closest to the drone is placed at the beginning of the sequence. Based on the total number of actual obstacle data points, the number of key data points is calculated, and the number of key data points is selected from the ordered dataset to form a key dataset.

4. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of unmanned aerial vehicles as described in claim 1, characterized in that, The specific steps for identifying connected regions in the environment are as follows: Cluster analysis was performed on key datasets to identify spatially clustered obstacle groups, and each identified cluster was treated as a composite obstacle region. Analyze the spatial relationships between complex obstacle areas to preliminarily identify connecting corridors and obstructed areas in the environment. The specific steps are as follows: Based on the geometric boundaries of each composite obstacle region, an environmental topology graph is constructed, each composite obstacle region is abstracted as a node in the environmental topology graph, and the minimum Euclidean distance between any two composite obstacle region boundaries is calculated. If the minimum distance between two composite obstacle areas is greater than the preset width threshold required for safe passage of the drone, it is determined that there is a connecting corridor between the two areas. The geometric features of the connecting corridor are defined by the line segment connecting the nearest point of the two composite obstacle areas and its vertical extension surface. If multiple complex obstacle areas are spatially adjacent and together form a closed area with a maximum inscribed circle diameter smaller than the preset width threshold required for safe passage of the drone, then the closed area is determined to be a blocking area.

5. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of an unmanned aerial vehicle as described in claim 1, characterized in that, The specific steps for establishing a spatial geographic data model based on a dual independent data structure are as follows: The node is used to represent an independent geographic location point, and its data unit contains at least a node ID and three-dimensional coordinates; The arc segment is used to represent a directed line segment, defined by a start node and an end node. Its data unit includes at least the arc segment ID, start node ID, end node ID, arc segment length, and the IDs of the polygons on the left and right sides of the arc segment. The polygon is used to represent a closed area, which is enclosed by a series of arc segments connected end to end. Its data unit contains at least a polygon ID and a sequence of arc segment IDs that make up the polygon. The polygon is used to identify obstacle areas or flyable airspace. The topological relationships between nodes, arcs, and polygons are automatically established and maintained by using the left and right polygon IDs and start and end node IDs recorded in the arc segment data unit.

6. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of an unmanned aerial vehicle as described in claim 5, characterized in that, The topological relationships include: the connection relationship between nodes and arc segments, the adjacency relationship between arc segments and polygons, and the adjacency relationship between polygons; The specific steps for decomposing environmental geometric features into a finite number of standard patterns and constructing a pattern library are as follows: Based on historical flight path planning data, local environmental geometric features are extracted. These local geometric features are composed of a predefined number of nodes, arcs, and polygons combined according to the topological relationships defined by the spatial geographic data model, and are used to characterize a typical path planning scenario. The extracted geometric features are categorized and defined as a finite number of standard patterns. Each standard pattern includes at least a pattern ID, a topology description, path generation rules, and a pattern library. The pattern ID is used to uniquely identify the pattern; the topology description is used to describe the connection and adjacency relationships of nodes, arcs, and polygons in the pattern; the path generation rules are used to predefine one or more optimal feasible paths for the pattern; the pattern library establishment is used to store all predefined standard patterns and their associated information as a queryable pattern library.

7. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of unmanned aerial vehicles as described in claim 1, characterized in that, The specific steps for transforming the path search problem into a pattern recognition and matching problem are as follows: Match the node-arc-polygon topology of the current local environment with the standard patterns in the pattern library; If a match is found with a standard pattern in the pattern library, the pre-stored path generation rules in the standard pattern are directly called to obtain a feasible path for the local area, thereby avoiding real-time, large-scale geometric search and calculation for the local environment. If a match with a standard pattern in the pattern library fails, a path search algorithm is activated to perform calculations. The features of the new environment and the final calculated optimal path are used as new samples, which are then learned and added to the pattern library to achieve self-expansion of the pattern library.

8. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of an unmanned aerial vehicle as described in claim 1, characterized in that, The specific steps for forming a connected initial feasible path are as follows: The mission start point, mission target point, and nodes located on the boundary of flyable airspace in the spatial data model are defined as edge nodes, and path search is initiated from the edge nodes. Starting from each edge node, query the arc segments and adjacent polygons connected to it, and perform real-time matching of the current local topology with the pattern library constructed in step S3; If a standard pattern is matched, the optimal connection rule or path segment pre-stored in that pattern will be used first to expand into the space and quickly connect to the next key node. Repeat the matching process described above, prioritizing the connection of path branches extended from different edge nodes, until one or more initial feasible paths consisting of key nodes and arc segments are formed that traverse the main flyable area.

9. The method for real-time trajectory planning and intelligent obstacle avoidance for autonomous flight of an unmanned aerial vehicle as described in claim 1, characterized in that, The specific steps for obtaining the feasible path composed of discrete line segments are as follows: S100, the region not covered by the initial feasible path skeleton and surrounded by obstacle polygons is identified as an unconnected region; S101, Combining the environmental data that was sorted and processed later in step S2, in the unconnected area, based on the established dual independent data structure, calculate all reachable nodes from the area to the initial feasible path and their corresponding costs. S102, evaluate the cost of all calculated reachable node paths, the path cost being a multi-objective weighted function whose factors include at least: path geometric length, path smoothness index, and minimum safe distance to obstacle boundaries; S103, by comparing the weighted total cost of all reachable node paths, the path with the lowest cost is selected as the optimal connection scheme, and its connection nodes with the skeleton path are determined; the optimal path is inserted as a new arc segment into the global path network, and the topological relationship between the relevant nodes and the polygon is updated to achieve structural fusion with the initial skeleton path. Repeat steps S100-S103 until all unconnected regions have been processed to generate a globally discrete feasible path that integrates the initial skeleton path and all supplementary branch paths. The globally discrete feasible path consists of an ordered sequence of nodes and their connecting arcs.

10. A real-time trajectory planning and intelligent obstacle avoidance system for autonomous flight of unmanned aerial vehicles (UAVs), wherein the system is used to implement the real-time trajectory planning and intelligent obstacle avoidance method for autonomous flight of UAVs as described in any one of claims 1-9, characterized in that, It includes modules for environmental perception and dynamic prediction, data reduction and macroscopic analysis, spatial modeling and pattern library construction, and hierarchical path search and fusion. The environmental perception and dynamic prediction module is used to acquire three-dimensional environmental perception data, including the coordinates and size of static obstacles and the position and speed of dynamic obstacles, in real time through the perception sensors mounted on the UAV; to set a planned time interval according to the movement speed of the dynamic obstacles, and to perform preliminary statistical analysis on the perception data within the preset continuous time segment to determine the regularity of the movement pattern of the dynamic obstacles. The data reduction and macroscopic analysis module is used to sort the 3D environment perception data, extract the key data at the top of the sort and perform cluster analysis to identify connected regions in the environment, thereby reducing the spatial scale and data volume of subsequent calculations in the early stage of global planning. The spatial modeling and pattern library construction module is used to establish a spatial geographic data model and a corresponding topological relationship model based on a dual independent data structure. The environmental geometric features are decomposed into a finite number of standard patterns, and a pattern library is constructed. The path search problem is partially transformed into a pattern recognition and matching problem to avoid large-scale repetitive computation. The hierarchical path search and fusion module is used to start from the edge region of the constructed spatial data model, perform pattern recognition and matching, prioritize the search and connection to form an initial feasible path. For areas that cannot be connected, combined with the environmental data that are ranked lower, a supplementary search is performed based on the established data structure and topological relationship. The minimum cost connection point between the area and the already connected path is calculated and added to the global path, resulting in a feasible path composed of discrete line segments.

Images