Method and device for generating a patrol path based on low-altitude flight
By constructing a 3D voxel map and a convex polyhedral safety corridor on the ground or in the cloud, and generating a set of linear inequalities, the problem of insufficient obstacle avoidance reliability and computational performance of path planning schemes in complex underground environments in existing technologies is solved, and efficient and safe underground inspection tasks are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2026-05-14
- Publication Date
- 2026-06-12
AI Technical Summary
Existing path planning technologies in complex underground environments have shortcomings in balancing real-time obstacle avoidance reliability with reducing the computational performance requirements of aircraft, especially when facing dynamic obstacles where collisions are likely and computational resources are limited.
By constructing a 3D voxel map and a convex polyhedral safety corridor on the ground or in the cloud, a system of linear inequalities is generated. The aircraft performs simple algebraic operations during flight to avoid obstacles in real time. Combined with lightweight constraint verification, the reliance on the onboard processor is reduced.
It significantly reduces the reliance on onboard processor computing power, improves obstacle avoidance reliability and system stability in dynamic environments, can sensitively detect temporary obstacles, and enhances the robustness and safety of aircraft operation in complex environments.
Smart Images
Figure CN122192332A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of low-altitude flight path planning technology, specifically relating to a method and device for generating inspection paths based on low-altitude flight. Background Technology
[0002] With the rapid development of modern infrastructure construction, the scale of enclosed and narrow spaces such as underground utility tunnels, rail transit tunnels, and large mines is increasing daily. The safety maintenance and regular inspection of these facilities have become crucial for ensuring urban operation and industrial production. Because these environments typically feature limited space, poor lighting conditions, and severe electromagnetic shielding, traditional manual inspection methods are not only labor-intensive and inefficient, but also pose challenges to the personal safety of inspection personnel in extreme situations such as hazardous gas leaks or structural instability. Against this backdrop, utilizing highly mobile low-altitude flight platforms—especially drones with autonomous navigation capabilities—carrying various sensors to conduct intelligent inspections deep into underground spaces has gradually become an industry consensus and an important direction for technological evolution.
[0003] In existing technological implementations, flight path planning schemes for complex underground environments can be mainly categorized into two types. The first type focuses on utilizing pre-built offline global maps and extracting digital twin models or high-precision laser point cloud data of the underground space to pre-calculate a guidance trajectory in a static environment using algorithms. This method performs well in ideal conditions where the environment is static and the structure is stable, providing clear guidance for the aircraft. However, underground spaces are often not absolutely static; temporary scaffolding erected during maintenance, scattered material piles, and even personnel and vehicles operating within the space all constitute dynamic obstacles. Once the actual working conditions deviate from the representation of the offline map, the aircraft, limited by its real-time local adjustment capabilities, is prone to collision risks, and the system's fault tolerance and robustness have room for optimization in dynamic scenarios. To improve the flexibility limitations of offline planning, the second type of scheme shifts to fully autonomous online environmental perception and real-time path reconstruction. However, this type of scheme typically requires the aircraft to be equipped with high-performance computing units, placing high demands on the aircraft's computational performance.
[0004] In summary, the existing technology has the following problems: in complex underground inspection environments, the existing path planning schemes still need improvement in balancing the reliability of real-time obstacle avoidance with reducing the computational performance requirements of the aircraft. Summary of the Invention
[0005] To address the aforementioned problems in the existing technology, this invention provides a method and device for generating inspection paths based on low-altitude flight. The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides a method for generating inspection paths based on low-altitude flight, comprising: Acquire three-dimensional point cloud data of the underground space to be inspected, and use the three-dimensional point cloud data to construct a three-dimensional voxel map; Using the three-dimensional voxel map and the minimum safe passage radius of the aircraft, the initial feasible airspace of the aircraft is determined; In the initial feasible airspace, the inspection start point and end point are set, and a global path search is performed in the initial feasible airspace to obtain a global initial path composed of multiple discrete path points. Multiple ordered smooth path segments are generated using the global initial path, and these multiple ordered smooth path segments constitute the inspection path of the aircraft. Convex polyhedra containing each smooth path segment are constructed respectively, wherein each convex polyhedron is a set of linear inequalities, and the convex polyhedra of the multiple ordered smooth path segments constitute the virtual safety corridor of the underground space. The inspection path and the set of linear inequalities are sent to the aircraft for local storage. During the inspection of the underground space according to the inspection path, the aircraft uses the set of linear inequalities of the smooth path segment to which its current position belongs as the constraint condition for each obstacle point it detects. When a preset number of obstacle points among the currently detected obstacle points make the constraint condition true, it indicates that the obstacle has intruded into the virtual safety corridor and obstacle avoidance is performed. Otherwise, the inspection continues according to the inspection path.
