An unmanned aerial vehicle unknown environment trajectory planning method based on incremental topological map

By using an incremental topology map method, point cloud data is updated in real time and geometric feature points are extracted to construct a topology map and dynamically update guidance nodes. This solves the problems of path planning efficiency and global guidance for UAVs in unknown environments, and improves navigation efficiency and obstacle avoidance capabilities.

CN123111276APending Publication Date: 2026-11-17BEIJING INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611198035.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Unmanned aerial vehicles (UAVs) struggle to efficiently plan feasible paths in unknown and complex environments. Existing methods suffer from problems such as loss of historical information, lack of global guidance, and high computational and storage overhead, which limit their ability to plan autonomous trajectories in complex environments.

Method used

An incremental topology map-based approach is adopted, which updates the sliding point cloud map by acquiring point cloud data in real time, extracts geometric feature points and constructs a topology map, dynamically updates guide nodes, generates heuristic guide paths, and combines trajectory optimization technology to achieve front-end path search and back-end trajectory optimization.

Benefits of technology

It enables efficient storage of historical information in unknown environments, provides global guidance, improves the navigation efficiency and adaptability of UAVs in complex environments, avoids dead ends, and ensures real-time obstacle avoidance capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123111276A_ABST
    Figure CN123111276A_ABST
Patent Text Reader

Abstract

This invention discloses a method for UAV trajectory planning in unknown environments based on incremental topology maps, comprising: real-time updating of a sliding point cloud map based on current point cloud data; obtaining a current geometric feature point set composed of geometric feature points based on the current point cloud data; preprocessing the geometric feature points in the current geometric feature point set and deleting redundant geometric feature points to obtain a second current geometric feature point set; establishing connections between elements in the second current geometric feature point set and some existing topology nodes in the current topology map; updating the guiding node set corresponding to the first current topology map based on the current geometric feature point set and the first current topology map, and determining the optimal node; determining the front-end search path based on the optimal node; and determining the flight position trajectory and yaw angle trajectory of the UAV from its current position to the guiding node. This invention improves the efficiency and adaptability of UAV trajectory planning in unknown and complex environments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) trajectory planning technology, and specifically to a method for UAV trajectory planning in unknown environments based on incremental topology maps. Background Technology

[0002] In recent years, drones have become an indispensable platform for industrial inspection, search and rescue and autonomous exploration. In most applications, drones need to operate in unknown and complex environments. Limited prior knowledge, irregular obstacles and limited onboard resources make efficient and reliable motion planning in these environments a major challenge.

[0003] To meet the real-time requirements of unknown environments, systems typically employ a rolling temporal planning framework, relying on a high-precision local map centered on the UAV. However, this strategy suffers from inherent limitations such as historical information loss and a lack of global guidance, causing UAVs to struggle to efficiently plan feasible paths when encountering complex structures like dead ends or narrow passages, resulting in them wandering back and forth in local areas. To address these issues, existing research employs different methods to reuse historical information and provide global guidance, but significant limitations remain: incremental search dynamically maintains the search tree, updating only nodes affected by environmental changes to quickly generate new paths, but its search tree size grows with the map, limiting its application in large-scale scenarios; multi-path set reuse strategies provide diverse guidance by maintaining paths of different homotopy classes, but require continuous updating and verification of a large number of paths, resulting in significant computational and storage overhead; existing solutions based on sparse road networks rely on additional dead zone detection mechanisms, leading to some fragmentation in the decision-making process. These problems severely restrict the autonomous trajectory planning capabilities of UAVs in complex and unknown environments. Therefore, it is necessary to build an efficient historical information representation and reuse mechanism, provide heuristic global guidance based on this, and closely integrate it with local planning, so as to avoid UAVs getting stuck in dead zones and ensure their real-time obstacle avoidance capabilities in complex environments. Summary of the Invention

[0004] In view of this, the present invention provides a method for UAV trajectory planning in unknown environments based on incremental topology maps, which can solve the above-mentioned technical problems.

[0005] To solve the above-mentioned technical problems, the present invention is implemented as follows.

[0006] A method for UAV trajectory planning in unknown environments based on incremental topology maps includes: Step S1: Before the UAV reaches the end of the flight mission, the UAV acquires the current point cloud data in real time; updates the sliding point cloud map in real time based on the current point cloud data; and acquires the current geometric feature point set composed of geometric feature points based on the current point cloud data. The geometric feature points include near-end channel feature points and far-end feature points. Step S2: Preprocess the geometric feature points in the current geometric feature point set to obtain the first current geometric feature point set; delete redundant geometric feature points in the first current geometric feature point set to obtain the second current geometric feature point set; establish connections between elements in the second current geometric feature point set and some existing topological nodes in the current topological map to obtain the first current topological map; update the guiding node set based on the current geometric feature point set and the first current topological map, calculate the guiding cost of the guiding nodes, and select the guiding node with the lowest guiding cost as the optimal node; determine the front-end search path corresponding to the first current topological map based on the optimal node. Step S3: Construct a safe flight corridor using the front-end search path as the line segment seed. Under the constraints of the safe flight corridor, take the current position of the UAV as the starting point and optimize the motion trajectory of the UAV in the process of reaching the optimal node along the front-end search path to determine the flight position trajectory and yaw angle trajectory of the UAV from the current position to the guiding node.

[0007] Preferably, the initial start and end points of the UAV flight mission are added to the initial topology map.

[0008] Preferably, in step S1, updating the sliding point cloud map in real time based on the current point cloud data includes: A sliding point cloud map window is constructed with the current location of the drone as the center to determine the range of the area mapped by the sliding point cloud map; historical point cloud data outside the range of the sliding point cloud map window is removed, and space is reserved in the sliding point cloud map for the area that the drone is about to fly into. The current point cloud data is downsampled based on a uniform grid structure to determine the observed point cloud in the processed current point cloud data; the sliding point cloud map is then used. The occupancy status of the location corresponding to the observation point cloud is updated. When an observation point cloud exists at a certain location in the sliding point cloud map, the occupancy status of that location is updated to "occupied"; when no observation point cloud exists, the occupancy status is "unoccupied". When the occupancy status is "occupied", it means that there is an object in the sliding point cloud map that affects the flight of the drone; when the occupancy status is "unoccupied", it means that it is uncertain whether there is an object in the sliding point cloud map that affects the flight of the drone.

[0009] Preferably, in step S1, a current geometric feature point set composed of geometric feature points is obtained based on the current point cloud data. The geometric feature points include near-end channel feature points and far-end feature points, including: Step S101: Convert the current point cloud data to a format based on the drone's current location. Using a spherical coordinate system with the origin, the current point cloud data is discretized and transformed according to a preset angular resolution to obtain a two-dimensional depth matrix. Each element in the depth matrix corresponds to a spatial direction determined by the polar angle and azimuth angle. The element value represents the depth value in that spatial direction, i.e., the distance between the point in the current point cloud data and the current position of the UAV. Based on the depth matrix, the corresponding depth gradient matrix is ​​calculated. Each element in the depth gradient matrix corresponds to a spatial direction determined by the polar angle and azimuth angle. The element value in the depth gradient matrix represents the rate of change of depth in the neighborhood of the discretized point in the current point cloud data in the corresponding spatial direction. Step S102: Determine the adaptive mutation threshold corresponding to each element in the depth gradient matrix. The formula for calculating the adaptive mutation threshold is as follows:

