A three-dimensional path planning method for unmanned aerial vehicles in urban environment

Abstract

The application discloses a kind of unmanned aerial vehicle three-dimensional track planning methods for urban environment, it is related to unmanned aerial vehicle track planning technical field.The method is blocked by space ray analysis track direction, and candidate obstacle avoidance point outside safety margin is dynamically generated using the geometric boundary of obstacle combined with unmanned aerial vehicle flight constraint condition;Topological optimization is carried out in discrete candidate point set based on cost function, and the initial path that meets the requirements of obstacle avoidance and the total cost is the smallest is quickly locked;Redundant inflection point is removed using bidirectional line-of-sight detection, and smooth curve is introduced to track fitting, so that it meets the kinematic constraint of unmanned aerial vehicle.The application extracts the geometric contour boundary of building and carries out sparse sampling search combined with heuristic cost function, to realize the efficient planning of unmanned aerial vehicle track in urban environment, while ensuring safe obstacle avoidance, reduce the memory and computing power consumption caused by global gridding modeling.

Description

Technical Field

[0001] This application relates to the field of UAV trajectory planning technology, and in particular to a UAV three-dimensional trajectory planning method for urban environments. Background Technology

[0002] In recent years, with the rapid development of drone technology, its applications in urban environments such as logistics delivery and emergency inspection have become increasingly widespread. However, the dense buildings and complex spatial structures of urban environments place extremely high demands on the autonomous obstacle avoidance and trajectory planning capabilities of drones, leading to a growing focus on three-dimensional path planning for drones both domestically and internationally.

[0003] Currently, mainstream UAV 3D path planning methods rely on constructing high-resolution global grid maps and applying heuristic search algorithms such as A* on this basis. For example, Liu et al. proposed an A*-APF algorithm based on spatio-temporal grid optimization in the journal Drones (Liu L, Ru L, Wang W, et al. UAV Path Planning in Threat Environment: A*-APF Algorithm for Spatio-Temporal Grid Optimization[J]. Drones, 2025, 9(9): 661). This method uses BeiDou grid codes to achieve multi-granularity spatial decomposition and fusion, optimizes heuristic search features from a spatio-temporal collaborative perspective, and combines a sliding window mechanism to guide local trajectories. The algorithm sets up a three-level network of kilometers, hundreds of meters, and tens of meters to divide the search space for different threat environments and flight mission requirements. However, its search space grows exponentially with the increase of map resolution, requiring high memory and computing resources. Moreover, the explorable range of the environment is limited by the grid resolution, resulting in high memory consumption and high computational latency, making it difficult to achieve real-time planning on platforms with limited airborne computing power.

[0004] In summary, although existing path planning models can generate collision-free safe trajectories, under the same start and end point conditions, their search space, memory usage, and computational latency are all random, making it difficult to achieve stable real-time performance when onboard computing power is limited. Summary of the Invention

[0005] Therefore, it is necessary to provide a UAV 3D trajectory planning method for urban environments to address the aforementioned technical problems.

[0006] The following technical solution is adopted in this specification: This specification provides a method for UAV 3D trajectory planning in urban environments, including: Obtain the drone's starting position and the preset target position; A virtual detection line of sight is established on the straight path between the starting position and the target position. The intersection point of the obstacle and the virtual detection line of sight on the straight path is taken as the line of sight obstruction point, and the physical boundary of the obstacle corresponding to the line of sight obstruction point is obtained. Candidate points are generated at a preset safety margin outside the physical boundary of each obstacle to obtain a set of discrete candidate points for bypassing. Using the starting position node, discrete candidate points, and target position node as search nodes, a heuristic search algorithm guided by a cost function is adopted. Starting from the starting position node, the candidate point with the minimum comprehensive cost is dynamically selected from the set of discrete candidate points for path expansion until the currently expanded node and the target position node meet the line-of-sight condition. Then, an initial discrete track is generated by backtracking, consisting of the starting position node, the selected discrete candidate point nodes, and the target position node. Redundant nodes are removed and curves are smoothed from the initial discrete tracks to generate the final tracks that conform to the UAV.

[0007] Furthermore, city buildings are simplified into regular axis-aligned bounding boxes to accelerate the calculation of intersection detection between the virtual detection line of sight and obstacles.