[0006] The present invention also provides an inspection path generation device based on low-altitude flight, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the above-mentioned inspection path generation method based on low-altitude flight.
[0007] The present invention also provides a low-altitude flight device, which pre-stores an inspection path and a set of linear inequalities for an underground space to be inspected, generated using the aforementioned low-altitude flight-based inspection path generation method. During the inspection of the underground space, the low-altitude flight device uses the set of linear inequalities of the smooth path segment to which its current position belongs as constraints for each currently detected obstacle. If a preset number of currently detected obstacles satisfy the constraints, it indicates that an obstacle has intruded into the virtual safety corridor, and obstacle avoidance is initiated. Otherwise, it indicates that the obstacle has not intruded into the virtual safety corridor, and the inspection continues according to the inspection path.
[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention simplifies complex spatial topological relationships into a system of linear inequalities that are easily processed by the aircraft by pre-completing complex underground space modeling and convex polyhedral safety corridor construction on the ground or in the cloud. During flight, the aircraft does not need to perform large-scale environmental reconstruction or complex path replanning calculations; it can determine the safety of the current environment simply through algebraic operations. Compared to solutions that require running synchronous positioning and mapping algorithms and real-time trajectory optimization algorithms on the airborne end, this invention significantly reduces the dependence on the computing power of the airborne processor, enabling small inspection aircraft with limited computing resources to perform complex underground inspection tasks, thus possessing higher engineering adaptability.
[0009] (2) This invention provides a clear geometric safety boundary for the aircraft by constructing a virtual safety corridor based on a sequence of convex polyhedra. Compared to planning schemes that rely solely on offline static maps, this invention introduces a real-time verification mechanism based on a system of linear inequalities, which can sensitively detect temporary obstacles (such as construction equipment, personnel, etc.) intruding into the corridor, significantly improving obstacle avoidance reliability and system stability in dynamic and complex environments. Furthermore, since the obstacle avoidance logic is based on direct coordinate substitution calculations, the response speed is extremely fast, effectively compensating for the lag in offline planning when facing dynamic environmental changes. Simultaneously, by presetting a threshold for the number of obstacle points triggered by obstacle avoidance, false alarms caused by sensor noise are filtered out, improving the operational robustness of the flight system in complex electromagnetic and lighting environments.
[0010] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description
[0011] Figure 1 This is a flowchart illustrating the inspection path generation method based on low-altitude flight provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a scene of drone inspection of underground utility tunnels provided in an embodiment of the present invention. Detailed Implementation
[0012] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0013] In inspection tasks in complex environments such as underground spaces, utility tunnels, and mines, traditional global path planning often struggles to cope with dynamically appearing obstacles due to factors such as confined space, poor lighting conditions, and electromagnetic shielding. Purely real-time local obstacle avoidance can easily lead to the aircraft getting stuck in local optima or causing collisions due to insufficient computing power. Therefore, this invention provides a method and device for generating inspection paths based on low-altitude flight. By constructing a virtual safety corridor offline based on a system of linear inequalities and combining it with online lightweight constraint verification, efficient and safe inspections in confined spaces are achieved. For example, this method can be executed by a cloud server.
[0014] Figure 1 This is a flowchart illustrating a method for generating inspection paths based on low-altitude flight, as provided in an embodiment of the present invention. Figure 1 As shown, the method includes: S101. Obtain the three-dimensional point cloud data of the underground space to be inspected, and use the three-dimensional point cloud data to construct a three-dimensional voxel map.