[0010] in, Represents the element values ​​in the depth gradient matrix. and These are parameters that control the rate of increase of the threshold and parameters that control the magnitude of increase, respectively. Near-range depth discontinuities are determined based on the depth gradient matrix. A near-range depth discontinuity is defined as an element in the depth gradient matrix whose value is greater than its corresponding adaptive abrupt change threshold, taken as the target element. The polar angle and azimuth angle of the target element are determined based on its position in the depth gradient matrix. The depth matrix is ​​then searched based on the polar angle and azimuth angle to determine the depth value corresponding to the target element. The coordinates of the point in the Cartesian coordinate system corresponding to the target element are determined, and these coordinates are taken as the near-range depth discontinuities. All near-range depth discontinuities form a set of near-range depth discontinuities. Distant points are determined based on the depth matrix. A distant point is defined as an element in the depth matrix whose value is greater than a distant distance threshold. The elements in the depth matrix are used to determine the polar angle and azimuth angle of the distant point based on its position in the depth gradient matrix, thus determining the coordinates of the point in the Cartesian coordinate system corresponding to the distant point; the distant point is then back-projected onto the UAV position. Center of the sphere, radius is On the virtual threshold sphere, that is, keeping the polar angle and azimuth angle of the far-distance point unchanged, For depth, based on the drone's location Using the center of the sphere as the coordinate system, transform to Cartesian coordinate space to obtain the set of projections of distant points; Step S103: Perform cascade clustering on the set of near-distance depth discontinuities, including: The DBSCAN algorithm is used to remove outliers from the set of near-depth discontinuities and to divide the remaining points in the set of near-depth discontinuities into several first initial spatial correlation clusters. The ISODATA algorithm is used to cluster each of the first initial spatial clusters to obtain several sub-clusters; the centroid of each sub-cluster represents a barrier boundary point. The centroid of each sub-cluster is offset by a preset angle to the empty region in the Cartesian coordinate system, and the offset point is the near-end channel feature point. Step S104: Perform cascade clustering on the distant point projection set, including: The DBSCAN algorithm is used to remove outliers from the near and far point projection sets and divide the remaining points in the far point projection set into several second initial spatial correlation clusters. The ISODATA algorithm is used to cluster each second initial spatially related cluster to obtain several sub-clusters; the centroid of each sub-cluster is used as the far-end feature point. Step S105: Both near-end channel feature points and far-end feature points are used as geometric feature points. All geometric feature points form the current geometric feature point set, which is denoted as the initial geometric feature point set for this planning.

[0011] Preferably, in step S2, the geometric feature points in the current geometric feature point set are preprocessed to obtain a first current geometric feature point set; redundant geometric feature points in the first current geometric feature point set are deleted to obtain a second current geometric feature point set; and connections are established between elements in the second current geometric feature point set and some existing topological nodes in the current topological map to obtain a first current topological map, including: Step S201: Define the first visibility criterion and the second visibility criterion; The first visibility criterion is as follows: if the line connecting two points in Cartesian coordinate system does not intersect any known obstacle, then the two points satisfy the first visibility criterion; if the two points do not satisfy the first visibility criterion, then it is determined that there must be an obstacle between the two points; however, if the two points satisfy the first visibility criterion, it cannot be determined that there must be no obstacle between the two points. The second visibility criterion is as follows: if the convex hull formed by two points in Cartesian coordinate system and the current position of the drone does not contain any known obstacles, then the two points satisfy the second visibility criterion; if the two points satisfy the second visibility criterion, it is determined that there are no obstacles between the two points, but if the two points do not satisfy the second visibility criterion, it cannot be determined that there are definitely obstacles between the two points. Step S202: Perform a preprocessing operation on the geometric feature points in the current geometric feature point set. The preprocessing operation is as follows: any two geometric feature points in the current geometric feature point set form a geometric feature point pair. When the distance between the two geometric feature points in the geometric feature point pair is less than a first preset threshold and the two geometric feature points satisfy the second visibility criterion, merge the two geometric feature points in the geometric feature point pair. The centroid of the merged geometric feature point is used as the position of the merged geometric feature point. The geometric feature points retained in the current geometric feature point set and the merged geometric feature points form a first current geometric feature point set. All elements in the first current geometric feature point set are first current geometric feature points. Step S203: Treat all topological nodes in the current topological map as existing topological nodes; determine the convex polyhedral properties of each existing topological node in the current topological map. Construct a set of convex polyhedra for each existing topological node. :

[0012] in, This is the number of each existing topology node in the current topology map. This represents the set of existing topological nodes in the current topological map; The first current geometric feature point set is removed from the location The first current geometric feature point within the set forms the second current geometric feature point set. The second set of current geometric feature points All elements in the table are the second current geometric feature points; Step S204: Add all second current geometric feature points as new geometric feature points to the current topology map; For each second current geometric feature point , i For each of the second current geometric feature points, perform the following operations: like If a neighboring node satisfies the first visibility criterion, then remove it from the current topology map. Otherwise, in With drones Uniform sampling along the line yields a set of sampling points. ,in, To collect the total number of elements in the point set, , For each sampling point Finding sampling points It satisfies the second visibility and is related to the sampling point The nearest existing topology node Calculate sampling points With existing topology nodes Connection cost of making a connection ; Determine the minimum connection cost from all connection costs, and identify the sampling point corresponding to the minimum connection cost, denoted as the intermediate connection point. Establish intermediate connection points With the second current geometric feature point Intermediate connection point With intermediate connection point Corresponding existing topology nodes By connecting the elements, we obtain the first intermediate topology map; in, ; Step S205: Perform the operation of inserting intermediate points and establishing connecting edges between the topological nodes of the first intermediate topological map: First, check all topological nodes pairwise. For any two topological nodes whose distance is greater than the second preset threshold, insert several intermediate nodes. The intermediate nodes are also considered as topological nodes. After all checks and insertions are performed, any two topological nodes whose distance is less than the third preset threshold form a pair of neighboring nodes. If the pair of neighboring nodes satisfies the second visibility requirement but no connecting edge has been established yet, add a connecting edge to the pair of neighboring nodes. Finally, the first current topological map is obtained. Step S206: Update the convex polyhedral properties of each topological node in the first current topological map, including: Press and drone Arrange all topological nodes in the first current topological map in ascending order of distance; perform the following operations on the topological nodes in the first current topological map in sequence: Based on obstacle information in the current point cloud data, using drones Generate convex polyhedra from point seeds When the topology node in the first current topology map is related to the drone The current distance is less than the distance between the topology node and the drone in the first current topology map. The previous minimum distance will then This refers to the convex polyhedron attribute corresponding to the topological node in the first current topological map.

[0013] Preferably, in step S2, the guidance node set is updated based on the current geometric feature point set and the first current topology map, the guidance cost of the guidance nodes is calculated, and the guidance node with the lowest guidance cost is taken as the optimal node; the front-end search path corresponding to the first current topology map is determined based on the optimal node, including: Step S207: The first set of guiding nodes when the current topology map has not been updated is denoted as... ,for any of the boot nodes Let its set of adjacent nodes be Then its feasible inheritance direction space Defined as:

[0014] in, Indicates adjacent nodes. Indicates the bootstrap node The corresponding position in the Cartesian coordinate system. Indicates the position of adjacent nodes. Represents any point in three-dimensional space. Let T be the real number field and T be the transpose; the feasible direction space of inheritance is formed by the intersection of the half-spaces of adjacent nodes pointing to the direction of the guiding node; Step S208: For any of the boot nodes All of them shall perform the following operations: When the boot node All newly added geometric feature points connected are located in Internal time, guiding node The bootstrap properties are inherited, and the bootstrap node will be... from Delete; When the boot node With drones L The distance is less than the preset third threshold, or all current geometric feature points in the current geometric feature point set generated by the current planning are not in the target area. Internal time, guiding node It no longer has the function of indicating unknown space, and will guide nodes. from Delete; Step S209: Add all the second current geometric feature points from the second current geometric feature point set to the second current geometric feature point set. ; Step S210: From The guiding node with the minimum guiding cost corresponding to the current planning period is selected as the optimal node for the current planning period; the formula for calculating the guiding cost is as follows:

[0015]

[0016] in, Indicates the bootstrap node The cost of guidance This represents the position of the guiding node in the Cartesian coordinate system. The global target point location, This is the current location of the drone. The function represents the actual cost from the drone to the pilot node. The function represents the heuristic cost of guiding the node to the global goal. For heuristic cost weighting, To project local target points onto the sliding point cloud map. This represents the propagation cost from the local target point to the guiding node. Represents the magnitude of a vector; Step S211: Use a graph search algorithm to determine the first guiding path from the UAV to the optimal node of the current planning period in the first current topology map; use the first guiding path as a guide to perform path planning in the sliding point cloud map and generate the front-end search path.