[0008] Furthermore, the physical boundaries of the obstacle include: Identify the points where the virtual detection line of sight is obstructed from the obstacle, and the surfaces where the obstructed lines of sight are located; The geometric boundary of the surface where the line of sight is obstructed is taken as the physical boundary of the obstacle.

[0009] Furthermore, the generation of the discrete candidate point set includes: If the surface where the line of sight is obstructed is the side of the obstacle, then based on a preset safety margin, extend outward along the horizontal and vertical directions of the side to generate lateral detour candidate points in the horizontal and vertical directions respectively. If the surface where the line of sight is blocked is the top surface of the obstacle, a high-level sliding strategy is adopted. On the top safe plane of the obstacle, the candidate point for flying over is obtained by extending outward along the horizontal edge direction vector. Combining the drone's flight altitude constraints and the permitted space range, the legality of the lateral detour candidate points and the overhead overpass candidate points is verified, and invalid points that exceed the flight altitude limit or enter a known no-fly zone are eliminated; If the candidate point for flying over is eliminated due to constraint verification, a backtracking mechanism is automatically triggered to generate a horizontal detour candidate point based on the lateral physical boundary of the obstacle. This horizontal detour candidate point, together with the lateral detour candidate point and the candidate point for flying over, constitutes the discrete candidate point set.

[0010] Furthermore, the cost function of the heuristic search algorithm is expressed as the sum of the cumulative travel distance cost from the starting position node to the current candidate node and the estimated remaining journey cost from the current candidate node to the target position node obtained using the Euclidean norm.

[0011] Furthermore, redundant nodes are removed through forward and reverse traversal pruning, specifically including: For all nodes in the initial discrete path, starting from the initial position node, the farthest subsequent node that satisfies the line-of-sight condition is obtained through forward traversal pruning, and redundant nodes between the initial position node and the farthest subsequent node are removed. Based on the nodes after removing redundant nodes, starting from the target location node, the farthest predecessor node that satisfies the visibility condition is obtained through reverse traversal pruning, and redundant nodes between the target location node and the farthest predecessor node are removed.

[0012] Furthermore, the curve smoothing process uses cubic Bézier curves to locally reconstruct the bends after redundant nodes are removed, so as to ensure that the UAV trajectory meets the continuity requirements at the bends.

[0013] This specification provides a UAV 3D trajectory planning system for urban environments, including: Location acquisition module: used to acquire the drone's starting position and the preset target position; The line-of-sight detection and track point generation module is used to establish a virtual line of sight on the straight path between the starting position and the target position, take the intersection point of the obstacle that intersects with the virtual line of sight on the straight path as the line-of-sight obstruction point, and obtain the physical boundary of the obstacle where the line-of-sight obstruction point is located; generate candidate points at a preset safety margin outside the physical boundary of each obstacle to obtain a set of discrete candidate points for intelligent detour; The heuristic track search module is used to take the starting position, discrete candidate points and target position as search nodes, and adopt a heuristic search algorithm guided by a cost function to dynamically select the discrete candidate point node with the minimum comprehensive cost from the starting position node and expand it until the starting position node and the target position node meet the line-of-sight condition, and backtrack to generate an initial discrete track composed of the starting position node, the selected discrete candidate point node and the target position node. The trajectory optimization and smoothing module is used to remove redundant nodes and smooth the curves of the initial discrete trajectory to generate a final trajectory that conforms to the UAV.

[0014] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for UAV three-dimensional trajectory planning in urban environments.

[0015] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for UAV three-dimensional trajectory planning in urban environments.