[0015] Here, the three-dimensional point cloud data of the underground space to be inspected is the original three-dimensional point cloud data of the underground space collected by lidar or depth camera.
[0016] Specifically, S101 is implemented through steps S1011 to S1013: S1011. Transform the coordinates of each point in the 3D point cloud data to the world coordinate system, and determine the spatial range of the underground space based on the minimum and maximum values of the coordinates of the 3D point cloud data after coordinate transformation.
[0017] Since the data collected by sensors is usually located in the sensor's own local coordinate system, a coordinate transformation operation must first be performed. By acquiring the real-time pose information provided by the aircraft's positioning system, each frame of local point cloud is transformed to a unified world coordinate system. Then, the maximum and minimum values of all point cloud coordinates in the X, Y, and Z axes of the world coordinate system are calculated to determine the overall bounding box range of the underground space to be inspected.
[0018] S1012. Based on the spatial range, an octree is used to divide the underground space into multiple voxels. Each voxel has a location coordinate. Furthermore, each voxel is determined to be an idle voxel or an voxel occupied by an obstacle based on whether it contains point cloud points. When a voxel contains point cloud points, the voxel's state value is 1, indicating that the voxel is occupied by an obstacle. When a voxel does not contain point cloud points, the voxel's state value is 0, indicating that the voxel is an idle voxel.
[0019] Specifically, an octree data structure is used to progressively divide the space within the bounding box. The core of the octree is to recursively divide the 3D space into eight equal sub-cubes (octets) until a preset minimum voxel resolution is reached, for example, 0.1m × 0.1m × 0.1m. During octree initialization, the entire bounding box is used as the root node, and then the octree is progressively subdivided downwards according to its splitting rules until the size of the child node equals the minimum voxel resolution (0.1m × 0.1m × 0.1m). For each voxel divided by the octree, if the voxel contains point cloud points, it is considered occupied by an obstacle, and its state value is set to 1; if the voxel does not contain point cloud points, it is considered free space, and its state value is set to 0. Furthermore, each voxel has a position coordinate, which is the geometric center coordinate of the voxel in the world coordinate system. Since this technique is existing, it will not be elaborated upon in this invention.
[0020] S1013. Associate the position coordinates of each voxel with its state value. All voxels associated with the position coordinates and state values constitute a three-dimensional voxel map of the underground space.
[0021] The octree-based storage method greatly compresses the amount of representational data for underground space, laying the foundation for subsequent fast queries.
[0022] S102. Using a three-dimensional voxel map and the minimum safe passage radius of the aircraft, determine the initial feasible airspace of the aircraft.
[0023] Specifically, S102 is implemented through steps S1021 to S1022: S1021. Calculate the Euclidean distance between each free voxel and the nearest voxel occupied by an obstacle in the three-dimensional voxel map, and obtain the minimum Euclidean distance corresponding to each free voxel.
[0024] Specifically, for each free voxel, the Euclidean distance between the free voxel and each voxel occupied by obstacles is calculated, and the smallest Euclidean distance is determined to obtain the minimum Euclidean distance corresponding to the free voxel.
[0025] S1022. Remove the idle voxels whose minimum Euclidean distance is less than the minimum safe passage radius of the aircraft from the three-dimensional voxel map, and take the set of the remaining and interconnected idle voxels in the three-dimensional voxel map as the initial feasible airspace of the aircraft; wherein, the minimum safe passage radius of the aircraft is the sum of the maximum physical envelope radius of the aircraft and the preset safety redundancy.
[0026] Specifically, the minimum safe passage radius of the aircraft is determined by comprehensively considering both the aircraft's maximum physical envelope radius and the safety redundancy set for sensor errors and dynamic disturbances. For example, the maximum physical envelope radius of the aircraft is half the diagonal distance of the rotor blades after deployment; the safety redundancy can be set according to the wind speed and positioning accuracy in the underground space to be inspected. For example, if the diagonal distance of the rotor blades after deployment is 0.6m and the safety redundancy is set to 0.2m, the minimum safe passage radius of the aircraft is 0.5m.