[0017] Preferably, guided by the first guiding path, path planning is performed within the sliding point cloud map to generate a front-end search path, including: Based on each path segment of the first guiding path, an axis-aligned bounding box is constructed. The search space for graph search of the sliding point cloud map is constrained within the bounding box set. A graph search algorithm is used to perform graph search on the sliding point cloud map to generate a front-end search path.

[0018] A UAV trajectory planning device for unknown environments based on incremental topology maps, comprising: Geometric feature point generation module: configured to acquire current point cloud data in real time before the UAV reaches the end of the flight mission; update the sliding point cloud map in real time based on the current point cloud data; and acquire the current geometric feature point set composed of geometric feature points based on the current point cloud data, including near-end channel feature points and far-end feature points. The front-end search module is configured to preprocess the geometric feature points in the current geometric feature point set to obtain a first current geometric feature point set; delete redundant geometric feature points in the first current geometric feature point set to obtain a second current geometric feature point set; establish connections between elements in the second current geometric feature point set and some existing topological nodes in the current topological map to obtain a first current topological map; update the guiding node set based on the current geometric feature point set and the first current topological map, calculate the guiding cost of the guiding nodes, and select the guiding node with the lowest guiding cost as the optimal node; and determine the front-end search path corresponding to the first current topological map based on the optimal node. The trajectory generation module is configured to construct a safe flight corridor using the front-end search path as the line segment seed. Under the constraint of the safe flight corridor, the module optimizes the motion trajectory of the UAV from its current position to the optimal node along the front-end search path, taking the current position of the UAV as the starting point, and determines the flight position trajectory and yaw angle trajectory of the UAV from its current position to the guide node.

[0019] The present invention provides an electronic device, characterized in that the electronic device comprises: A processor is used to execute multiple instructions; Memory, used to store multiple instructions; The plurality of instructions are to be stored in the memory and loaded and executed by the processor as described above.

[0020] The present invention provides a non-transitory computer-readable storage medium storing computer instructions, characterized in that the computer instructions are used to cause the computer to perform the method described above.

[0021] The present invention has the following beneficial effects: This invention constructs a hierarchical environment representation and planning method based on point clouds. The top layer extracts geometric features from the point cloud and incrementally constructs a topology map to provide global guidance. The bottom layer maintains a fine-grained sliding point cloud map centered on the UAV to achieve real-time obstacle avoidance. Based on this, a hierarchical planning architecture is formed, consisting of front-end path search and back-end trajectory optimization. Specifically, by utilizing the depth information of the original point cloud to extract two types of features: near-end depth discontinuities and far-end structural points, newly generated feature points are dynamically integrated into the historical topology as the UAV moves, forming a compact and scalable environment representation. A heuristically guided front-end path is generated through the design of a dynamic update mechanism for guiding nodes and a cost-based selection mechanism. The trajectory optimization part comprehensively considers dynamics and environmental constraints to generate a smooth and executable flight trajectory.

[0022] This invention achieves efficient storage and reuse of historical information by incrementally constructing a topology map, improving the navigation efficiency and adaptability of UAVs in unknown and complex environments while ensuring real-time planning. Through a geometric feature extraction method based on point cloud depth, key nodes with topological significance can be stably acquired without relying on additional environmental representations. Incremental maintenance of the topology structure reduces environmental information storage overhead, constructing sparse spatial connections to improve global path search efficiency. Dynamic updates of topology nodes and a cost-based selection mechanism eliminate the need for additional dead zone detection and explicit backtracking logic. Natural exit from dead ends is achieved solely through the underlying autonomous iteration of the guiding node set, continuously guiding the UAV towards the global target. This is of great significance for promoting the practical application of UAVs in emergency rescue, autonomous exploration, and other fields. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the UAV trajectory planning method for unknown environments based on incremental topology maps according to the present invention.

[0024] Figure 2(A) and Figure 2(B) are schematic diagrams of the sampling point connection cost and the optimal intermediate connection point selection when integrating geometric feature points in this invention, respectively.

[0025] Figure 3This is a schematic diagram illustrating the inheritance and removal of the guide node in this invention.

[0026] Figure 4 This is a schematic diagram of the topology map construction result of the present invention.

[0027] Figure 5 This is a schematic diagram of the trajectory planning results of the present invention.

[0028] Figure 6 This is a schematic diagram of the three-axis velocity and acceleration curves over time in the trajectory planning results of this invention. Detailed Implementation

[0029] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] like Figure 1 As shown, specifically, this invention proposes a method for UAV trajectory planning in unknown environments based on incremental topology maps, including: Step S1: Before the UAV reaches the end of the flight mission, the UAV acquires the current point cloud data in real time; updates the sliding point cloud map in real time based on the current point cloud data; and acquires the current geometric feature point set composed of geometric feature points based on the current point cloud data. The geometric feature points include near-end channel feature points and far-end feature points. Step S2: Preprocess the geometric feature points in the current geometric feature point set to obtain the first current geometric feature point set; delete redundant geometric feature points in the first current geometric feature point set to obtain the second current geometric feature point set; establish connections between elements in the second current geometric feature point set and some existing topological nodes in the current topological map to obtain the first current topological map; update the guiding node set corresponding to the first current topological map based on the current geometric feature point set and the first current topological map, calculate the guiding cost of the guiding nodes, and select the guiding node with the lowest guiding cost as the optimal node; determine the front-end search path corresponding to the first current topological map based on the optimal node. Step S3: Construct a safe flight corridor using the front-end search path as the line segment seed. Under the constraints of the safe flight corridor, take the current position of the UAV as the starting point and optimize the motion trajectory of the UAV in the process of reaching the optimal node along the front-end search path to determine the flight position trajectory and yaw angle trajectory of the UAV from the current position to the guiding node.

[0031] This involves adding the initial start and end points of the drone flight mission to the initial topology map.

[0032] Further, in step S1, updating the sliding point cloud map in real time based on the current point cloud data includes: A sliding point cloud map window is constructed with the current location of the drone as the center to determine the range of the area mapped by the sliding point cloud map; historical point cloud data outside the range of the sliding point cloud map window is removed, and space is reserved in the sliding point cloud map for the area that the drone is about to fly into. The current point cloud data is downsampled based on a uniform grid structure to determine the observed point cloud in the processed current point cloud data; the sliding point cloud map is then used. The occupancy status of the location corresponding to the observation point cloud is updated. When an observation point cloud exists at a certain location in the sliding point cloud map, the occupancy status of that location is updated to "occupied"; when no observation point cloud exists, the occupancy status is "unoccupied". When the occupancy status is "occupied", it means that there is an object in the sliding point cloud map that affects the flight of the drone; when the occupancy status is "unoccupied", it means that it is uncertain whether there is an object in the sliding point cloud map that affects the flight of the drone.