[0016] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: In the UAV 3D trajectory planning method for urban environments provided in this specification, a virtual line of sight is established on the straight path between the starting position and the target position, and candidate points outside the safety margin are dynamically generated using the physical boundaries of obstacles. An optimization is performed on the discrete candidate point set based on a cost function to quickly find the initial discrete trajectory with the minimum total cost. Redundant nodes in the initial discrete trajectory are then removed and the curve is smoothed to generate the final trajectory suitable for the UAV. Specifically, by employing spatial ray detection and heuristic search techniques, the physical boundaries of obstacles are analyzed and detour candidate points are dynamically generated. This transforms the search problem in continuous 3D space into topological optimization of a discrete candidate point set, avoiding indiscriminate blind node expansion and eliminating the need to quickly lock the optimal trajectory under global grid modeling, significantly reducing memory usage and computational burden. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This document provides a flowchart illustrating a UAV 3D trajectory planning method for urban environments. Figure 2 This is a schematic diagram of line-of-sight detection and intelligent detour candidate point generation provided in this specification. Figure 3 This is a schematic diagram of track pruning based on bidirectional line-of-sight detection provided in this manual; Figure 4 This is a schematic diagram of trajectory smoothing based on cubic Bézier curves provided in this specification. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0019] Existing UAV 3D path planning methods can also rely on local sensor information for real-time obstacle avoidance without global modeling. For example, Jang et al. proposed a voxel-based optimal path planning method in the ISPRS International Journal of Geo-Information (Jang M, Kim D, Lee J. A Voxel-BasedOptimal Path Planning Method for UAV Navigation in Smart Cities. ISPRS International Journal of Geo-Information. 2025 Nov 23;14(12):457.). This method first quantifies multiple dynamic risks such as obstacles and weather through a voxel model to construct a 3D flight risk map; then, it uses an extended 3D wavefront algorithm to calculate the risk-weighted distance and performs path smoothing through local waypoint interpolation. This method introduces a flight risk index parameter to balance the tendency of the trajectory to take the shortest path and avoid comprehensive risks.

[0020] Zhao et al. proposed an adaptive path planning algorithm for UAVs based on A* algorithm and artificial potential field method in the journal *Drones* (Zhao J, Zhang Y, Ning L, et al. Adaptive Path Planning of UAVBased on A* Algorithm and Artificial Potential Field Method[J]. Drones, 2026,10(1): 93.). This method uses the A* algorithm at the front end to find discrete global paths in the grid, and utilizes the attraction and repulsion model of the artificial potential field method at the back end for local obstacle avoidance and trajectory smoothing. However, the front-end node expansion of this method lacks geometric feature guidance and exhibits indiscriminate divergence.

[0021] Zhu Wenjie et al. proposed a method for UAV urban low-altitude logistics path planning based on an improved A* algorithm in the journal *Computer Engineering and Applications* (Zhu Wenjie, Li Wei, Wang Ziyan. Improved A* algorithm for UAV urban low-altitude logistics path planning [J]. *Computer Engineering and Applications*, 2025, 61(19): 302-310.). This method first constructs a 3D map using a raster method, and then optimizes the heuristic cost function of the A* algorithm by introducing a 3D vector cross product and a dynamic weight mechanism, thereby reducing the search for redundant nodes. However, this method relies on high-resolution 3D global raster modeling, leading to an exponential explosion in the search space.

[0022] Specifically, the existing technology has the following problems that need to be solved: Existing methods rely heavily on high-resolution 3D global rasterization modeling. When faced with large-scale urban environments, the search space tends to explode exponentially, resulting in large memory consumption and high computational latency. Existing methods require gradually exploring the surrounding space to expand nodes, but because they fail to directly extract and utilize the three-dimensional physical boundaries of obstacles for effective guidance, the search process is prone to indiscriminate and blind divergence.

[0023] To address the aforementioned technical shortcomings, this invention employs spatial ray detection technology to extract building geometric information during the 3D path search process. Through intersection detection, it directly analyzes the impact axis of obstacles and dynamically generates a small number of candidate points outside their safety margin. This allows the generated trajectory to quickly bypass buildings while conforming to the kinematic constraints of the UAV. The aim is to solve the technical problems of high memory consumption, low search efficiency, and susceptibility to local deadlocks in existing UAV 3D path planning methods applied in complex urban environments.

[0024] The keywords involved in this invention include: Unmanned Aerial Vehicle (UAV): An UAV is an unmanned aircraft controlled by radio remote control equipment and its own program control device. Its safe operation relies on intelligent three-dimensional spatial navigation technology. With the help of trajectory planning and obstacle avoidance algorithms, it finds a collision-free path in three-dimensional space and ultimately generates a safe and continuous trajectory that can avoid spatial obstacles and meet the kinematic constraints of the aircraft itself.

[0025] Spatial Ray Detection: Spatial ray detection is a geometric analytical technique that detects occlusion and physical boundaries of target objects in three-dimensional space by projecting virtual rays. It quickly determines the visibility between two points by calculating the geometric relationship between the ray and the spatial entity, and extracts the spatial features of the obstacle surface. This technique replaces traditional global high-resolution raster modeling, performing local detection on demand, effectively reducing memory consumption and computational burden.