[0027] Specifically, after removing idle voxels whose minimum Euclidean distance is less than the minimum safe passage radius of the aircraft from the 3D voxel map, the remaining idle voxels are obtained. Next, a 26-neighborhood definition is used (i.e., if two voxel cubes have any face, edge, or corner in contact, they are considered connected), and a breadth-first search algorithm is used to find interconnected idle voxels among the remaining idle voxels. These interconnected idle voxels constitute connected components. Finally, the connected component with the largest volume is selected as the initial feasible airspace of the aircraft. Since this technique is existing, it will not be elaborated upon in this invention.
[0028] S103. Set the inspection start point and end point in the initial feasible airspace, and perform a global path search in the initial feasible airspace to obtain a global initial path composed of multiple discrete path points.
[0029] For example, the improved Rapidly-exploring Random TreesStar (RRT*) algorithm can be used to perform a global path search in the initial feasible space, or other improved RRT algorithms can be used to perform a global path search in the initial feasible space, which will not be elaborated in this invention.
[0030] In some embodiments, the specific process of performing a global path search in the initial feasible spatial domain is as follows: S1. Create a random tree T with a root node P, a set of nodes V, and a set of edges E. Initialize a cost record, which records the cumulative cost of each node. The cumulative cost of a node is the cost from the start of the inspection. The cumulative cost to this node, where the root node is the inspection starting point. The nodes in the initial node set V are the starting points for the inspection. The initial edge set E is empty; any node The expression for the cumulative cost is: ,in, and These are the weighting coefficients, It is 0.6. It is 0.4. Indicates the starting point of the inspection To the node The cumulative path length, For nodes The Euclidean distance indicates the distance from that location to the nearest obstacle.
[0031] S2. Within the boundary of the initial feasible space, with probability... Select the sampling points to obtain the sampling points. , This indicates that the sampling point is the end point of the inspection. The probability of (i.e., the target bias probability); S3. Among all nodes of the current random tree T, find the distance from the sampling point. The nearest node From node To sampling point Directional expansion preset step size , obtain the expansion point Check by node With extension point The edges formed Whether it is completely contained within the initial feasible space, where the checking method is to check the edges. Discrete sampling is performed on the points, where the sampling step size is less than or equal to the voxel resolution, and the edges are verified. Does each sampling point on the edge belong to the initial feasible spatial region? If the space is not completely contained within the initial feasible space, return to step S2 above and continue execution; otherwise, execute step S4 below.
[0032] S4. Find all distance expansion points in the current random tree T. The distance is less than the preset neighborhood radius The nodes are obtained to form a node set. ; S5. In the node set In the middle, select to start from the inspection starting point To the extension point The node with the lowest cumulative cost is selected as the parent node. and expand the points Add the edges to the node set V. Add the edge set E and calculate the edge from the inspection starting point. To the parent node The cumulative cost And calculate the parent node With extension point Euclidean distance between ,Will and The sum as the extension point The latest cumulative cost ; S6. For the set of nodes Except for the parent node Every node outside Calculate the starting point of the inspection To the node The cumulative cost and extension points With nodes Euclidean distance between ,judge and Is the sum less than If so, then the node The parent node is changed and update and update the edge set E; S7. Computation Node Inspection Termination Point The Euclidean distance between them, if the Euclidean distance is less than or equal to the preset step size. Then the inspection will end at the designated point. Add it as a new node to the random tree T and create an edge. , will node As the end point of the inspection The parent node, and update ,in, yes and The sum of these elements completes the search, and the root node is obtained as the starting point for the inspection. The endpoint is the end point of the inspection. The random tree T; conversely, if the Euclidean distance is greater than the preset step size. If the above S2 is not executed, the process continues until the root node is identified as the starting point for the inspection. The leaf node is the end point of the inspection. A random tree T, or, until the preset number of iterations is reached; S8. After obtaining the final random tree T, inspect the starting point. Inspection Termination Point and from the starting point of the inspection To the end point of the inspection The polyline formed by the nodes traversed serves as the global initial path. It can be represented as ,in, for , for , It is the final random tree T starting from the inspection start point To the end point of the inspection The nodes traversed.
[0033] S104. Utilize the global initial path to generate multiple ordered smooth path segments, which together constitute the aircraft's inspection path.