[0033] Further, in step S1, a current geometric feature point set composed of geometric feature points is obtained based on the current point cloud data. The geometric feature points include near-end channel feature points and far-end feature points, including: Step S101: Convert the current point cloud data to a format based on the drone's current location. Using a spherical coordinate system with the origin, the current point cloud data is discretized and transformed according to a preset angular resolution to obtain a two-dimensional depth matrix. Each element in the depth matrix corresponds to a spatial direction determined by the polar angle and azimuth angle. The element value represents the depth value in that spatial direction, i.e., the distance between the point in the current point cloud data and the current position of the UAV. Based on the depth matrix, the corresponding depth gradient matrix is ​​calculated. Each element in the depth gradient matrix corresponds to a spatial direction determined by the polar angle and azimuth angle. The element value in the depth gradient matrix represents the rate of change of depth in the neighborhood of the discretized point in the current point cloud data in the corresponding spatial direction. Step S102: Determine the adaptive mutation threshold corresponding to each element in the depth gradient matrix. The formula for calculating the adaptive mutation threshold is as follows:

[0034] in, Represents the element values ​​in the depth gradient matrix. and These are parameters that control the rate of increase of the threshold and parameters that control the magnitude of increase, respectively. Near-range depth discontinuities are determined based on the depth gradient matrix. A near-range depth discontinuity is defined as an element in the depth gradient matrix whose value is greater than its corresponding adaptive abrupt change threshold, taken as the target element. The polar angle and azimuth angle of the target element are determined based on its position in the depth gradient matrix. The depth matrix is ​​then searched based on the polar angle and azimuth angle to determine the depth value corresponding to the target element. The coordinates of the point in the Cartesian coordinate system corresponding to the target element are determined, and these coordinates are taken as the near-range depth discontinuities. All near-range depth discontinuities form a set of near-range depth discontinuities. Distant points are determined based on the depth matrix. A distant point is defined as an element in the depth matrix whose value is greater than a distant distance threshold. The elements in the depth matrix are used to determine the polar angle and azimuth angle of the distant point based on its position in the depth gradient matrix, thus determining the coordinates of the point in the Cartesian coordinate system corresponding to the distant point; the distant point is then back-projected onto the UAV position. Center of the sphere, radius is On the virtual threshold sphere, that is, keeping the polar angle and azimuth angle of the far-distance point unchanged, For depth, based on the drone's location Using the center of the sphere as the coordinate system, transform to Cartesian coordinate space to obtain the set of projections of distant points; Step S103: Perform cascade clustering on the set of near-distance depth discontinuities, including: The DBSCAN algorithm is used to remove outliers from the set of near-depth discontinuities and to divide the remaining points in the set of near-depth discontinuities into several first initial spatial correlation clusters. The ISODATA algorithm is used to cluster each of the first initial spatial clusters to obtain several sub-clusters; the centroid of each sub-cluster represents a barrier boundary point. The centroid of each sub-cluster is offset by a preset angle to the empty region in the Cartesian coordinate system, and the offset point is the near-end channel feature point. Step S104: Perform cascade clustering on the distant point projection set, including: The DBSCAN algorithm is used to remove outliers from the near and far point projection sets and divide the remaining points in the far point projection set into several second initial spatial correlation clusters. The ISODATA algorithm is used to cluster each second initial spatially related cluster to obtain several sub-clusters; the centroid of each sub-cluster is used as the far-end feature point. Step S105: Both near-end channel feature points and far-end feature points are used as geometric feature points. All geometric feature points form the current geometric feature point set, which is denoted as the initial geometric feature point set for this planning.

[0035] In this invention, the closer the region is to the drone, the smaller the mutation threshold.

[0036] Further, in step S2, the geometric feature points in the current geometric feature point set are preprocessed to obtain a first current geometric feature point set; redundant geometric feature points in the first current geometric feature point set are deleted to obtain a second current geometric feature point set; connections are established between elements in the second current geometric feature point set and some existing topological nodes in the current topological map to obtain a first current topological map, including: Step S201: Define the first visibility criterion and the second visibility criterion; The first visibility criterion is as follows: if the line connecting two points in Cartesian coordinate system does not intersect any known obstacle, then the two points satisfy the first visibility criterion; if the two points do not satisfy the first visibility criterion, then it is determined that there must be an obstacle between the two points; however, if the two points satisfy the first visibility criterion, it cannot be determined that there must be no obstacle between the two points. The second visibility criterion is as follows: if the convex hull formed by two points in Cartesian coordinate system and the current position of the drone does not contain any known obstacles, then the two points satisfy the second visibility criterion; if the two points satisfy the second visibility criterion, it is determined that there are no obstacles between the two points, but if the two points do not satisfy the second visibility criterion, it cannot be determined that there are definitely obstacles between the two points. Step S202: Perform a preprocessing operation on the geometric feature points in the current geometric feature point set. The preprocessing operation is as follows: any two geometric feature points in the current geometric feature point set form a geometric feature point pair. When the distance between the two geometric feature points in the geometric feature point pair is less than a first preset threshold and the two geometric feature points satisfy the second visibility criterion, merge the two geometric feature points in the geometric feature point pair. The centroid of the merged geometric feature point is used as the position of the merged geometric feature point. The geometric feature points retained in the current geometric feature point set and the merged geometric feature points form a first current geometric feature point set. All elements in the first current geometric feature point set are first current geometric feature points. Step S203: Treat all topological nodes in the current topological map as existing topological nodes; determine the convex polyhedral properties of each existing topological node in the current topological map. Construct a set of convex polyhedra for each existing topological node. :

[0037] in, This is the number of each existing topology node in the current topology map. This represents the set of existing topological nodes in the current topological map; The first current geometric feature point set is removed from the location The first current geometric feature point within the set forms the second current geometric feature point set. The second set of current geometric feature points All elements in the table are the second current geometric feature points; Step S204: Add all second current geometric feature points as new geometric feature points to the current topology map; For each second current geometric feature point , i For each of the second current geometric feature points, perform the following operations: like If a neighboring node satisfies the first visibility criterion, then remove it from the current topology map. Otherwise, in With drones Uniform sampling along the line yields a set of sampling points. ,in, To collect the total number of elements in the point set, , For each sampling point Finding sampling points It satisfies the second visibility and is related to the sampling point The nearest existing topology node Calculate sampling points With existing topology nodes Connection cost of making a connection ; Determine the minimum connection cost from all connection costs, and identify the sampling point corresponding to the minimum connection cost, denoted as the intermediate connection point. Establish intermediate connection points With the second current geometric feature point Intermediate connection point With intermediate connection point Corresponding existing topology nodes By connecting the elements, we obtain the first intermediate topology map; in, ; Step S205: Perform the operation of inserting intermediate points and establishing connecting edges between the topological nodes of the first intermediate topological map: First, check all topological nodes pairwise. For any two topological nodes whose distance is greater than the second preset threshold, insert several intermediate nodes. The intermediate nodes are also considered as topological nodes. After all checks and insertions are performed, any two topological nodes whose distance is less than the third preset threshold form a pair of neighboring nodes. If the pair of neighboring nodes satisfies the second visibility requirement but no connecting edge has been established yet, add a connecting edge to the pair of neighboring nodes. Finally, the first current topological map is obtained. Step S206: Update the convex polyhedral properties of each topological node in the first current topological map, including: Press and drone Arrange all topological nodes in the first current topological map in ascending order of distance; perform the following operations on the topological nodes in the first current topological map in sequence: Based on obstacle information in the current point cloud data, using drones Generate convex polyhedra from point seeds When the topology node in the first current topology map is related to the drone The current distance is less than the distance between the topology node and the drone in the first current topology map. The previous minimum distance will then This refers to the convex polyhedron attribute corresponding to the topological node in the first current topological map.

[0038] In this invention, step S206 realizes the processing of the already traversed space. Corrections.

