A three-dimensional path planning method for urban low-altitude unmanned aerial vehicles

Abstract

The present application belongs to the technical field of unmanned aerial vehicle three-dimensional path planning, and particularly relates to a three-dimensional path planning method for urban low-altitude unmanned aerial vehicles. The present application significantly reduces the size of the three-dimensional search space, reduces the search complexity, improves the planning speed, and ensures that the unmanned aerial vehicle maintains a safe distance from obstacles and reduces the collision risk by using the obstacle inflation and dangerous area avoidance strategy. The bidirectional pruning and intermediate node insertion optimize the path, making the unmanned aerial vehicle flight path smooth and continuous, which is beneficial for actual flight control execution. The present application considers the free space distribution of multiple height layers, obstacle density and urban complex terrain, can generate a feasible path in high-density building areas, and can be combined with various unmanned aerial vehicle task scenarios.

Description

Technical Field

[0001] This invention belongs to the field of UAV three-dimensional path planning technology, specifically relating to a three-dimensional path planning method for urban low-altitude UAVs. Background Technology

[0002] With the rapid development of urban drone applications, the demand for drones to perform tasks such as logistics transportation, inspection and monitoring, and emergency rescue in low-altitude environments is increasing. The low-altitude economy has become a national strategic emerging industry, and drones, as the core carrier of this economy, face a critical bottleneck in achieving large-scale commercial applications in complex urban environments due to their autonomous navigation path planning technology. The A* search algorithm (also known as the A* algorithm) has become the mainstream algorithm for global path planning of drones due to its good stability and ability to guarantee finding the optimal path in a static, completely known environment. This algorithm typically prioritizes search nodes by constructing a node evaluation function that includes both the actual path cost and the heuristic estimated cost, thereby gradually approaching the target node. However, in complex urban low-altitude environments, the traditional A* algorithm still has certain limitations in practical applications. Existing technologies often directly apply the A* algorithm to two-dimensional or three-dimensional spatial path search and uniformly adopt a fixed form of heuristic function, such as an estimation method based on Euclidean distance, without distinguishing between different stages of the path planning process. In reality, in multi-stage path planning, the requirements for heuristic functions differ significantly across different spatial dimensions and planning stages.

[0003] In existing technologies, when traditional A* search algorithms are applied to urban low-altitude 3D path planning, they directly extend the 2D 8-neighborhood search to a 3D 26-neighborhood search. They do not design adaptation strategies for the structured environment of urban low-altitude areas (layered building heights, clear obstacle distribution in horizontal space). Some 3D path planning methods mainly establish flight space models through gridding or sampling, which is insufficient for risk assessment around obstacles. The planned paths may be close to buildings or other obstacles, posing a collision risk. In the path search process, there is a lack of a comprehensive cost evaluation mechanism, making it impossible to achieve a balance between path length, flight safety, and search efficiency. There is also a lack of effective path optimization and pruning mechanisms, resulting in too many path inflection points and poor smoothness, which affects the stability and efficiency of UAV flight. Summary of the Invention

[0004] The purpose of this invention is to provide a three-dimensional path planning method for urban low-altitude unmanned aerial vehicles (UAVs), which can achieve efficient, safe and smooth three-dimensional path planning in complex urban low-altitude environments, provide reliable support for the autonomous flight of UAVs in urban low-altitude areas, and improve flight safety and mission execution efficiency.

[0005] The specific technical solution adopted by this invention is as follows:

[0006] A three-dimensional path planning method for urban low-altitude unmanned aerial vehicles (UAVs) includes:

[0007] The urban low-altitude flight area is discretized into a three-dimensional raster to establish a three-dimensional spatial model, and the three-dimensional coordinates of the UAV's flight start and end points are obtained to set the allowable flight altitude range and safe distance.

[0008] A layered traversal is performed within the range from the starting point altitude to the permissible flight altitude. Based on the area constructed by projecting the starting and ending points, the altitude layers that meet preset conditions are selected as key altitude layers.

[0009] Project the start and end points onto each critical height layer, and perform path search at each critical height layer to obtain a two-dimensional path point sequence.

[0010] The two-dimensional path point sequences of each key height layer are merged to obtain a two-dimensional path projection set, which is then extended to the free space grid along the height direction to construct a restricted three-dimensional search space. The restricted three-dimensional search space is then expanded to a neighborhood to obtain an extended restricted three-dimensional search space.

[0011] Perform path search within the extended restricted 3D search space to obtain the initial 3D path from the starting point to the ending point;

[0012] The initial 3D path is pruned bidirectionally, and intermediate nodes are inserted during the pruning process to obtain an optimized path;

[0013] The obstacle grid is expanded based on the safety distance to form a danger zone, and the danger zone is avoided during the path search and optimization process to obtain the final path.

[0014] In a preferred embodiment, the urban low-altitude flight area is discretized using a three-dimensional rasterization process to establish a three-dimensional spatial model. The three-dimensional coordinates of the UAV's flight start and end points are obtained, and the permissible flight altitude range and safe distance are set, including:

[0015] The urban low-altitude flight area is discretized in three-dimensional space according to a preset spatial resolution along the horizontal and vertical directions to form multiple three-dimensional grid units.

[0016] The system acquires flight area data, which includes building data, terrain data, and airspace restriction data. Based on the flight area data, it obtains the spatial occupancy status of each three-dimensional grid cell. Three-dimensional grid cells that are occupied by obstacles or located in no-fly zones are marked as obstacle grids, and the rest are marked as free space grids. A three-dimensional spatial model is then constructed.

[0017] Acquire drone mission information, and based on the drone mission information, obtain the three-dimensional coordinates of the drone's flight start and end points, and map the three-dimensional coordinates to the corresponding three-dimensional grid cells according to the preset spatial resolution;

[0018] Obtain the UAV flight performance parameters and airspace management constraints, and obtain the permissible flight altitude range based on the UAV flight performance parameters and airspace management constraints;

[0019] The system acquires the drone's external dimensions and the environmental complexity of the flight area, and combines this with preset flight safety rules to determine the safe distance between the drone and obstacles.

[0020] In a preferred embodiment, a layered traversal is performed within the range from the altitude corresponding to the starting point to the permissible flight altitude. This is expanded based on the area constructed using the projections of the starting and ending points, and altitude layers that meet preset conditions are selected as key altitude layers, including:

[0021] Within the range from the altitude corresponding to the drone's flight start point to the allowable flight altitude, the three-dimensional spatial model is divided into layers according to a preset altitude interval to obtain multiple candidate altitude layers;

[0022] In each candidate altitude layer, the three-dimensional coordinates of the UAV's flight start and end points are projected onto the corresponding altitude layer to obtain the projection position for each altitude layer;

[0023] The basic region is obtained based on the projection position, and the basic region is spatially expanded according to a preset expansion distance to obtain the candidate search region;

[0024] Within the candidate search area, perform connected component labeling on the free space raster to obtain free space connected regions, and obtain the corresponding number of connected regions and the area of ​​each connected region;

[0025] The number of obstacle grids in the candidate search area is obtained, and the obstacle grid density is obtained based on the number of obstacle grids per unit volume. The obstacle distribution gradient is obtained based on the density change between spatially adjacent grids.