[0034] Specifically, interpolation methods such as quintic spline curves or polynomial trajectory optimization can be used to smooth the generated global initial path, thereby obtaining a smooth path. This invention does not limit the specific interpolation method selected. Smoothing eliminates abrupt curvature changes at path inflections, making the path smooth and thus satisfying the continuity of velocity and acceleration to adapt to the motion constraints of the aircraft. After obtaining a smooth path, it is then segmented to obtain multiple ordered smooth path segments. It should be noted that the purpose of segmentation is to make each segment approximately a straight line. In some embodiments, to ensure good continuity of the subsequently constructed virtual safety corridor and provide a smooth constraint switching buffer for the inspection aircraft during cross-segment flight, there can be 10% to 15% overlap between adjacent smooth path segments.
[0035] S105. Construct convex polyhedra containing each smooth path segment, wherein each convex polyhedron is a set of linear inequalities, and multiple ordered convex polyhedra of smooth path segments constitute a virtual safety corridor in the underground space.
[0036] Specifically, the above S105 is achieved through steps S1051 to S1055: S1051, For each smooth path segment Select smooth path segments The midpoint is the reference point. , with reference point Centered on, with Construct a cube as the seed convex polyhedron with radius r, and use the seed convex polyhedron as the current convex polyhedron. And initialize an empty set of hyperplanes, where, The preset empirical coefficient is less than 1. For reference point The Euclidean distance between the free voxel and the nearest voxel occupied by an obstacle.
[0037] For example, The value ranges from 0.1 to 0.3.
[0038] S1052, in the current convex polyhedron Around the boundary, find a point that will make the current convex polyhedron... Hyperplane separated from the nearest obstacle and the hyperplane Add to the hyperplane set.
[0039] Specifically, for the current convex polyhedron For each vertex, the search direction is the direction along which the vertex points outward from the convex polyhedron (i.e., away from the center). Perform a ray search until a voxel occupied by an obstacle is encountered. Record the boundary point of the voxel occupied by the obstacle (called the obstacle boundary point). The coordinates of the boundary point of a voxel are represented as follows: , These are the position coordinates of the voxel. For voxel resolution, and m, , and respectively search direction The symbols on the X, Y, and Z axes of the world coordinate system. , respectively search direction The components on the X, Y, and Z axes of the world coordinate system, and, for , for , for ,and, , , It is a sign function. Based on the current convex polyhedron. A hyperplane is determined by solving a semidefinite programming subproblem, which identifies the boundary points of all obstacles corresponding to the vertices of the hyperplane. The hyperplane obtained by solving This enables the current convex polyhedron The hyperplane is completely located on one side of the hyperplane, and all obstacle boundary points corresponding to its vertices are located on the other side of the hyperplane. The solved hyperplane... Add to the hyperplane set.
[0040] S1053. Using all hyperplanes in the current set of hyperplanes as constraints, maximize the current convex polyhedron by solving a semidefinite programming problem. The volume is used to obtain a new convex polyhedron after expansion. .
[0041] Specifically, the desired new convex polyhedron after expansion will be determined. Parameterized as: ,in, Represents any coordinate point in three-dimensional space. This represents the number of hyperplanes in the current set of hyperplanes. Indicates the first Normal vectors of the hyperplane, T It is the transpose symbol. Indicates the first The offsets of the hyperplanes to be optimized. To approximately maximize the volume of the convex polyhedron, a strategy of maximizing the volume of its inscribed ellipsoid is adopted, and the following semidefinite programming problem is constructed: ,in, The shape matrix of the ellipsoid. Centered at the ellipsoid Represents the L2 norm. express The determinant value, Represents a logarithmic determinant. The constraints are represented. The optimal ellipsoid is obtained by solving this semidefinite programming problem, and then the offset can be determined based on the relationship between the optimal ellipsoid and each hyperplane. Thus, a new convex polyhedron is obtained after expansion. Specifically, after obtaining the optimal ellipsoid, for the normal vector... The ellipsoid along the normal vector The support function value is This value represents the ellipsoid in the direction The signed distance between the farthest boundary point on the ellipsoid and the origin (or reference point). To ensure that the convex polyhedron is exactly inscribed within the ellipsoid (i.e., each hyperplane is a supporting hyperplane of the ellipsoid), we should take... At this point, the ellipsoid lies entirely within the convex polyhedron, and each hyperplane is tangent to the ellipsoid. This is how it is determined. This approach ensures that the convex polyhedron contains the ellipsoid (thus maximizing the volume) while avoiding wasted volume due to an excessively distant hyperplane. It should be noted that constraints need to be added during the solution process to ensure smooth path segments. All points on are located Internally, the constraint is ,in, It is a smooth path segment Any sampling point on the surface.