[0039] Further, in step S2, the guiding node set corresponding to the first current topology map is updated based on the current geometric feature point set and the first current topology map; the guiding cost of the guiding nodes is calculated; and the guiding node with the lowest guiding cost is taken as the optimal node; the front-end search path corresponding to the first current topology map is determined based on the optimal node, including: Step S207: The first set of guiding nodes when the current topology map has not been updated is denoted as... ,for any of the boot nodes Let its set of adjacent nodes be Then its feasible inheritance direction space Defined as:

[0040] in, Indicates adjacent nodes. Indicates the bootstrap node The corresponding position in the Cartesian coordinate system. Indicates the position of adjacent nodes. Represents any point in three-dimensional space. Let T be the real number field and T be the transpose; the feasible direction space of inheritance is formed by the intersection of the half-spaces of adjacent nodes pointing to the direction of the guiding node; Step S208: For any of the boot nodes All of them shall perform the following operations: When the boot node All newly added geometric feature points connected are located in Internal time, guiding node The bootstrap properties are inherited, and the bootstrap node will be... from Delete; When the boot node With drones L The distance is less than the preset third threshold, or all current geometric feature points in the current geometric feature point set generated by the current planning are not in the target area. Internal time, guiding node It no longer has the function of indicating unknown space, and will guide nodes. from Delete; Step S209: Add all the second current geometric feature points from the second current geometric feature point set to the second current geometric feature point set. ; Step S210: From The guiding node with the minimum guiding cost corresponding to the current planning period is selected as the optimal node for the current planning period; the formula for calculating the guiding cost is as follows:

[0041]

[0042] in, Indicates the bootstrap node The cost of guidance This represents the position of the guiding node in the Cartesian coordinate system. The global target point location, This is the current location of the drone. The function represents the actual cost from the drone to the pilot node. The function represents the heuristic cost of guiding the node to the global goal. For heuristic cost weighting, To project local target points onto the sliding point cloud map. This represents the propagation cost from the local target point to the guiding node. Represents the magnitude of a vector; Step S211: Use a graph search algorithm to determine the first guiding path from the UAV to the optimal node of the current planning period in the first current topology map; use the first guiding path as a guide to perform path planning in the sliding point cloud map and generate the front-end search path.

[0043] Furthermore, guided by the first guiding path, path planning is performed within the sliding point cloud map to generate a front-end search path, including: Based on each path segment of the first guiding path, an axis-aligned bounding box is constructed. The search space for graph search of the sliding point cloud map is constrained within the bounding box set. A graph search algorithm is used to perform graph search on the sliding point cloud map to generate a front-end search path.

[0044] In this invention, based on the dynamically maintained set of guiding nodes, guiding nodes are selected for rolling planning according to the principle of minimum cost, wherein the guiding nodes... The cost is defined as:

[0045] in, For node position, The target point location, The actual cost of getting from the drone to the node is obtained by using a graph search algorithm on the topology map. The heuristic cost from the node to the global goal represents the estimated cost. This is the heuristic cost weight. Since the heuristic cost needs to reflect the additional cost of bypassing known obstacles, it cannot be simply represented as the Euclidean distance between two points. Therefore, the wavefront propagation method is used to estimate it: like Not on local map If the target is located within a local area, it is first projected onto the local map boundary, and then extended outwards along the projection direction by a certain distance to obtain the local target point. Subsequently, with Starting from a point on the map, perform one Dijkstra's propagation within the map to calculate the distance from each point on the local map to the destination. The cost of dissemination For guide nodes outside the local map, the route to the global target point is still used. The heuristic cost of the guiding node can be approximated using Euclidean distance.

[0046] To improve computational efficiency, a resolution reduction strategy is employed during the propagation process, and an early termination condition is set. Propagation terminates when the propagation distance exceeds a set threshold or the computation time exceeds an upper limit. This method requires only one propagation to obtain heuristic values ​​for multiple guiding nodes within the local map, while accurately reflecting the impact of known obstacles on path costs. In each rolling planning step, the guiding node with the lowest cost is selected from the candidate set. .

[0047] In this invention, A is used The algorithm obtains the actual cost of the bootstrap node and searches for a coarse bootstrap path on the topology graph that points to the optimal bootstrap node.

[0048] Step S3 involves constructing a safe flight corridor using the front-end search path as a line segment seed. Under the constraints of the safe flight corridor, and taking the UAV's current position as the starting point, the motion trajectory along the front-end search path to the optimal node is optimized to determine the UAV's flight position trajectory and yaw angle trajectory from its current position to the guiding node, including: A safe flight corridor is constructed using the front-end search path as the line segment seed. The continuous time trajectory of the UAV is parameterized. Under the premise of satisfying the dynamic constraints of the UAV, a trajectory optimization problem model is constructed. The front-end search path is used as the initial value and optimized under the constraint of the safe flight corridor to obtain the flight position trajectory and yaw angle trajectory of the UAV from the current position to the guidance node.

[0049] In this invention, the Configuration Space Iterative Region Expansion (CIRI) method is used to construct a safe flight corridor in free space based on local point cloud obstacle information, thereby constraining the feasible space for trajectory optimization. The MINCO method is used to parameterize the continuous time trajectory of the UAV, transforming the trajectory optimization problem into a finite-dimensional optimization problem with respect to intermediate points and time allocation. The L-BFGS algorithm is then used to iteratively solve the problem based on the objective function and its gradient information, thereby obtaining a smooth trajectory that satisfies both safety and dynamic constraints.

[0050] The present invention also provides a specific embodiment of a method for planning the trajectory of a UAV in an unknown environment based on an incremental topology map.

[0051] Example 1: This invention also provides a specific embodiment of a UAV trajectory planning method for unknown environments based on incremental topology maps, as shown in Figure 2- Figure 6 As shown.

[0052] Step 1: Update the local sliding map centered on the drone's current location. The map range is updated synchronously during the drone's movement, historical data exceeding the range is removed, and update space is reserved for newly entered areas. In this embodiment, the local map range is set to 50. 30 3m 3 The data update method within the sliding map is as follows: First, the newly acquired raw point cloud data is preprocessed by downsampling based on a uniform grid structure. Then, the occupancy status of the observed point cloud locations within the local map is updated without performing ray projection operations.

[0053] Step 2: Extract spatial geometric feature points based on point cloud depth information, which will serve as candidate nodes for subsequent topology graph construction; Step 2.1: Convert the newly received point cloud data to a format based on the drone's location. In a spherical coordinate system with the origin at the origin, and with a fixed angular resolution, the azimuth angle is discretized. With elevation angle A two-dimensional depth matrix is ​​constructed, and then the depth gradient matrix is ​​calculated, where each element represents the rate of change of depth in the neighborhood of that point. In this embodiment, the discrete angular resolution is set to 1°.

[0054] Step 2.2: Set an adaptive mutation threshold based on the distance from the point to the drone:

[0055] in, The depth of the representative point, i.e., the distance from the drone, and The control parameters representing the threshold growth rate and growth magnitude are respectively, and in this embodiment, we take... , The closer an area is to the drone, the smaller the mutation threshold. Based on this threshold, points with gradient values ​​greater than the corresponding location threshold are extracted from the depth gradient matrix, forming a set of near-distance depth discontinuities.

[0056] Step 2.3: Perform cascaded clustering on the depth discontinuous point set: First, the DBSCAN algorithm is used to remove outliers and form initial spatially correlated clusters; then, the ISODATA algorithm is used to adaptively subdivide each cluster to obtain several sub-clusters, and the centroid of each sub-cluster represents an obstacle boundary point. The near-end channel feature points are obtained by offsetting the obstacle boundary points to the empty region by a certain angle.