[0026] Heuristic Search: Heuristic search is an intelligent optimization mechanism that guides the algorithm to prioritize exploring the most likely path to the target node using a specific cost evaluation function. By comprehensively evaluating the current movement cost and the estimated remaining distance, it dynamically selects the optimal development direction in the state space. This mechanism overcomes the indiscriminate blind exploration in traditional path planning, enabling the search process to converge quickly to the target point and effectively improving pathfinding efficiency.

[0027] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0028] Figure 1 This document presents a flowchart illustrating a UAV 3D trajectory planning method for urban environments, comprising three core modules: a line-of-sight detection and trajectory point generation module, a heuristic trajectory search module, and a trajectory optimization and smoothing module. The line-of-sight detection and trajectory point generation module analyzes the obstructed trajectory orientation using spatial rays and dynamically generates candidate obstacle avoidance points outside the safety margin using the geometric boundaries of obstacles. The heuristic trajectory search module performs topology optimization on a discrete set of candidate points based on a cost function, quickly identifying the initial path that meets obstacle avoidance requirements and minimizes the total cost. The trajectory optimization and smoothing module uses bidirectional line-of-sight detection to eliminate redundant inflection points and introduces a smooth curve for trajectory fitting, ensuring it conforms to UAV kinematic constraints. This invention aims to establish a lightweight representation of path planning problems in urban environments and, by combining spatial ray detection and obstacle physical feature extraction, generates efficient and safe UAV 3D obstacle avoidance trajectories through heuristic optimization and trajectory smoothing.

[0029] S101: Line of sight detection and waypoint generation module.

[0030] like Figure 2 As shown, in the current node (Initial Index) ) and target node Establish a virtual line of sight between them It is used to pre-determine whether there are obstacles on a straight path: (1); in This is the parameterized time constant.

[0031] To reduce onboard computing and storage overhead, this invention simplifies complex urban architecture into regular axis-aligned bounding boxes. When a ray is detected passing through a building, the point of visual obstruction is calculated using spatial geometry. Subsequently, according to And the physical boundary of the building, the safety margin at the edge of the interception surface In addition, a set of intelligent detour candidate points is calculated. : (2); in Let be the set of normal vectors perpendicular to the hit axis. Here, to find the obstacle avoidance path with minimum cost, the system determines candidate points based on a strategy of "multi-dimensional coverage, optimal detour": if When a point hits the side of a building, the algorithm will simultaneously generate candidate points in both horizontal and vertical dimensions, one along the horizontal vector. Extend outwards The system uses distance-based positioning to quickly lock onto lateral corner points for smooth maneuvering, while simultaneously extending along the height vector to the top of buildings, providing an overhead flight path. During this process, the system retrieves preset flight height restrictions. The system verifies the legitimacy of candidate points in real time based on no-fly zone information. If a candidate point exceeds the height limit or enters a no-fly zone, that branch will be assigned a very high cost weight to ensure that the generated path always remains within the permitted space. Correspondingly, if... Upon hitting the top of a building, the algorithm prioritizes a "high-level sliding" strategy, attempting to generate candidate overflight points on a safe plane at the top of the building, guiding the drone to traverse over from above. From a spatial geometry perspective, since the detection line of sight passes through the top of the building, it must also intersect with the space to the side of the building. Therefore, when the system's real-time verification detects that the "overflight" path exceeds the flight height limit... If the algorithm fails due to entering a no-fly zone, it will immediately trigger a rollback mechanism. This mechanism will abandon the top surface path and instead consider a side detour. Its specific generation principle is completely consistent with the processing logic for the aforementioned "ray directly hitting the side": that is, directly utilizing the lateral bounding box boundary of the building, extending outwards horizontally. Distance is used to supplement and generate lateral detour candidate points in the horizontal dimension, ultimately achieving automatic optimization between "lateral detour" and "vertical overpass" by combining real-time path cost and multiple flight constraints.

[0032] S102: Heuristic track search module.

[0033] After obtaining the discrete candidate point set generated by the aforementioned line-of-sight detection module, a cost-guided heuristic search mechanism is used to quickly lock the optimal topological path among a large number of spatial combinations. By maintaining a priority queue, the comprehensive cost is selected in real time. The node with the smallest value is chosen as the exploration direction, and the cost function is defined as follows: (3); in This represents the distance from the starting point to the current candidate node. The cumulative cost of travel distance, This is the remaining distance from the current point to the destination, calculated using the Euclidean norm.