[0042] S1054. Determine the new convex polyhedron after expansion. Compared to current convex polyhedra The volume increment.
[0043] S1055. When the volume increment is less than the preset increment threshold, the expanded new convex polyhedron will be... As the final constructed component containing smooth path segments convex polyhedron Otherwise, the expanded new convex polyhedron As the new current convex polyhedron Then return to S1052 above and continue execution until the final constructed smooth path segment is obtained. convex polyhedron .
[0044] In some embodiments, when the number of iterations of the above process reaches a preset number, the newly expanded convex polyhedron obtained in the last iteration can be... As the final constructed component containing smooth path segments convex polyhedron For example, the preset number of times can be 32 or 64, and the present invention does not limit the value of the preset number of times.
[0045] For example, the final constructed result contains smooth path segments. convex polyhedron Represented as , and, when Depend on When composed of multiple hyperplanes Expressed using the following system of linear inequalities: ; , ; in, Indicates the first Normal vectors of the hyperplane, Indicates the first The offset of each hyperplane, It is the transpose symbol. .
[0046] S106. The inspection path and the set of linear inequalities are sent to the aircraft for local storage. During the inspection of the underground space according to the inspection path, the aircraft uses the set of linear inequalities of the smooth path segment to which its current position belongs as the constraint condition for each obstacle point it detects. When a preset number of obstacle points among the currently detected obstacle points make the constraint condition true, it indicates that the obstacle has intruded into the virtual safety corridor and obstacle avoidance is performed. Otherwise, the inspection continues according to the inspection path.
[0047] Specifically, the inspection path of the underground space to be inspected, along with a set of linear inequalities constraining that path, is constructed and then sent to the aircraft for local storage. When the aircraft is performing its inspection mission, it flies along this path and simultaneously scans the surrounding environment at a preset frequency using its environmental sensing sensors (e.g., lidar) to acquire point cloud data. This point cloud data is then converted to the world coordinate system. The aircraft, equipped with an inertial measurement unit and a multi-line lidar, uses a point cloud matching algorithm to obtain the relative pose to the starting point. This is combined with known reference points in the underground utility tunnel for cumulative error correction, thereby enabling the acquisition of the real-time pose in the world coordinate system. The specific principle behind this is known and will not be elaborated upon here. After obtaining its own position, the drone retrieves a set of linear inequalities from the smooth path segment to which its position belongs, and verifies whether each point cloud point obtained makes the retrieved set of linear inequalities true. If a preset number of point cloud points make the retrieved set of linear inequalities true, it indicates that an obstacle has intruded into the virtual safety corridor of the current area, and the drone performs corresponding obstacle avoidance processing. The specific obstacle avoidance range can be determined based on the number of point cloud points that make the retrieved set of linear inequalities true. If the number of point cloud points that make the retrieved set of linear inequalities true does not meet the preset number, it indicates that the obstacle has not intruded into the virtual safety corridor of the current area, and the drone continues to inspect subsequent areas according to the inspection path. It should be noted that the specific value of the preset number can be set according to actual needs, such as 1, 2, or 5, etc., and this invention does not limit this. It should also be noted that this invention does not limit the principle of how the aircraft avoids obstacles; existing obstacle avoidance mechanisms can be used.
[0048] The present invention also provides an inspection path generation device based on low-altitude flight, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the above-described method for generating inspection paths based on low-altitude flight.