[0057] Step 2.4: Set the distance threshold Filter points with a depth greater than the threshold and back-project them onto the drone's location. Center of the sphere, radius is On the virtual threshold sphere, the set of distant point projections is obtained. In this embodiment, we take... m. The same cascaded clustering strategy (DBSCAN+ISODATA) is applied to this set to extract the centers of each sub-cluster as distant feature points representing the distant topological structure.

[0058] Step 2.5: The near-end channel feature points and the far-end feature points together constitute the geometric feature point set of the current frame. , as candidate nodes for subsequent topology graph construction.

[0059] Step 3: Integrate the extracted geometric feature points into the existing topology to achieve incremental updates and maintenance of the topology map; Step 3.1: Process the geometric feature points generated in Step 2 Preprocessing is performed to merge feature points that are too close to each other and meet the Type II visibility requirement. The centroid of the merged point is then taken.

[0060] Step 3.2: To avoid generating redundant nodes in areas already traversed by the drone, the system assigns each topology node... Maintain the properties of a convex polyhedron Then define a set of convex polyhedra. To represent the space that has been traversed:

[0061] in, This indicates that a set of topological nodes already exists. Secondary filtering is performed on the feature points: points located at... Feature points within the area form a point set. .

[0062] Step 3.3: For the feature points that are still retained at this point Process them one by one: If in the following radius of the center If a node satisfying Type I visibility exists within the range, remove it. , This is called the filtering threshold. In this embodiment, we take... m; otherwise, in and Uniform sampling along the line yields a set of sampling points. ,in , For each sampling point Find the nearest existing node that satisfies Type II visibility. And calculate the connection cost:

[0063] Select the sampling point with the lowest cost (referred to as intermediate connection points), establish connections between them and feature points respectively. and the corresponding nearest visible node The connection of (called anchor points) enables the integration of new feature points into the topology, as shown in Figure 2.

[0064] Step 3.4: After processing all feature points, insert intermediate nodes for feasible connections that are too far apart, and add connections for pairs of neighboring nodes that are visible but not connected in Type II.

[0065] Step 3.5: Based on the current point cloud obstacle information, with Generate convex polyhedra from point seeds Based on the minimum distance update criterion, first determine the distance to the minimum distance update criterion. The nearest node, if from If the distance to the node is less than the distance recorded during the node's last update, then use... Replace the convex polyhedron associated with this node. This process enables the replacement of the traversed space. Corrections.

[0066] Step 4: Dynamically maintain the set of guiding nodes, realize the generation, inheritance and removal of guiding nodes; calculate the heuristic cost of each guiding node based on the wavefront propagation method, and select the optimal guiding node through cost evaluation; perform graph search on the topology map to obtain the topology path pointing to the optimal guiding node, and generate the front-end fine path under its constraints. Step 4.1: Dynamically maintain the bootstrap node set ,like Figure 3 As shown. Before inserting new nodes and edges in step three, for any guiding node... Let its set of adjacent nodes be Then its feasible inheritance direction space is defined as:

[0067] in, Indicates adjacent nodes. Indicates the position of adjacent nodes. This represents any point in three-dimensional space. The space is formed by the intersection of half-spaces pointing from adjacent nodes to the guiding node.

[0068] When a certain existing boot node A bootstrap attribute is considered to be inherited and is removed from the bootstrap node set if the following conditions are met: In step S4, node It becomes the anchor point connecting the newly added feature points, and the connection... All intermediate connection points satisfy .

[0069] When a certain existing boot node A node is removed when one of the following conditions is met, indicating that the node no longer has the function of further indicating the unknown space: (1) Unmanned aerial vehicles and The distance is less than the set threshold In this embodiment, take m, i.e.

[0070] (2) None of the original geometric feature points generated during this planning period are located in within, that is

[0071] Finally, nodes formed by feature points newly added to the topology graph during the planning cycle automatically become guiding nodes.

[0072] Step 4.2: Based on the dynamically maintained set of guiding nodes, select guiding nodes for rolling planning according to the principle of minimum cost, wherein the guiding nodes... The cost is defined as:

[0073] in, For node position, The target point location, The actual cost from the drone to the node is obtained by using a graph search algorithm on the topology map. Ideally, A is used. The algorithm obtains the actual cost of the bootstrap node; The heuristic cost from the node to the global goal represents the estimated cost. To heuristically determine the cost weights, in this embodiment, we take... Since the heuristic cost needs to reflect the additional cost of bypassing known obstacles, it cannot be simply expressed as the Euclidean distance between two points. Therefore, the wavefront propagation method is used to estimate it: like Not on local map If the target is located within a local area, it is first projected onto the local map boundary, and then extended outwards along the projection direction by a certain distance to obtain the local target point. Subsequently, with Starting from a point on the map, perform one Dijkstra's propagation within the map to calculate the distance from each point on the local map to the destination. The cost of dissemination For guide nodes outside the local map, the route to the global target point is still used. The heuristic cost of the guiding node can be approximated using Euclidean distance.

[0074] To improve computational efficiency, a resolution reduction strategy is employed during the propagation process, and an early termination condition is set. Propagation terminates when the propagation distance exceeds a set threshold or the computation time exceeds an upper limit. This method requires only one propagation to obtain heuristic values ​​for multiple guiding nodes within the local map, while accurately reflecting the impact of known obstacles on path costs. In each rolling planning step, the guiding node with the lowest cost is selected from the candidate set. .

[0075] Step 4.3: After obtaining the guiding node for the current planning period, firstly, use a graph search algorithm on the topology graph to generate a coarse guiding path pointing to that guiding node. Ideally, use A... The algorithm searches for a guiding path; guided by this path, fine-grained path planning is performed within the local map: axis-aligned bounding boxes are constructed for each path segment of the coarse path, and the search space of the graph search is constrained within the set of bounding boxes to improve search efficiency, thereby obtaining a front-end path that satisfies obstacle avoidance constraints.

[0076] Step 5: Construct a safe flight corridor using the front-end search path as the line segment seed; based on this, parameterize the continuous time trajectory of the UAV. Under the premise of satisfying the UAV dynamic constraints, construct a trajectory optimization problem model with intermediate points and time allocation as optimization variables and minimizing flight time and control variables as the objective function. Using the front-end search path as the initial value, optimize under the constraints of the safe flight corridor to finally obtain a smooth and feasible position trajectory and yaw angle trajectory. In this embodiment, the UAV dynamic constraints are set as follows: , .

[0077] Preferably, the Configuration Space Iterative Region Expansion (CIRI) method is used to construct a safe flight corridor in free space based on local point cloud obstacle information, thereby constraining the feasible space for trajectory optimization. The MINCO method is used to parameterize the continuous time trajectory of the UAV, transforming the trajectory optimization problem into a finite-dimensional optimization problem with respect to intermediate points and time allocation. The L-BFGS algorithm is then used to iteratively solve the problem based on the objective function and its gradient information, thereby obtaining a smooth trajectory that satisfies both safety and dynamic constraints.

[0078] In this embodiment, with a size of 116 90 3m 3 The method for constructing topological maps and planning trajectories for UAVs in an environment was validated. Environmental mapping was completed by remotely controlling the movement of the UAV, and the mapping results are as follows: Figure 4 As shown; set the start and end points on the map, perform autonomous trajectory planning, and the planned trajectory result is as follows. Figure 5 As shown, Figure 6 The diagram shows how the three-axis velocity and acceleration of the trajectory change over time.