[0026] The difference in the number of connected regions, the difference in the area of ​​connected regions, the difference in the density of obstacle grids, and the difference in the distribution gradient between the current candidate height layer and the adjacent candidate height layers are obtained and compared with the preset change threshold corresponding to each difference. When any difference exceeds the corresponding preset change threshold, the corresponding candidate height layer is determined as the critical height layer.

[0027] In a preferred embodiment, the start and end points are projected onto each critical height level, and a path search is performed at each critical height level to obtain a two-dimensional path point sequence, including:

[0028] The three-dimensional coordinates of the UAV's flight start and end points are projected onto each key height layer to obtain the corresponding two-dimensional start and end points. Based on the grid resolution of the three-dimensional spatial model, the two-dimensional start and end points are mapped to grid nodes in the corresponding key height layer.

[0029] In each critical height layer, the grid node corresponding to the two-dimensional starting point is taken as the starting node, and the grid node corresponding to the two-dimensional ending point is taken as the target node;

[0030] Obtain a two-dimensional search space composed of free space grids and perform path search to obtain a two-dimensional path point sequence from the starting node to the target node.

[0031] In a preferred embodiment, the two-dimensional path point sequences of each key height layer are merged to obtain a two-dimensional path projection set, which is then extended along the height direction to a free space grid to construct a restricted three-dimensional search space. Furthermore, the restricted three-dimensional search space is expanded into a neighborhood to obtain an extended restricted three-dimensional search space, including:

[0032] The two-dimensional path point sequences of each key height layer are merged to obtain a two-dimensional path projection set;

[0033] Based on each two-dimensional coordinate point in the two-dimensional path projection set, the vertical direction is expanded layer by layer within the allowable flight altitude range. The grid cells in the corresponding altitude layer are mapped to three-dimensional grid points, and the three-dimensional grid points located in the free space grid are selected to construct the initial restricted three-dimensional search space.

[0034] Based on the initial restricted 3D search space, the neighborhood grid of each 3D grid point is expanded to include grid cells that are adjacent to the 3D grid points in the initial restricted 3D search space and satisfy the free space constraints, thus obtaining the expanded restricted 3D search space.

[0035] In a preferred embodiment, path searching is performed within an extended, restricted 3D search space to obtain an initial 3D path from the starting point to the ending point, including:

[0036] Within an extended, constrained 3D search space, a node-based 3D path search process is constructed. The neighboring 3D grid nodes of the current node are progressively expanded, and the actual path cost from the starting node to the current node is obtained for each candidate node.

[0037] The estimated cost is constructed based on the three-dimensional Euclidean distance between the current node and the target node, and a node evaluation function is formed by combining the actual path cost. Candidate nodes are sorted and selected to guide the path search toward the target node.

[0038] When the path search reaches the target node, the path is backtracked according to the parent-child relationship between each node to generate a three-dimensional path point sequence from the starting node to the target node, which serves as the initial three-dimensional path for the UAV in the urban low-altitude environment.

[0039] In a preferred embodiment, candidate nodes are sorted and selected to guide the path search toward the target node, including:

[0040] Based on the obstacle grid in the 3D spatial model, the spatial distance between each free space grid cell and its adjacent obstacle grid cells is obtained;

[0041] Each free space grid cell is assigned a corresponding risk value based on its spatial distance.

[0042] Candidate nodes are sorted based on risk cost, and nodes are selected in order of increasing comprehensive cost for expansion to complete the path search process.

[0043] In a preferred embodiment, the initial 3D path is subjected to bidirectional pruning, and intermediate nodes are inserted during the pruning process to obtain an optimized path, including:

[0044] The first round of pruning is performed on the 3D initial path point sequence. Starting from the starting node of the 3D initial path point sequence and moving towards the target node, the drivability of non-adjacent path points in the path is checked. When it is determined that two path points are straight-line connected and there are no obstacles blocking them, the intermediate redundant path points between the two path points are deleted to reduce the number of path inflection points.

[0045] After completing the first round of pruning of the three-dimensional initial path point sequence, when the distance between adjacent path points exceeds the preset distance threshold or the path segment does not meet the smoothness constraint, intermediate nodes are inserted in the corresponding path segment at preset intervals to refine the path.

[0046] After inserting intermediate nodes, pruning is performed again on the updated path point sequence to remove redundant nodes in the newly inserted nodes that meet the conditions of straight-line connectivity and no obstacles, thereby reducing the number of path points while ensuring path smoothness.

[0047] After completing the second pruning process, an optimized 3D path point sequence is obtained, which serves as the optimized path for the UAV.

[0048] In a preferred embodiment, the obstacle grid is expanded according to a safety distance to form a danger zone, and the danger zone is avoided during path search and optimization to obtain the final path, including:

[0049] Based on the safety distance, the obstacle grid in the 3D spatial model is expanded, and adjacent grid cells within the safety distance are marked as impassable grids, forming a danger zone.

[0050] During the path search process, dangerous areas are used as constraints, and only free space grid nodes that are not marked as dangerous areas are expanded to limit the path search range.

[0051] During the path optimization process, each path segment obtained through path search is detected. When a path point is located in a dangerous area or spatially intersects with a dangerous area, the corresponding path segment is adjusted so that the adjusted path avoids the dangerous area.

[0052] After successfully avoiding the danger zone, the final path is obtained.

[0053] And, a three-dimensional path planning terminal for urban low-altitude unmanned aerial vehicles, comprising:

[0054] One or more processors;

[0055] A storage device on which one or more programs are stored;

[0056] When one or more programs are executed by one or more processors, the one or more processors implement a three-dimensional path planning method for urban low-altitude drones.

[0057] The technical effects achieved by this invention are as follows:

[0058] This invention significantly reduces the size of the 3D search space, lowers search complexity, and improves planning speed by filtering key height layers and constructing a restricted 3D search space. It utilizes obstacle expansion and danger zone avoidance strategies to ensure that the UAV maintains a safe distance from obstacles, reducing collision risks. Bidirectional pruning and intermediate node insertion optimize the path, making the UAV's flight path smooth and continuous, which is conducive to actual flight control execution. Considering the free space distribution of multiple height layers, obstacle density, and complex urban terrain, it can generate feasible paths in high-density building areas and can be combined with various UAV mission scenarios. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method provided by the present invention. Detailed Implementation

[0060] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0061] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0062] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0063] Furthermore, the present invention will be described in detail with reference to the schematic diagrams. When describing the embodiments of the present invention in detail, the schematic diagrams are merely examples for ease of explanation and should not limit the scope of protection of the present invention.

[0064] Please see the appendix Figure 1 As shown, a three-dimensional path planning method for urban low-altitude unmanned aerial vehicles (UAVs) is provided, including:

[0065] S1. The urban low-altitude flight area is discretized into a three-dimensional raster to establish a three-dimensional spatial model, and the three-dimensional coordinates of the UAV's flight start and end points are obtained. The allowable flight altitude range and safe distance are set.

[0066] S2. Perform a layered traversal within the range from the starting point's altitude to the permissible flight altitude. Expand upon the region constructed by projecting the starting and ending points, and select altitude layers that meet preset conditions as key altitude layers:

[0067] S3. Project the starting point and the ending point onto each critical height layer, and perform path search in each critical height layer to obtain a two-dimensional path point sequence;

[0068] S4. Merge the two-dimensional path point sequences of each key height layer to obtain a two-dimensional path projection set, and extend it along the height direction to the free space grid to construct a restricted three-dimensional search space. Then, expand the neighborhood of the restricted three-dimensional search space to obtain an extended restricted three-dimensional search space.