[0034] The search algorithm then starts from the starting point. Initially, continuously extract from the priority queue. The candidate point with the lowest cost is selected as the current point of consideration. ,like With the target point If the line-of-sight condition is met, the search iteration stops. At this point, The parent node points directly to the current node: (4); Once the search iteration stops, the algorithm retrieves data from the parent node pointer. Backtracking Generate the initial discrete path: (5); S103: Track optimization and smoothing module.

[0035] Since the initial trajectory generated by the aforementioned module is a polygonal line composed of discrete points, containing numerous sharp angles, it does not conform to the kinematic constraints of the UAV. This module processes the trajectory in two steps: Step 1: Track pruning based on two-way line-of-sight detection like Figure 3 As shown, an initial discrete track point set is defined. ,in Starting from, As the endpoint, here for The total number of waypoints in the system And it must be an integer. The algorithm starts from the current node. (Initial Index) Starting from the index, traverse the index in reverse order. (from Decrease to Find the furthest visible point that satisfies the collision safety constraints: If the current node With the point to be measured straight line segments between All nodes p on the network are at a distance greater than the safety margin from the building. Then the point to be measured The furthest visible point to meet collision safety constraints.

[0036] Once determined and If the conditions for safe visibility between them are met, then the index is determined. All intermediate nodes are redundant and should be deleted.

[0037] Once the forward pruning from the starting point to the ending point is completed, the sequence of retained track nodes is reversed, and the above-mentioned search process for the farthest visible point is executed again for reverse pruning to completely eliminate all redundant inflection points.

[0038] Step 2: Local reconstruction based on cubic Bézier curves like Figure 4 As shown, for the angle retained after pruning, a cubic Bézier curve equation is introduced. B (t Perform local reconstruction to ensure track continuity at turns: (6); in, and The points of tangency are on both sides of the corner. and These are control points generated based on steering distance. Here are the control points. and The value of depends on the angle between two adjacent broken line segments and the preset turning radius, tangent point and The position is determined by cutting along the broken lines to both sides of the inflection point at a set distance or ratio.

[0039] Compared with the prior art, the method described in this invention can achieve the following beneficial effects: Innovative and Advanced Methodology: Addressing the issues of high memory consumption and computational demands caused by existing UAV 3D path planning methods that rely on high-resolution global raster modeling, this invention employs spatial ray detection and heuristic search techniques. Through line-of-sight detection and track point generation modules, it analyzes the physical boundaries of obstacles and dynamically generates detour candidate points, enabling rapid locking of the optimal track without requiring global raster modeling. Furthermore, this invention introduces bidirectional line-of-sight detection pruning and cubic Bézier curves for local reconstruction, eliminating redundant points and sharp angles in the path, ensuring that the final generated track has the shortest distance and conforms to the kinematic constraints of actual UAV flight.

[0040] Innovation and Advancement: Based on computational geometry and heuristic optimization theory, this invention constructs a 3D trajectory planning algorithm framework for UAVs in urban environments. Building upon this, to meet the needs of practical engineering applications and performance evaluation, we conducted simulations in both simple and complex urban scenarios. Using metrics such as algorithm path length, computation time, and number of sampling points, we comprehensively evaluated the superior performance of this invention in achieving computationally lightweight design, efficient search, and ensuring the safe and stable operation of the aircraft.

[0041] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for planning the three-dimensional flight path of unmanned aerial vehicles (UAVs) in urban environments, characterized in that, include: Obtain the drone's starting position and the preset target position; A virtual detection line of sight is established on the straight path between the starting position and the target position. The intersection point of the obstacle and the virtual detection line of sight on the straight path is taken as the line of sight obstruction point, and the physical boundary of the obstacle corresponding to the line of sight obstruction point is obtained. Candidate points are generated at a preset safety margin outside the physical boundary of each obstacle to obtain a set of discrete candidate points for bypassing. The starting position node, discrete candidate points, and target position node are used as search nodes. A heuristic search algorithm guided by a cost function is used on the search nodes. Starting from the starting position node, the candidate point with the minimum comprehensive cost is dynamically selected from the set of discrete candidate points for path expansion until the currently expanded node and the target position node meet the line-of-sight condition. Then, an initial discrete track is generated by backtracking, consisting of the starting position node, the selected discrete candidate point nodes, and the target position node. Redundant nodes are removed and curves are smoothed from the initial discrete tracks to generate the final tracks that conform to the UAV.