[0079] The present invention also provides a UAV trajectory planning device for unknown environments based on incremental topology maps, comprising: Geometric feature point generation module: configured to acquire current point cloud data in real time before the UAV reaches the end of the flight mission; update the sliding point cloud map in real time based on the current point cloud data; and acquire the current geometric feature point set composed of geometric feature points based on the current point cloud data, including near-end channel feature points and far-end feature points. The front-end search module is configured to preprocess the geometric feature points in the current geometric feature point set to obtain a first current geometric feature point set; delete redundant geometric feature points in the first current geometric feature point set to obtain a second current geometric feature point set; establish connections between elements in the second current geometric feature point set and some existing topological nodes in the current topological map to obtain a first current topological map; update the guiding node set corresponding to the first current topological map based on the current geometric feature point set and the first current topological map, calculate the guiding cost of the guiding nodes, and select the guiding node with the lowest guiding cost as the optimal node; and determine the front-end search path corresponding to the first current topological map based on the optimal node. The trajectory generation module is configured to construct a safe flight corridor using the front-end search path as the line segment seed. Under the constraint of the safe flight corridor, the module optimizes the motion trajectory of the UAV from its current position to the optimal node along the front-end search path, taking the current position of the UAV as the starting point, and determines the flight position trajectory and yaw angle trajectory of the UAV from its current position to the guide node.

[0080] The specific embodiments described above only illustrate the design principles of the present invention. The shapes and names of the components in this description may differ and are not limited. Therefore, those skilled in the art can modify or make equivalent substitutions to the technical solutions described in the foregoing embodiments; and these modifications and substitutions do not depart from the inventive spirit and technical solutions of the present invention, and should all fall within the protection scope of the present invention.

Claims

1. A method for UAV trajectory planning in unknown environments based on incremental topology maps, characterized in that, include: Step S1: Before the drone reaches the end of the flight mission, the drone acquires the current point cloud data in real time; The sliding point cloud map is updated in real time based on the current point cloud data; The current geometric feature point set is obtained based on the current point cloud data. The geometric feature points include near-end channel feature points and far-end feature points. Step S2: Preprocess the geometric feature points in the current geometric feature point set to obtain the first current geometric feature point set; Delete redundant geometric feature points from the first set of current geometric feature points to obtain the second set of current geometric feature points; Establish connections between the elements of the second current geometric feature point set and some existing topological nodes in the current topological map to obtain the first current topological map; Based on the current set of geometric feature points and the first current topology map, update the set of guiding nodes, calculate the guiding cost of the guiding nodes, and select the guiding node with the lowest guiding cost as the optimal node; determine the front-end search path corresponding to the first current topology map based on the optimal node; Step S3: Construct a safe flight corridor using the front-end search path as the line segment seed. Under the constraints of the safe flight corridor, take the current position of the UAV as the starting point and optimize the motion trajectory of the UAV in the process of reaching the optimal node along the front-end search path to determine the flight position trajectory and yaw angle trajectory of the UAV from the current position to the guiding node.

2. The method as described in claim 1, characterized in that, Add the initial start and end points of the drone flight mission to the initial topology map.

3. The method as described in claim 1, characterized in that, In step S1, updating the sliding point cloud map in real time based on the current point cloud data includes: A sliding point cloud map window is constructed with the current location of the drone as the center to determine the range of the area mapped by the sliding point cloud map; historical point cloud data outside the range of the sliding point cloud map window is removed, and space is reserved in the sliding point cloud map for the area that the drone is about to fly into. The current point cloud data is downsampled based on a uniform grid structure to determine the observed point cloud in the processed current point cloud data; the sliding point cloud map is then used. The occupancy status of the location corresponding to the observation point cloud is updated. When an observation point cloud exists at a certain location in the sliding point cloud map, the occupancy status of that location is updated to "occupied"; when no observation point cloud exists, the occupancy status is "unoccupied". When the occupancy status is "occupied", it means that there is an object in the sliding point cloud map that affects the flight of the drone; when the occupancy status is "unoccupied", it means that it is uncertain whether there is an object in the sliding point cloud map that affects the flight of the drone.

4. The method as described in claim 3, characterized in that, In step S1, a current set of geometric feature points is obtained based on the current point cloud data. These geometric feature points include near-end channel feature points and far-end feature points, including: Step S101: Convert the current point cloud data to a format based on the drone's current location. Using a spherical coordinate system with the origin, the current point cloud data is discretized and transformed according to a preset angular resolution to obtain a two-dimensional depth matrix. Each element in the depth matrix corresponds to a spatial direction determined by the polar angle and azimuth angle. The element value represents the depth value in that spatial direction, i.e., the distance between the point in the current point cloud data and the current position of the UAV. Based on the depth matrix, the corresponding depth gradient matrix is ​​calculated. Each element in the depth gradient matrix corresponds to a spatial direction determined by the polar angle and azimuth angle. The element value in the depth gradient matrix represents the rate of change of depth in the neighborhood of the discretized point in the current point cloud data in the corresponding spatial direction. Step S102: Determine the adaptive mutation threshold corresponding to each element in the depth gradient matrix. The formula for calculating the adaptive mutation threshold is as follows: in, Represents the element values ​​in the depth gradient matrix. and These are parameters that control the rate of increase of the threshold and parameters that control the magnitude of increase, respectively. Near-range depth discontinuities are determined based on the depth gradient matrix. A near-range depth discontinuity is defined as an element in the depth gradient matrix whose value is greater than its corresponding adaptive abrupt change threshold, taken as the target element. The polar angle and azimuth angle of the target element are determined based on its position in the depth gradient matrix. The depth matrix is ​​then searched based on the polar angle and azimuth angle to determine the depth value corresponding to the target element. The coordinates of the point in the Cartesian coordinate system corresponding to the target element are determined, and these coordinates are taken as the near-range depth discontinuities. All near-range depth discontinuities form a set of near-range depth discontinuities. Distant points are determined based on the depth matrix. A distant point is defined as an element in the depth matrix whose value is greater than a distant distance threshold. The elements in the depth matrix are used to determine the polar angle and azimuth angle of the distant point based on its position in the depth gradient matrix, thus determining the coordinates of the point in the Cartesian coordinate system corresponding to the distant point; the distant point is then back-projected onto the UAV position. Center of the sphere, radius is On the virtual threshold sphere, that is, keeping the polar angle and azimuth angle of the far-distance point unchanged, For depth, based on the drone's location Using the center of the sphere as the coordinate system, transform to Cartesian coordinate space to obtain the set of projections of distant points; Step S103: Perform cascade clustering on the set of near-distance depth discontinuities, including: The DBSCAN algorithm is used to remove outliers from the set of near-depth discontinuities and to divide the remaining points in the set of near-depth discontinuities into several first initial spatial correlation clusters. The ISODATA algorithm is used to cluster each of the first initial spatial clusters to obtain several sub-clusters; the centroid of each sub-cluster represents a barrier boundary point. The centroid of each sub-cluster is offset by a preset angle to the empty region in the Cartesian coordinate system, and the offset point is the near-end channel feature point. Step S104: Perform cascade clustering on the distant point projection set, including: The DBSCAN algorithm is used to remove outliers from the near and far point projection sets and divide the remaining points in the far point projection set into several second initial spatial correlation clusters. The ISODATA algorithm is used to cluster each second initial spatially related cluster to obtain several sub-clusters; the centroid of each sub-cluster is used as the far-end feature point. Step S105: Both near-end channel feature points and far-end feature points are used as geometric feature points. All geometric feature points form the current geometric feature point set, which is denoted as the initial geometric feature point set for this planning.