[0069] S5. Perform path search within the extended restricted 3D search space to obtain the initial 3D path from the starting point to the ending point;

[0070] S6. Perform bidirectional pruning on the initial 3D path and insert intermediate nodes during the pruning process to obtain an optimized path;

[0071] S7. Expand the obstacle grid according to the safety distance to form a danger zone, and avoid the danger zone during the path search and optimization process to obtain the final path.

[0072] As described in steps S1 to S7 above, the urban low-altitude flight area is discretized using a three-dimensional rasterization process, and a three-dimensional spatial model is established. This is achieved by dividing the flight area into three-dimensional raster units with fixed spatial resolution in both horizontal and vertical directions. Each raster unit is labeled as an obstacle or free space, thus accurately representing the complex urban environment, including buildings, terrain, and airspace restrictions. The three-dimensional coordinates of the UAV's starting and ending points are mapped to the corresponding raster units. Combined with the UAV's flight performance parameters, airspace management constraints, and dimensional information, the permissible flight altitude range and safe distance from obstacles are determined. This is achieved by performing a layered traversal from the starting point's corresponding altitude to the permissible flight altitude range, and then projecting the model onto the base area constructed from the starting and ending points. The algorithm expands upon the candidate altitude layers by selecting key altitude layers that meet preset conditions. It analyzes the free space connectivity, obstacle density, and gradient changes of different altitude layers to eliminate unsuitable altitude layers, retaining only safe and efficient flight altitude layers. Layer-by-layer spatial feature analysis optimizes the search range, reducing computational complexity while ensuring path feasibility. A two-dimensional path search is performed within each key altitude layer, projecting the three-dimensional coordinates of the start and end points onto the two-dimensional altitude layer and mapping them as grid nodes. The path search employs an improved A* algorithm, combining actual path cost, estimated cost, and risk value. Dynamic weighting and node expansion strategies optimize search efficiency. By merging the two-dimensional paths from each altitude layer, a two-dimensional path projection set is obtained, and the path is then mapped along the altitude direction. Extending to a free-space grid, a restricted 3D search space is constructed, and further neighborhood expansion forms an extended restricted 3D search space. This extends the 2D path into a feasible traversable region in 3D space, limiting the spatial range of the 3D path search while ensuring path continuity along the altitude direction. This reduces the 3D search space, improves search efficiency, and preserves the possibility of flight at multiple altitudes. Based on the extended restricted 3D search space, 3D path search is performed. By comprehensively evaluating the actual path cost of each candidate node, the estimated cost to the target node, and the risk value, an initial 3D path is generated. This ensures that the path not only reaches the destination but also stays spatially away from obstacles and dangerous areas, guaranteeing flight safety and feasibility. Bidirectional pruning and intermediate node insertion are performed to optimize the path shape. Pruning detects the straight-line traversability between non-adjacent path points, removes redundant nodes, and reduces the number of inflection points. Intermediate node insertion is not directly used for path smoothing, but rather indirectly improves path smoothness by increasing the density of path points and providing more candidate connection conditions for bidirectional pruning. Hazard zones are created by expanding the obstacle grid at safe distances and are used as constraints in path search and optimization. During the path search phase, nodes within hazardous zones are avoided from expanding. During the optimization phase, adjustments are made to the intersections of path segments and hazardous zones to ensure the final path completely avoids hazardous zones. Potential collision risks are incorporated into path planning through spatial expansion and constraint application.To ensure the safety of UAVs flying in complex urban low-altitude environments, the final path is feasible, safe, smooth, and efficient. It provides optimal 3D flight routes while guaranteeing safety. Through key altitude layer selection and the construction of a constrained 3D search space, the size of the 3D search space is significantly reduced, search complexity is lowered, and planning speed is improved. Obstacle expansion and danger zone avoidance strategies ensure a safe distance between the UAV and obstacles, reducing collision risk. Two-way pruning and intermediate node insertion optimize the path, making the UAV flight path smooth and continuous, which is beneficial for actual flight control execution. Considering the free space distribution at multiple altitude layers, obstacle density, and complex urban terrain, it can generate feasible paths in high-density building areas and can be combined with various UAV mission scenarios (such as inspection, logistics, and photography).

[0073] In a preferred embodiment, the urban low-altitude flight area is discretized using a three-dimensional rasterization process to establish a three-dimensional spatial model. The three-dimensional coordinates of the UAV's flight start and end points are obtained, and the allowable flight altitude range and safe distance are set, including:

[0074] S101. The urban low-altitude flight area is discretized in three-dimensional space according to a preset spatial resolution along the horizontal and vertical directions to form multiple three-dimensional grid units.

[0075] S102. Obtain flight area data, which includes building data, terrain data and airspace restriction data. Based on the flight area data, obtain the spatial occupancy status of each three-dimensional grid cell. Three-dimensional grid cells that are occupied by obstacles or located in no-fly zones are marked as obstacle grids, and the rest are marked as free space grids. Construct a three-dimensional space model.

[0076] S103. Obtain UAV mission information, and based on the UAV mission information, obtain the three-dimensional coordinates of the UAV's flight start point and end point, and map the three-dimensional coordinates to the corresponding three-dimensional grid cells according to the preset spatial resolution.

[0077] S104. Obtain the UAV flight performance parameters and airspace management constraints, and obtain the permissible flight altitude range based on the UAV flight performance parameters and airspace management constraints;

[0078] S105. Obtain the external dimensions of the UAV and the environmental complexity information of the flight area, and combine them with preset flight safety rules to obtain the safe distance between the UAV and obstacles.

[0079] As described in steps S101 to S105 above, the urban low-altitude flight area is discretized into a three-dimensional raster, dividing the continuous three-dimensional space into several fixed-size cubic units (raster units) in the horizontal and vertical directions. The preset spatial resolution can be determined by combining the UAV's flight accuracy, perception capabilities, and path planning accuracy requirements. For example, smaller raster units are selected in densely built-up areas to ensure detail capture, while in open areas, the raster units can be appropriately enlarged to reduce computational load. Each raster unit can be determined by its center coordinates and side length in three-dimensional space, forming an overall grid structure. Building data, terrain data, and airspace restriction data of the flight area are acquired. These data are compared with the three-dimensional raster to determine whether each raster unit is occupied by obstacles or located in a no-fly zone, and it is marked as an obstacle raster. Unoccupied raster units are marked as free space raster units. All raster units and their occupancy status form a three-dimensional spatial model. The three-dimensional coordinates of the starting point and ending point are obtained through UAV mission information, which can be obtained through the mission planning system or user input, including the latitude and longitude of the geographical location, elevation data, and the UAV's own data. The initial flight altitude of the drone is used to acquire three-dimensional coordinates, which are then mapped to corresponding grid cells according to a preset spatial resolution. This transforms continuous coordinates into nodes in discrete space, facilitating path search within the three-dimensional grid model. The drone's flight performance parameters (such as maximum flight altitude, rate of climb, and turning ability) and airspace management constraints (such as no-fly zones, altitude restrictions, and flight density restrictions) are used to determine the permissible flight altitude range, ensuring that the drone operates within legal and safe airspace. The drone's external dimensions (such as fuselage length and wingspan) and environmental complexity information (such as building density and obstacle distribution density) are combined with preset flight safety rules to determine the minimum safe distance between the drone and obstacles. These rules can be determined based on national aviation regulations, drone manufacturer recommendations, and actual flight test data, providing a high-precision three-dimensional environmental representation. This enables the drone to perceive the spatial distribution of obstacles, buildings, and no-fly zones, providing a reliable basis for path planning and ensuring flight safety. By setting flight altitude restrictions and safe distances, the planned path is kept within legal limits, avoiding collisions with obstacles.