[0049] This invention also provides a low-altitude flight device. The low-altitude flight device pre-stores an inspection path and a set of linear inequalities for an underground space to be inspected, generated using the aforementioned low-altitude flight-based inspection path generation method. During the inspection of the underground space, the low-altitude flight device uses the set of linear inequalities of the smooth path segment to which its current position belongs as constraints for each currently detected obstacle. If a preset number of obstacles among the currently detected obstacles satisfy the constraints, it indicates that an obstacle has intruded into the virtual safety corridor, and obstacle avoidance is initiated. Otherwise, it indicates that the obstacle has not intruded into the virtual safety corridor, and the inspection continues according to the inspection path. For example... Figure 2 This is a schematic diagram illustrating a scenario where a drone inspects an underground utility tunnel. (Example) Figure 2 As shown, when the aircraft is a drone and the underground area to be inspected is an underground utility tunnel, the drone obtains the inspection path and set of linear inequalities of the underground utility tunnel to be inspected from the cloud server in advance and stores them locally. Then, it uses the locally stored inspection path and set of linear inequalities of the underground utility tunnel to perform the inspection task of the underground utility tunnel.
[0050] The present invention has the following beneficial effects: (1) This invention simplifies complex spatial topological relationships into a system of linear inequalities that are easily processed by the aircraft by pre-completing complex underground space modeling and convex polyhedral safety corridor construction on the ground or in the cloud. During flight, the aircraft does not need to perform large-scale environmental reconstruction or complex path replanning calculations; it can determine the safety of the current environment simply through algebraic operations. Compared to solutions that require running synchronous positioning and mapping algorithms and real-time trajectory optimization algorithms on the airborne end, this invention significantly reduces the dependence on the computing power of the airborne processor, enabling small inspection aircraft with limited computing resources to perform complex underground inspection tasks, thus possessing higher engineering adaptability.
[0051] (2) This invention provides a clear geometric safety boundary for the aircraft by constructing a virtual safety corridor based on a sequence of convex polyhedra. Compared to planning schemes that rely solely on offline static maps, this invention introduces a real-time verification mechanism based on a system of linear inequalities, which can sensitively detect temporary obstacles (such as construction equipment, personnel, etc.) intruding into the corridor, significantly improving obstacle avoidance reliability and system stability in dynamic and complex environments. Furthermore, since the obstacle avoidance logic is based on direct coordinate substitution calculations, the response speed is extremely fast, effectively compensating for the lag in offline planning when facing dynamic environmental changes. Simultaneously, by presetting a threshold for the number of obstacle points triggered by obstacle avoidance, false alarms caused by sensor noise are filtered out, improving the operational robustness of the flight system in complex electromagnetic and lighting environments.
[0052] (3) This invention utilizes an improved fast search random tree algorithm combined with smooth segmentation processing to generate inspection paths that not only satisfy the geometric constraints of narrow underground spaces but also conform to the dynamic characteristics of aircraft. By customizing convex polyhedral constraints for each path segment, the shape of the safety corridor can adaptively adjust to the orientation of the underground utility tunnel or tunnel. This segmentation constraint mechanism ensures that the aircraft remains within a predefined physical safety zone during long-distance inspections, avoiding the risk of course deviation due to global cumulative errors, and maintaining stable inspection performance even in extremely narrow spaces. Thus, it can maintain high-precision path tracking and safety assurance capabilities in confined spaces.
[0053] (4) This invention decouples global path planning from local security verification, realizing the technical logic of "offline computing power for online efficiency". When the aircraft performs inspections, it forms a real-time "safety shield" through a set of locally stored linear inequalities. This mechanism avoids the risk of system crash due to communication interruption or insufficient computing power on the edge. After detecting obstacle intrusion, timely obstacle avoidance rather than blindly executing the preset path greatly reduces the probability of collision damage to the aircraft, thereby ensuring the integrity of the underground inspection task and improving the automation level of the system in an unattended environment.
[0054] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.
[0055] In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. While different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce a good effect.