5. The method as described in claim 4, characterized in that, In step S2, the geometric feature points in the current geometric feature point set are preprocessed to obtain the first current geometric feature point set; Delete redundant geometric feature points from the first set of current geometric feature points to obtain the second set of current geometric feature points; Establish connections between elements in the second current geometric feature point set and some existing topological nodes in the current topological map to obtain the first current topological map, including: Step S201: Define the first visibility criterion and the second visibility criterion; The first visibility criterion is as follows: if the line connecting two points in Cartesian coordinate system does not intersect any known obstacle, then the two points satisfy the first visibility criterion; if the two points do not satisfy the first visibility criterion, then it is determined that there must be an obstacle between the two points; however, if the two points satisfy the first visibility criterion, it cannot be determined that there must be no obstacle between the two points. The second visibility criterion is as follows: if the convex hull formed by two points in Cartesian coordinate system and the current position of the drone does not contain any known obstacles, then the two points satisfy the second visibility criterion; if the two points satisfy the second visibility criterion, it is determined that there are no obstacles between the two points, but if the two points do not satisfy the second visibility criterion, it cannot be determined that there are definitely obstacles between the two points. Step S202: Perform a preprocessing operation on the geometric feature points in the current geometric feature point set. The preprocessing operation is as follows: any two geometric feature points in the current geometric feature point set form a geometric feature point pair. When the distance between the two geometric feature points in the geometric feature point pair is less than a first preset threshold and the two geometric feature points satisfy the second visibility criterion, merge the two geometric feature points in the geometric feature point pair. The centroid of the merged geometric feature point is used as the position of the merged geometric feature point. The geometric feature points retained in the current geometric feature point set and the merged geometric feature points form a first current geometric feature point set. All elements in the first current geometric feature point set are first current geometric feature points. Step S203: Treat all topological nodes in the current topological map as existing topological nodes; determine the convex polyhedral properties of each existing topological node in the current topological map. Construct a set of convex polyhedra for each existing topological node. : in, This is the number of each existing topology node in the current topology map. This represents the set of existing topological nodes in the current topological map; The first current geometric feature point set is removed from the location The first current geometric feature point within the set forms the second current geometric feature point set. The second set of current geometric feature points All elements in the table are the second current geometric feature points; Step S204: Add all second current geometric feature points as new geometric feature points to the current topology map; For each second current geometric feature point , i For each of the second current geometric feature points, perform the following operations: like If a neighboring node satisfies the first visibility criterion, then remove it from the current topology map. Otherwise, in With drones Uniform sampling along the line yields a set of sampling points. ,in, To collect the total number of elements in the point set, , For each sampling point Finding sampling points It satisfies the second visibility and is related to the sampling point The nearest existing topology node Calculate sampling points With existing topology nodes Connection cost of making a connection ; Determine the minimum connection cost from all connection costs, and identify the sampling point corresponding to the minimum connection cost, denoted as the intermediate connection point. Establish intermediate connection points With the second current geometric feature point Intermediate connection point With intermediate connection point Corresponding existing topology nodes By connecting the elements, we obtain the first intermediate topology map; in, ; Step S205: Perform the operation of inserting intermediate points and establishing connecting edges between the topological nodes of the first intermediate topological map: First, check all topological nodes pairwise. For any two topological nodes whose distance is greater than the second preset threshold, insert several intermediate nodes. The intermediate nodes are also considered as topological nodes. After all checks and insertions are performed, any two topological nodes whose distance is less than the third preset threshold form a pair of neighboring nodes. If the pair of neighboring nodes satisfies the second visibility requirement but no connecting edge has been established yet, add a connecting edge to the pair of neighboring nodes. Finally, the first current topological map is obtained. Step S206: Update the convex polyhedral properties of each topological node in the first current topological map, including: Press and drone Arrange all topological nodes in the first current topological map in ascending order of distance; perform the following operations on the topological nodes in the first current topological map in sequence: Based on obstacle information in the current point cloud data, using drones Generate convex polyhedra from point seeds When the topology node in the first current topology map is related to the drone The current distance is less than the distance between the topology node and the drone in the first current topology map. The previous minimum distance will then This refers to the convex polyhedron attribute corresponding to the topological node in the first current topological map.

6. The method as described in claim 5, characterized in that, In step S2, the guidance node set is updated based on the current geometric feature point set and the first current topology map, the guidance cost of the guidance node is calculated, and the guidance node with the lowest guidance cost is taken as the optimal node. The front-end search path corresponding to the first current topology map is determined based on the optimal node, including: Step S207: The first set of guiding nodes when the current topology map has not been updated is denoted as... ,for any of the boot nodes Let its set of adjacent nodes be Then its feasible inheritance direction space Defined as: in, Indicates adjacent nodes. Indicates the bootstrap node The corresponding position in the Cartesian coordinate system. Indicates the position of adjacent nodes. Represents any point in three-dimensional space. Let T be the real number field and T be the transpose; the feasible direction space of inheritance is formed by the intersection of the half-spaces of adjacent nodes pointing to the direction of the guiding node; Step S208: For any of the boot nodes All of them shall perform the following operations: When the boot node All newly added geometric feature points connected are located in Internal time, guiding node The bootstrap properties are inherited, and the bootstrap node will be... from Delete; When the boot node With drones L The distance is less than the preset third threshold, or all current geometric feature points in the current geometric feature point set generated by the current planning are not in the target area. Internal time, guiding node It no longer has the function of indicating unknown space, and will guide nodes. from Delete; Step S209: Add all the second current geometric feature points from the second current geometric feature point set to the second current geometric feature point set. ; Step S210: From The guiding node with the minimum guiding cost corresponding to the current planning period is selected as the optimal node for the current planning period; the formula for calculating the guiding cost is as follows: in, Indicates the bootstrap node The cost of guidance This represents the position of the guiding node in the Cartesian coordinate system. The global target point location, This is the current location of the drone. The function represents the actual cost from the drone to the pilot node. The function represents the heuristic cost of guiding the node to the global goal. To heuristically weigh costs, To project local target points onto the sliding point cloud map. This represents the propagation cost from the local target point to the guiding node. Represents the magnitude of a vector; Step S211: Use a graph search algorithm to determine the first guiding path from the UAV to the optimal node of the current planning period in the first current topology map; use the first guiding path as a guide to perform path planning in the sliding point cloud map and generate the front-end search path.

7. The method as described in claim 6, characterized in that, Guided by the first guiding path, path planning is performed within the sliding point cloud map to generate the front-end search path, including: Based on each path segment of the first guiding path, an axis-aligned bounding box is constructed. The search space for graph search of the sliding point cloud map is constrained within the bounding box set. A graph search algorithm is used to perform graph search on the sliding point cloud map to generate a front-end search path.

8. A UAV trajectory planning device for unknown environments based on incremental topology maps, characterized in that, include: Geometric feature point generation module: configured to allow the drone to acquire current point cloud data in real time before it reaches the end of its flight mission; The sliding point cloud map is updated in real time based on the current point cloud data; The current geometric feature point set is obtained based on the current point cloud data. The geometric feature points include near-end channel feature points and far-end feature points. Front-end search module: configured to preprocess the geometric feature points in the current geometric feature point set to obtain the first current geometric feature point set; Delete redundant geometric feature points from the first set of current geometric feature points to obtain the second set of current geometric feature points; Establish connections between the elements of the second current geometric feature point set and some existing topological nodes in the current topological map to obtain the first current topological map; Based on the current set of geometric feature points and the first current topology map, update the set of guiding nodes, calculate the guiding cost of the guiding nodes, and select the guiding node with the lowest guiding cost as the optimal node; determine the front-end search path corresponding to the first current topology map based on the optimal node; The trajectory generation module is configured to construct a safe flight corridor using the front-end search path as the line segment seed. Under the constraint of the safe flight corridor, the module optimizes the motion trajectory of the UAV from its current position to the optimal node along the front-end search path, taking the current position of the UAV as the starting point, and determines the flight position trajectory and yaw angle trajectory of the UAV from its current position to the guide node.

9. An electronic device, characterized in that, The electronic device includes: A processor is used to execute multiple instructions; Memory, used to store multiple instructions; The plurality of instructions are to be stored in the memory and loaded by the processor and executed as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method as described in any one of claims 1-7.