[0080] In a preferred embodiment, a layered traversal is performed within the range from the altitude corresponding to the starting point to the allowable flight altitude. This is expanded based on the region constructed using the projections of the starting and ending points, and altitude layers that meet preset conditions are selected as key altitude layers, including:

[0081] S201. Within the range from the altitude corresponding to the starting point of the UAV flight to the allowable flight altitude, the three-dimensional spatial model is divided into layers according to the preset altitude interval to obtain multiple candidate altitude layers.

[0082] S202. In each candidate altitude layer, project the three-dimensional coordinates of the UAV's flight start and end points to the corresponding altitude layer to obtain the projection position corresponding to each altitude layer.

[0083] S203. Obtain the basic region based on the projection position, and spatially expand the basic region according to a preset expansion distance to obtain the candidate search region;

[0084] S204. Within the candidate search area, perform connected component labeling on the free space raster to obtain free space connected regions, and obtain the corresponding number of connected regions and the area of ​​each connected region.

[0085] S205. Obtain the number of obstacle grids in the candidate search area, obtain the obstacle grid density based on the number of obstacle grids per unit volume, and obtain the obstacle distribution gradient based on the density change between spatially adjacent grids.

[0086] S206. Obtain the difference in the number of connected regions, the difference in the area of ​​connected regions, the difference in the density of obstacle grids, and the difference in the distribution gradient between the current candidate height layer and the adjacent candidate height layers, and compare them with the preset change threshold corresponding to each difference. When any difference exceeds the corresponding preset change threshold, the corresponding candidate height layer is determined as the key height layer.

[0087] As described in steps S201 to S206 above, a layered traversal is performed from the UAV's starting altitude to the allowable flight altitude range. To obtain candidate altitude layers, a preset altitude interval needs to be determined. This altitude interval can be set based on the UAV's vertical maneuverability, perception resolution, and the complexity of altitude changes in the flight environment. For example, in densely built-up areas or areas with multiple layers of obstacles, a smaller altitude interval can be selected to capture altitude differences. In open airspace, the altitude interval can be appropriately increased to reduce computational load. Dividing the three-dimensional spatial model vertically according to this altitude interval yields multiple parallel candidate altitude layers. Each altitude layer is equivalent to a two-dimensional projection of space at a specific altitude, thus forming a series of reference planes for path search. The three-dimensional coordinates of the drone's starting and ending points are projected onto each candidate altitude layer to obtain the two-dimensional projection position of each altitude layer. The projection method replaces the altitude component of the three-dimensional coordinates with the altitude value of the candidate altitude layer while retaining the horizontal coordinates, thus obtaining the starting and ending point projections at that altitude layer. A base region is constructed centered on the projection position. A rectangular or circular coverage area is generated near the line connecting the starting and ending points of the two-dimensional projection to ensure that it includes potential feasible paths. The base region is then expanded horizontally and vertically according to a preset expansion distance. This expansion distance can be determined based on the drone's maneuverability, safety margin, and obstacle distribution to obtain the candidate search area. Connectivity labeling is performed on the free space grid within the candidate search area, i.e., for each The free-space grid performs a neighborhood traversal, grouping interconnected and continuous free grids into the same connected region. This process identifies the continuity of traversable space within a region and counts the number and area of ​​each connected region, which can be used to evaluate the traversability of that height level. Larger connected regions generally indicate that that height level is more suitable for path planning. The number of obstacle grids within the candidate search area is obtained, and the obstacle grid density is calculated based on the number of obstacle grids per unit volume, reflecting the obstacle distribution at that height level. By comparing the difference in obstacle grid density between spatially adjacent grids, density variation is obtained, and the obstacle distribution gradient is acquired, reflecting the spatial variation between dense and sparse obstacle areas within the height level. This is crucial for determining traversability during path planning. The passageway and high-risk avoidance areas are valuable references. The current candidate altitude layer is compared with adjacent altitude layers to obtain differences in the number of connected regions, the area of ​​connected regions, the density of obstacle grids, and the distribution gradient. These differences are then compared with corresponding preset change thresholds. These thresholds can be set based on UAV safety requirements, environmental complexity, and historical path planning experience. When any difference exceeds a threshold, it indicates a significant change in the spatial structure or obstacle distribution of that altitude layer, potentially forming a new feasible path or risk area. Therefore, this is identified as a critical altitude layer, reducing the computational load of 3D path search while ensuring that the planned path covers key spatial features. Path search is performed only at critical altitude layers, rather than across the entire altitude range, saving computational resources.The critical height layer covers areas where obstacle distribution varies significantly in space, ensuring that path planning does not overlook potential feasible paths.

[0088] In a preferred embodiment, the start and end points are projected onto each critical height layer, and a path search is performed at each critical height layer to obtain a two-dimensional path point sequence, including:

[0089] S301. Project the three-dimensional coordinates of the UAV's flight start and end points to each key height layer to obtain the corresponding two-dimensional start and two-dimensional end points. Based on the grid resolution of the three-dimensional spatial model, map the two-dimensional start and two-dimensional end points to the grid nodes in the corresponding key height layer.

[0090] S302. In each critical height layer, the grid node corresponding to the two-dimensional starting point is taken as the starting node, and the grid node corresponding to the two-dimensional ending point is taken as the target node.

[0091] S303. Obtain the two-dimensional search space composed of free space grid and perform path search to obtain the two-dimensional path point sequence from the starting node to the target node.