2. The UAV three-dimensional trajectory planning method for urban environments as described in claim 1, characterized in that, Urban buildings are simplified into regular axis-aligned bounding boxes to accelerate the calculation of intersection detection between the virtual detection line of sight and obstacles.

3. The UAV three-dimensional trajectory planning method for urban environments as described in claim 1, characterized in that, The physical boundaries of the obstacle include: Identify the points where the virtual detection line of sight is obstructed from the obstacle, and the surfaces where the obstructed lines of sight are located; The geometric boundary of the surface where the line of sight is obstructed is taken as the physical boundary of the obstacle.

4. The UAV three-dimensional trajectory planning method for urban environments as described in claim 1, characterized in that, The generation of the discrete candidate point set includes: If the surface where the line of sight is obstructed is the side of the obstacle, then based on a preset safety margin, extend outward along the horizontal and vertical directions of the side to generate lateral detour candidate points in the horizontal and vertical directions respectively. If the surface where the line of sight is blocked is the top surface of the obstacle, a high-level sliding strategy is adopted. On the top safe plane of the obstacle, the candidate point for flying over is obtained by extending outward along the horizontal edge direction vector. Combining the drone's flight altitude constraints and the permitted space range, the legality of the lateral detour candidate points and the overhead overpass candidate points is verified, and invalid points that exceed the flight altitude limit or enter a known no-fly zone are eliminated; If the candidate point for flying over is eliminated due to constraint verification, a backtracking mechanism is automatically triggered to generate a horizontal detour candidate point based on the lateral physical boundary of the obstacle. This horizontal detour candidate point, together with the lateral detour candidate point and the candidate point for flying over, constitutes the discrete candidate point set.

5. The UAV three-dimensional trajectory planning method for urban environments as described in claim 1, characterized in that, The cost function of the heuristic search algorithm is expressed as the sum of the cumulative travel distance cost from the starting position node to the current candidate node and the estimated remaining journey cost from the current candidate node to the target position node obtained using the Euclidean norm.

6. The UAV three-dimensional trajectory planning method for urban environments as described in claim 1, characterized in that, Redundant nodes are removed from the initial discrete track through forward and reverse traversal pruning, including: For all nodes in the initial discrete path, starting from the initial position node, the farthest subsequent node that satisfies the line-of-sight condition is obtained through forward traversal pruning, and redundant nodes between the initial position node and the farthest subsequent node are removed. Based on the nodes after removing redundant nodes, starting from the target location node, the farthest predecessor node that satisfies the visibility condition is obtained through reverse traversal pruning, and redundant nodes between the target location node and the farthest predecessor node are removed.

7. The UAV three-dimensional trajectory planning method for urban environments as described in claim 1, characterized in that, The curve smoothing process uses cubic Bézier curves to locally reconstruct the bends after redundant nodes are removed, so as to ensure that the UAV trajectory meets the continuity at the bends.

8. A three-dimensional trajectory planning system for unmanned aerial vehicles (UAVs) in urban environments, characterized in that: include: Location acquisition module: used to acquire the drone's starting position and the preset target position; The line-of-sight detection and track point generation module is used to establish a virtual line of sight on the straight path between the starting position and the target position, take the intersection point of the obstacle that intersects with the virtual line of sight on the straight path as the line-of-sight obstruction point, and obtain the physical boundary of the obstacle where the line-of-sight obstruction point is located; generate candidate points at a preset safety margin outside the physical boundary of each obstacle to obtain a set of discrete candidate points for intelligent detour; The heuristic track search module is used to take the starting position, discrete candidate points and target position as search nodes. It adopts a heuristic search algorithm guided by a cost function to dynamically select the discrete candidate point node with the minimum comprehensive cost from the starting position node and expand it until the starting position node and the target position node meet the line-of-sight condition. Then, it backtracks to generate an initial discrete track composed of the starting position node, the selected discrete candidate point nodes and the target position node. The trajectory optimization and smoothing module is used to remove redundant nodes and smooth the curves of the initial discrete trajectory to generate a final trajectory that conforms to the UAV.

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

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.

Images