[0092] As described in steps S301 to S303 above, the three-dimensional start and end points of the UAV are projected onto each key height layer. The projection method keeps the horizontal coordinates (i.e., x and y coordinates) unchanged and replaces the height values ​​of the three-dimensional coordinates with the height values ​​of the key height layers, thereby obtaining the start and end points of the two-dimensional coordinates. Utilizing the grid resolution of the three-dimensional spatial model, the two-dimensional coordinates are mapped to the corresponding two-dimensional grid nodes. The mapping method typically involves dividing the two-dimensional coordinates by the horizontal resolution and rounding to determine their index position in the two-dimensional grid. Each key height layer forms two-dimensional start and end point nodes that can be used for searching. The mapped two-dimensional start node is used as the starting node for path searching. The two-dimensional endpoint node serves as the target node. This node definition not only identifies the start and end positions of the search but also serves as the benchmark for node evaluation and expansion in the path search algorithm. The two-dimensional search space is obtained, which is a two-dimensional grid composed of all free-space grids in the critical height layer. Free-space grids are grid cells not occupied by obstacles. By traversing the two-dimensional grid matrix of this height layer, grid cells not marked as obstacles are collected, thus obtaining the two-dimensional search space. Path search is performed in this two-dimensional search space, using an improved A* algorithm. Algorithm features include an improved node evaluation function (the evaluation function of each candidate node is determined by the actual path cost (from the starting point to the current node)). The path consists of the cumulative cost and the estimated cost (the predicted cost from the current node to the destination), and introduces a dynamic weight coefficient. The dynamic weight coefficient gradually increases with the search depth, making the search more biased towards global exploration in the early stage and enhancing convergence in the later stage, thereby improving search efficiency while ensuring path reachability. The path generation strategy is optimized (during the node expansion process, neighboring nodes are selected based on the movement direction of the current node and its parent node, and only neighboring nodes whose angle with the current movement direction is less than a preset threshold are expanded. This directional constraint can avoid invalid node expansion, reduce computation, and improve the smoothness of the generated path). The path generation strategy is optimized by sorting and selecting candidate nodes. Based on a node evaluation function, the system gradually expands from the starting node to the target node. As the target node is expanded, a sequence of path points from the two-dimensional starting point to the two-dimensional ending point can be generated by backtracking. Each path point corresponds to a two-dimensional grid node, forming a two-dimensional path at the key altitude level. By decomposing the three-dimensional problem into a two-dimensional search problem, the search space at each key altitude level is significantly reduced, thereby improving the path search efficiency. Dynamic weight coefficients and turning constraints reduce unnecessary inflection points, generating smoother two-dimensional paths. This provides a good foundation for three-dimensional path synthesis and optimization, and can be flexibly applied to urban low-altitude environments with different altitude levels and densities, adapting to complex building distributions and flight mission requirements.

[0093] In a preferred embodiment, the two-dimensional path point sequences of each key height layer are merged to obtain a two-dimensional path projection set, which is then extended along the height direction to a free space grid to construct a restricted three-dimensional search space. Furthermore, the restricted three-dimensional search space is expanded into a neighborhood to obtain an extended restricted three-dimensional search space, including:

[0094] S401. Merge the two-dimensional path point sequences of each critical height layer to obtain a two-dimensional path projection set;

[0095] S402. Based on each two-dimensional coordinate point in the two-dimensional path projection set, expand layer by layer in the vertical direction within the allowable flight altitude range, map the grid cells in the corresponding altitude layer to three-dimensional grid points, and filter the three-dimensional grid points located in the free space grid to construct the initial restricted three-dimensional search space.

[0096] S403. Based on the initial restricted three-dimensional search space, the neighborhood grid of each three-dimensional grid point is expanded to include grid cells that are adjacent to the three-dimensional grid points in the initial restricted three-dimensional search space and satisfy the free space constraint conditions, thus obtaining the expanded restricted three-dimensional search space.

[0097] As described in steps S401 to S403 above, all two-dimensional path points generated in all altitude layers are projected onto the same plane according to their horizontal coordinates (x, y). Duplicate or adjacent two-dimensional coordinate points are deduplicated and recorded uniformly to form a two-dimensional path projection set. This set not only contains the overall distribution information of paths in each key altitude layer, but also preserves the continuity of paths along the horizontal direction. Redundant information between two-dimensional path point sequences is eliminated, while key passage areas are preserved, making the two-dimensional path projection set the basis for constraining the three-dimensional path search. Based on each two-dimensional coordinate point in the two-dimensional path projection set, the vertical direction is expanded layer by layer within the allowable flight altitude range. Starting from the lowest allowable flight altitude, every preset vertical resolution, the two-dimensional coordinate points are mapped to the three-dimensional grid cells of the corresponding altitude layer to form three-dimensional grid points. The mapped three-dimensional grid points are filtered, retaining only points located within the free space grid, i.e., excluding points occupied by obstacles or belonging to no-fly zones. A grid point, along with a two-dimensional path, forms a continuous, passable three-dimensional region, thus constructing an initial constrained three-dimensional search space. This initial constrained space represents the passable extension of the two-dimensional path projection in the height direction, used to constrain the search range in the three-dimensional path search and avoid the expansion of invalid or dangerous areas. For each grid point in the initial three-dimensional grid space, its six or twenty-six neighboring grids are checked (selected according to the six-neighbor or twenty-six-neighbor strategy). Neighboring grids that meet the free space constraints are included in the search space. Free space constraints typically include that the grid is not occupied by obstacles, does not exceed the allowable flight altitude range, and maintains a safe distance from the surrounding environment. After the neighborhood expansion, the initial constrained three-dimensional search space is expanded to form an extended constrained three-dimensional search space, so that the three-dimensional search space not only covers the direct extension of the two-dimensional path projection but also includes the surrounding passable auxiliary grids, providing more flexibility for path search and improving path feasibility.

[0098] In a preferred embodiment, path search is performed within an extended restricted 3D search space to obtain an initial 3D path from the starting point to the ending point, including:

[0099] S501. Within the extended restricted 3D search space, construct a 3D path search process based on node expansion, progressively expand the 3D grid nodes in the neighborhood of the current node, and obtain the actual path cost of each candidate node from the starting node to the current node.

[0100] S502. Construct an estimated cost based on the three-dimensional Euclidean distance between the current node and the target node, and combine it with the actual path cost to form a node evaluation function. Sort and select candidate nodes to guide the path search toward the target node.

[0101] S503. When the path search reaches the target node, the path backtracking is performed according to the parent-child relationship between each node to generate a three-dimensional path point sequence from the starting node to the target node, which serves as the three-dimensional initial path of the UAV in the urban low-altitude environment.

[0102] As described in steps S501 to S503 above, the search space is discretized into a three-dimensional grid, with each grid cell corresponding to a node. Starting from the starting node, the search space is expanded layer by layer to the neighboring nodes (usually six or twenty-six neighbors). During the expansion process, the expansion operation is only performed on neighboring nodes that meet the free space constraints and preset passage conditions. For example, the neighboring node is not occupied by obstacles, has not entered a no-fly zone, and maintains a necessary safe distance. For each candidate node, the actual path cost from the starting node to that node needs to be obtained. The actual path cost can be obtained by accumulating the Euclidean distance or weighted distance along the path through the nodes. At the same time, it can be weighted by combining factors such as movement direction, turning change, and altitude change to truly reflect the free space constraints. The cost of human-machine movement in complex 3D space, improving path search efficiency and guiding the search towards the target node, requires constructing an estimated cost for each candidate node. The estimated cost is typically the 3D Euclidean distance from the current node to the target node, reflecting the shortest spatial distance from the node to the destination. During path search, the node evaluation function is composed of both the actual path cost and the estimated cost. In the 2D critical height layer path planning stage, to improve the guidance capability and search efficiency, a dynamic weight adjustment mechanism is introduced for the estimated cost, giving it different degrees of influence at different search stages. This achieves an adaptive balance between global search capability and target convergence speed. However, when performing path search in a 3D constrained search space, due to the search... With the spatial constraints effectively reduced, the path search range is significantly narrowed. Therefore, a node evaluation function is constructed based on the estimated cost of three-dimensional Euclidean distance. A fixed weighting method is used to combine the actual path cost and the estimated cost to ensure the stability and computational efficiency of the search process. Through a phased heuristic function design, targeted optimization of the search strategy is achieved at different path planning stages. This improves the overall path planning efficiency and path quality while ensuring path reachability. The dynamic weight heuristic function only applies to the two-dimensional path planning stage, while a fixed weight heuristic function is used in the three-dimensional path search stage to avoid the instability introduced by dynamic weights in high-dimensional space. For example, when the search depth is shallow, the weight of the estimated cost can be increased to accelerate the search. The search process involves convergence, and as the search depth increases, the weight of the actual path cost can be increased to optimize path smoothness. A node evaluation function is used to rank candidate nodes, prioritizing the expansion of nodes with lower overall costs. This allows the path search to quickly move towards the destination within a limited space and effectively avoids the expansion of invalid nodes, thus improving search efficiency. Once the path search reaches the target node, path backtracking is performed based on the parent-child relationships between each node. During the expansion process, each node records its parent node, indicating which neighboring node it was expanded from. The path backtracking process starts from the target node and traces upwards along the parent nodes until it returns to the starting node. During the backtracking process, the 3D coordinates of each node are recorded sequentially, generating a continuous sequence of 3D path points.The resulting 3D initial path maintains drivability from the starting point to the ending point, reflecting the optimal or near-optimal path selection along the free space in the search space. This provides a basic path for the safe flight of UAVs in urban low-altitude environments, ensuring that the path only passes through drivable areas, reducing collision risks, and providing safety guarantees for UAV low-altitude flight.

[0103] In a preferred embodiment, the candidate nodes are sorted and selected to guide the path search toward the target node, including:

[0104] S50201. Based on the obstacle grid in the 3D spatial model, obtain the spatial distance between each free space grid cell and its adjacent obstacle grid cells;

[0105] S50202. Assign a corresponding risk value to each free space grid cell based on the spatial distance;

[0106] S50203. Based on the risk cost value, the candidate nodes are sorted, and nodes are selected in order of increasing comprehensive cost to expand the path search process.

[0107] As described in steps S50201 to S50203 above, based on the established three-dimensional spatial model, for each free space grid cell, obstacle grids in its neighborhood are searched, and the shortest Euclidean distance from the center point of the free grid to the center point of the neighboring obstacle grid is measured. This yields the safety distance of each free grid relative to surrounding obstacles. This spatial distance not only reflects the relative position of the flight path and obstacles, but also assigns a risk value to each free space grid based on the spatial distance. The risk value typically varies with the distance between the free grid and the obstacle; the closer the distance, the higher the risk value, and the farther the distance, the lower the risk value. This can be mapped using an inverse distance function or an exponential function. For example, the risk value can be set as the reciprocal of the minimum obstacle distance or in an exponentially decaying form, thus forming a continuous risk value. The risk cost distribution, which aggregates the risk cost of each free grid, generates a complete 3D spatial risk cost distribution. This distribution reflects the potential danger level at different locations within the entire constrained search space. All candidate nodes are sorted according to their risk cost, and nodes are expanded sequentially from smallest to largest comprehensive cost. The lower the comprehensive cost of a node, the shorter and safer its corresponding path. Therefore, prioritizing the expansion of these nodes can accelerate search convergence, prevent paths from approaching obstacles or entering potentially dangerous areas, and maintain the global optimality or near-optimality of the path search. The search process controls computational complexity, improves the safety and flightability of the generated path, reduces unnecessary node expansion, lowers computational load, accelerates path search speed, and enables path planning to adapt to different environmental complexities, achieving efficient and reliable 3D path generation.

[0108] In a preferred embodiment, the initial 3D path is subjected to bidirectional pruning, and intermediate nodes are inserted during the pruning process to obtain an optimized path, including:

[0109] S601. Perform the first round of pruning on the three-dimensional initial path point sequence. Starting from the starting node of the three-dimensional initial path point sequence and moving towards the target node, perform drivability detection on non-adjacent path points in the path. When it is determined that two path points are straight-line connected and there are no obstacles blocking them, delete the intermediate redundant path points between the two path points to reduce the number of path inflection points.

[0110] S602. After completing the first round of pruning of the three-dimensional initial path point sequence, when the distance between adjacent path points exceeds the preset distance threshold or the path segment does not meet the smoothness constraint, intermediate nodes are inserted in the corresponding path segment at preset intervals to refine the path.

[0111] S603. After inserting intermediate nodes, perform pruning again on the updated path point sequence to delete redundant nodes in the newly inserted nodes that meet the conditions of straight-line connectivity and no obstacles, thereby reducing the number of path points while ensuring path smoothness.

[0112] S604. After completing the second pruning process, the optimized three-dimensional path point sequence is obtained, which serves as the optimized path for the UAV.

[0113] As described in steps S601 to S604 above, during the first round of pruning of the 3D initial path point sequence, unidirectional pruning is performed from the starting node to the target node. The current node is taken as the reference node, and subsequent path points that are not adjacent to it are selected sequentially along the path direction as candidate connection nodes. The traversability of the connection between the reference node and the candidate connection nodes is checked. The spatial connection between the two nodes is discretely sampled according to a preset step size, or by traversing the 3D grid cells traversed by the connection, it is determined whether the connection crosses the obstacle grid or enters a restricted area. At the same time, it is determined whether the connection meets the preset safety distance requirements between the connection and the surrounding obstacles. When the detection result shows that the connection is in free space and meets the requirements of the surrounding obstacles, the connection is considered safe. When the safety distance constraint is met, the two nodes are considered to be directly connected. All intermediate path points between these two nodes are considered redundant and deleted, thus compressing the original multi-segment polyline path into a single straight path. The updated connecting node is used as the new reference node, and the same process continues until no more distant nodes satisfying the direct connection condition can be found. This significantly reduces redundant nodes in the path while ensuring path traversability, resulting in a more concise path point sequence. After the first round of pruning, the resulting path point sequence is traversed, and the spatial relationship between adjacent path points is checked sequentially. When the distance between adjacent path points exceeds a preset distance threshold, or the path segment is in the direction of... If the changes do not meet the preset smoothness constraints (e.g., the transitions between adjacent path segments are too abrupt), then the path segment is determined to need refinement. For path segments requiring refinement, the number of intermediate nodes to be inserted is determined based on the spatial length of the path segment and the preset node spacing requirements. Multiple intermediate nodes are generated by interpolation at uniform intervals on the path segment. The generated intermediate nodes are located on the spatial connection line of the original path segment and maintain a continuous distribution in three-dimensional space. During the insertion of intermediate nodes, the validity of each newly generated node is verified. The verification includes determining whether the node is located within the free space grid, whether it meets the safe distance requirements between it and surrounding obstacles, and whether it is within the allowable flight altitude range. Only when all these conditions are met is the node added to the path point sequence. This effectively avoids excessively long or abrupt path segments, making the spatial distribution of the path more uniform and smooth. After inserting intermediate nodes, the updated path point sequence is pruned again, similar to the first round of pruning. Reverse pruning is performed from the target node to the starting node, further compressing the path point sequence. Any node in the updated path point sequence is used as the reference node, and non-adjacent nodes are selected forward as candidate connection nodes. The drivability of the spatial connection between two nodes is then checked using the same method as the first round of pruning, including discrete sampling or grid traversal of the connection path.The system then determines if there are any obstacles or violations of safety distance constraints. If the detection results indicate that two nodes can be directly connected, all intermediate nodes between them are deleted, including nodes inserted in the previous step and unnecessary nodes from the original nodes. This eliminates redundant nodes introduced by interpolation refinement while retaining key nodes that are practically significant for path smoothness. Through further pruning, the path maintains smoothness while further reducing the number of nodes, achieving further optimization of the path structure. After this second pruning process, the final path point sequence is organized and arranged in order from the starting node to the target node, forming a continuous three-dimensional path point sequence. Each adjacent node pair in this path point sequence is then finally verified. Each segment of the path must meet the following requirements: the path segment does not cross any obstacle grid in space; the path segment always stays within the allowable flight altitude range; the path segment maintains a preset safe distance from surrounding obstacles; and the overall path meets preset smoothness requirements without sharp turns. Based on these requirements, the sequence of path points is used as the final optimized path for the UAV. This optimized path is not only spatially continuous and feasible but also structurally simpler and smoother, and can be directly used for UAV path tracking and flight control. This forms a processing mechanism that first performs redundancy compression, then refines and compensates, and finally performs further compression and optimization, achieving a balance between the number of nodes, spatial smoothness, and flight safety. This significantly improves the practical feasibility of 3D paths in complex low-altitude urban environments.

[0114] In a preferred embodiment, the obstacle grid is expanded according to a safety distance to form a danger zone, and the danger zone is avoided during path search and optimization to obtain the final path, including:

[0115] S701. Based on the safety distance, the obstacle grid in the three-dimensional space model is expanded, and adjacent grid cells within the safety distance are marked as impassable grids, forming a danger zone.

[0116] S702. During the path search process, dangerous areas are used as constraints, and only free space grid nodes that are not marked as dangerous areas are expanded to limit the path search range.

[0117] S703. During the path optimization process, each path segment obtained through path search is detected. When a path point is located in a dangerous area or spatially intersects with a dangerous area, the corresponding path segment is adjusted so that the adjusted path avoids the dangerous area.

[0118] S704. After completing the danger zone avoidance, the final path is obtained.

[0119] As described in steps S701 to S704 above, after completing the 3D path search and optimization, based on a preset flight safety distance, the obstacle grid in the 3D spatial model is expanded. The expansion process extends horizontally and vertically along the 3D space, centered on each obstacle grid, covering neighboring grid cells with a radius equal to the safety distance. Grid cells within the expanded range are marked as impassable danger grids, constructing a complete danger zone. This restricts the entire space within which the UAV might approach obstacles due to its size, flight attitude deviation, or control error, thus eliminating potential collision risks in advance during path planning. During the path search process, candidate path nodes are only allowed to select free-space grids not marked as danger zones as expansion targets. The path search process automatically avoids danger zones. In the 3D path generation stage, the UAV's planned path will not traverse the expanded obstacle space, thereby reducing the risk of collisions during flight. Combined with the prior risk cost mechanism, the A* algorithm can prioritize safe paths away from danger zones while maintaining path accessibility and efficiency. In the path optimization stage, the path... The 3D path obtained through path search (including the initial path and the optimized path after pruning and node insertion) undergoes hazard area detection. For each path segment and its continuous nodes, it is determined whether they are located within a hazard area or intersect with a hazard area. If an intersection or close proximity is found, the path segment is adjusted by moving nodes, reconstructing the path segment, or inserting intermediate nodes to avoid the hazard area while maintaining the continuity and smoothness of the path. During the adjustment process, it is necessary to ensure that the path remains within a passable free space grid and meets flight performance parameters and steering constraints, thereby ensuring that the UAV can safely execute the planned path. After completing hazard area avoidance and path adjustment, a complete final path is obtained. This path inherits the smoothness and flight efficiency of the previously optimized path and strictly avoids hazard areas around obstacles, providing safety assurance for the autonomous flight of UAVs in complex low-altitude urban environments. It significantly reduces the collision risk during UAV flight, and the flight path is more in line with actual flight safety requirements. At the same time, it can maintain the optimization of path length and flight efficiency while ensuring safety, achieving a balance between safety and efficiency.

[0120] And, a three-dimensional path planning terminal for urban low-altitude unmanned aerial vehicles, comprising:

[0121] One or more processors;

[0122] A storage device on which one or more programs are stored;

[0123] When one or more programs are executed by one or more processors, the one or more processors implement a three-dimensional path planning method for urban low-altitude drones.

[0124] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A three-dimensional path planning method for urban low-altitude unmanned aerial vehicles (UAVs), characterized in that, include: The urban low-altitude flight area is discretized into a three-dimensional raster to establish a three-dimensional spatial model, and the three-dimensional coordinates of the UAV's flight start and end points are obtained to set the allowable flight altitude range and safe distance. A layered traversal is performed within the range from the starting point altitude to the allowed flight altitude. Based on the region constructed by projecting the starting point and the ending point, the region is expanded, and the height layer that meets the preset conditions is selected as the key height layer. Specifically, within the range from the starting point altitude of the UAV to the allowed flight altitude, the three-dimensional spatial model is divided into layers according to the preset height interval to obtain multiple candidate height layers. The difference in the number of connected regions, the difference in the area of ​​connected regions, the difference in the density of obstacle grids, and the difference in the distribution gradient between the current candidate height layer and the adjacent candidate height layers are obtained and compared with the preset change threshold corresponding to each difference. When any difference exceeds the corresponding preset change threshold, the corresponding candidate height layer is determined as the key height layer. Project the start and end points onto each critical height layer, and perform path search at each critical height layer to obtain a two-dimensional path point sequence. The two-dimensional path point sequences of each key height layer are merged to obtain a two-dimensional path projection set, which is then extended to the free space grid along the height direction to construct a restricted three-dimensional search space. The restricted three-dimensional search space is then expanded to a neighborhood to obtain an extended restricted three-dimensional search space. Perform path search within the extended restricted 3D search space to obtain the initial 3D path from the starting point to the ending point; The initial 3D path is pruned bidirectionally, and intermediate nodes are inserted during the pruning process to obtain an optimized path; The obstacle grid is expanded based on the safety distance to form a danger zone, and the danger zone is avoided during the path search and optimization process to obtain the final path.

2. The three-dimensional path planning method for urban low-altitude unmanned aerial vehicles according to claim 1, characterized in that, The urban low-altitude flight area is discretized using a three-dimensional rasterization process to establish a three-dimensional spatial model. The three-dimensional coordinates of the UAV's flight start and end points are obtained, and the permissible flight altitude range and safe distance are set, including: The urban low-altitude flight area is discretized in three-dimensional space according to a preset spatial resolution along the horizontal and vertical directions to form multiple three-dimensional grid units. The system acquires flight area data, which includes building data, terrain data, and airspace restriction data. Based on the flight area data, it obtains the spatial occupancy status of each three-dimensional grid cell. Three-dimensional grid cells that are occupied by obstacles or located in no-fly zones are marked as obstacle grids, and the rest are marked as free space grids. A three-dimensional spatial model is then constructed. Acquire drone mission information, and based on the drone mission information, obtain the three-dimensional coordinates of the drone's flight start and end points, and map the three-dimensional coordinates to the corresponding three-dimensional grid cells according to the preset spatial resolution; Obtain the UAV flight performance parameters and airspace management constraints, and obtain the permissible flight altitude range based on the UAV flight performance parameters and airspace management constraints; The system acquires the drone's external dimensions and the environmental complexity of the flight area, and combines this with preset flight safety rules to determine the safe distance between the drone and obstacles.

3. The three-dimensional path planning method for urban low-altitude unmanned aerial vehicles according to claim 1, characterized in that, A layered traversal is performed within the range from the starting point altitude to the permissible flight altitude. Based on the area constructed using the projections of the starting and ending points, the region is expanded, and altitude layers that meet preset conditions are selected as key altitude layers, including: Within the range from the altitude corresponding to the drone's flight start point to the allowable flight altitude, the three-dimensional spatial model is divided into layers according to a preset altitude interval to obtain multiple candidate altitude layers; In each candidate altitude layer, the three-dimensional coordinates of the UAV's flight start and end points are projected onto the corresponding altitude layer to obtain the projection position for each altitude layer; The basic region is obtained based on the projection position, and the basic region is spatially expanded according to the preset expansion distance to obtain the candidate search region. The basic region is a rectangular or circular coverage area generated near the line connecting the start and end points of the two-dimensional projection, with the projection position as the center. Within the candidate search area, perform connected component labeling on the free space raster to obtain free space connected regions, and obtain the corresponding number of connected regions and the area of ​​each connected region; The number of obstacle grids in the candidate search area is obtained, and the obstacle grid density is obtained based on the number of obstacle grids per unit volume. The obstacle distribution gradient is obtained based on the density change between spatially adjacent grids. The difference in the number of connected regions, the difference in the area of ​​connected regions, the difference in the density of obstacle grids, and the difference in the distribution gradient between the current candidate height layer and the adjacent candidate height layers are obtained and compared with the preset change threshold corresponding to each difference. When any difference exceeds the corresponding preset change threshold, the corresponding candidate height layer is determined as the critical height layer.

4. The three-dimensional path planning method for urban low-altitude UAVs according to claim 1, characterized in that, Projecting the start and end points onto each critical height level, and performing path search at each critical height level, yields a two-dimensional path point sequence, including: The three-dimensional coordinates of the UAV's flight start and end points are projected onto each key height layer to obtain the corresponding two-dimensional start and end points. Based on the grid resolution of the three-dimensional spatial model, the two-dimensional start and end points are mapped to grid nodes in the corresponding key height layer. In each critical height layer, the grid node corresponding to the two-dimensional starting point is taken as the starting node, and the grid node corresponding to the two-dimensional ending point is taken as the target node; Obtain a two-dimensional search space composed of free space grids and perform path search to obtain a two-dimensional path point sequence from the starting node to the target node.

5. The three-dimensional path planning method for urban low-altitude unmanned aerial vehicles according to claim 1, characterized in that, The two-dimensional path point sequences of each key height layer are merged to obtain a two-dimensional path projection set, which is then extended along the height direction to the free space grid to construct a restricted three-dimensional search space. Furthermore, the restricted three-dimensional search space is expanded by neighborhood expansion to obtain an extended restricted three-dimensional search space, including: The two-dimensional path point sequences of each key height layer are merged to obtain a two-dimensional path projection set; Based on each two-dimensional coordinate point in the two-dimensional path projection set, the vertical direction is expanded layer by layer within the allowable flight altitude range. The grid cells in the corresponding altitude layer are mapped to three-dimensional grid points, and the three-dimensional grid points located in the free space grid are selected to construct the initial restricted three-dimensional search space. Based on the initial restricted 3D search space, the neighborhood grid of each 3D grid point is expanded to include grid cells that are adjacent to the 3D grid points in the initial restricted 3D search space and satisfy the free space constraints, thus obtaining the expanded restricted 3D search space.

6. The three-dimensional path planning method for urban low-altitude unmanned aerial vehicles according to claim 1, characterized in that, Perform path search within the extended restricted 3D search space to obtain the initial 3D path from the starting point to the ending point, including: Within an extended, constrained 3D search space, a node-based 3D path search process is constructed. The neighboring 3D grid nodes of the current node are progressively expanded, and the actual path cost from the starting node to the current node is obtained for each candidate node. The estimated cost is constructed based on the three-dimensional Euclidean distance between the current node and the target node, and a node evaluation function is formed by combining the actual path cost. Candidate nodes are sorted and selected to guide the path search toward the target node. When the path search reaches the target node, the path is backtracked according to the parent-child relationship between each node to generate a three-dimensional path point sequence from the starting node to the target node, which serves as the initial three-dimensional path for the UAV in the urban low-altitude environment.

7. The three-dimensional path planning method for urban low-altitude unmanned aerial vehicles according to claim 6, characterized in that, Sort and select candidate nodes to guide the path search toward the target node, including: Based on the obstacle grid in the 3D spatial model, the spatial distance between each free space grid cell and its adjacent obstacle grid cells is obtained; Each free space grid cell is assigned a corresponding risk value based on its spatial distance. Candidate nodes are sorted based on risk cost, and nodes are selected in order of increasing comprehensive cost for expansion to complete the path search process.

8. The three-dimensional path planning method for urban low-altitude unmanned aerial vehicles according to claim 1, characterized in that, The initial 3D path is pruned bidirectionally, and intermediate nodes are inserted during the pruning process to obtain an optimized path, including: The first round of pruning is performed on the 3D initial path point sequence. Starting from the starting node of the 3D initial path point sequence and moving towards the target node, the drivability of non-adjacent path points in the path is checked. When it is determined that two path points are straight-line connected and there are no obstacles blocking them, the intermediate redundant path points between the two path points are deleted to reduce the number of path inflection points. After completing the first round of pruning of the three-dimensional initial path point sequence, when the distance between adjacent path points exceeds the preset distance threshold or the path segment does not meet the smoothness constraint, intermediate nodes are inserted in the corresponding path segment at preset intervals to refine the path. After inserting intermediate nodes, pruning is performed again on the updated path point sequence to remove redundant nodes in the newly inserted nodes that meet the conditions of straight-line connectivity and no obstacles, thereby reducing the number of path points while ensuring path smoothness. After completing the second pruning process, an optimized 3D path point sequence is obtained, which serves as the optimized path for the UAV.

9. The three-dimensional path planning method for urban low-altitude unmanned aerial vehicles according to claim 1, characterized in that, The obstacle grid is expanded based on a safety distance to form a danger zone. This danger zone is then avoided during path search and optimization to obtain the final path, which includes: Based on the safety distance, the obstacle grid in the 3D spatial model is expanded, and adjacent grid cells within the safety distance are marked as impassable grids, forming a danger zone. During the path search process, dangerous areas are used as constraints, and only free space grid nodes that are not marked as dangerous areas are expanded to limit the path search range. During the path optimization process, each path segment obtained through path search is detected. When a path point is located in a dangerous area or spatially intersects with a dangerous area, the corresponding path segment is adjusted so that the adjusted path avoids the dangerous area. After successfully avoiding the danger zone, the final path is obtained.

10. A three-dimensional path planning terminal for urban low-altitude unmanned aerial vehicles (UAVs), characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When one or more programs are executed by one or more processors, the one or more processors implement the three-dimensional path planning method for urban low-altitude unmanned aerial vehicles as described in any one of claims 1 to 9.

Images