[0056] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A method for generating inspection paths based on low-altitude flight, characterized in that, include: Acquire three-dimensional point cloud data of the underground space to be inspected, and use the three-dimensional point cloud data to construct a three-dimensional voxel map; Using the three-dimensional voxel map and the minimum safe passage radius of the aircraft, the initial feasible airspace of the aircraft is determined; In the initial feasible airspace, the inspection start point and end point are set, and a global path search is performed in the initial feasible airspace to obtain a global initial path composed of multiple discrete path points. Multiple ordered smooth path segments are generated using the global initial path, and these multiple ordered smooth path segments constitute the inspection path of the aircraft. Convex polyhedra containing each smooth path segment are constructed respectively, wherein each convex polyhedron is a set of linear inequalities, and the convex polyhedra of the multiple ordered smooth path segments constitute the virtual safety corridor of the underground space. The inspection path and the set of linear inequalities are sent to the aircraft for local storage. During the inspection of the underground space according to the inspection path, the aircraft uses the set of linear inequalities of the smooth path segment to which its current position belongs as the constraint condition for each obstacle point it detects. When a preset number of obstacle points among the currently detected obstacle points make the constraint condition true, it indicates that the obstacle has intruded into the virtual safety corridor and obstacle avoidance is performed. Otherwise, the inspection continues according to the inspection path.
2. The inspection path generation method based on low-altitude flight according to claim 1, characterized in that, The construction of a 3D voxel map using the 3D point cloud data includes: The coordinates of each point in the three-dimensional point cloud data are transformed to the world coordinate system, and the spatial range of the underground space is determined based on the minimum and maximum values of the coordinates of the three-dimensional point cloud data after coordinate transformation. Based on the spatial range, an octree is used to divide the underground space into multiple voxels. Each voxel has a location coordinate. Furthermore, each voxel is determined to be either an idle voxel or an voxel occupied by an obstacle based on whether it contains point cloud points. When a voxel contains point cloud points, the voxel's state value is 1, indicating that the voxel is occupied by an obstacle. When a voxel does not contain point cloud points, the voxel's state value is 0, indicating that the voxel is an idle voxel. The position coordinates of each voxel are associated with its state value. All voxels associated with the position coordinates and state values constitute a three-dimensional voxel map of the underground space.
3. The inspection path generation method based on low-altitude flight according to claim 2, characterized in that, The process of determining the initial feasible airspace of the aircraft using the three-dimensional voxel map and the minimum safe passage radius of the aircraft includes: Calculate the Euclidean distance between each free voxel and the nearest voxel occupied by an obstacle in the three-dimensional voxel map to obtain the minimum Euclidean distance corresponding to each free voxel. Empty voxels whose minimum Euclidean distance is less than the minimum safe passage radius of the aircraft are removed from the three-dimensional voxel map, and the set of the remaining and interconnected empty voxels in the three-dimensional voxel map is taken as the initial feasible airspace of the aircraft. The minimum safe passage radius of the aircraft is the sum of the maximum physical envelope radius of the aircraft and the preset safety redundancy.
4. The inspection path generation method based on low-altitude flight according to claim 1, characterized in that, The process of generating multiple ordered smooth path segments using the global initial path includes: The global initial path is smoothed to obtain a smoothed path; The smooth path is segmented to obtain multiple ordered smooth path segments.
5. The inspection path generation method based on low-altitude flight according to claim 2, characterized in that, The construction of convex polyhedra containing each smooth path segment includes: S1, for each smooth path segment Select the smooth path segment The midpoint is the reference point. With the aforementioned reference point Centered on, with Construct a cube as the seed convex polyhedron with radius r, and use the seed convex polyhedron as the current convex polyhedron. And initialize an empty set of hyperplanes, where, The preset empirical coefficient is less than 1. The reference point The Euclidean distance between the available voxel and the nearest voxel occupied by an obstacle; S2, in the current convex polyhedron Around the boundary, find a point that will hold the current convex polyhedron. Hyperplane separated from the nearest obstacle and the hyperplane Add the hyperplane set; S3. Using all hyperplanes in the current set of hyperplanes as constraints, maximize the current convex polyhedron by solving a semidefinite programming problem. The volume is used to obtain a new convex polyhedron after expansion. ; S4. Determine the new convex polyhedron after expansion. Compared to the current convex polyhedron The volume increment; S5. When the volume increment is less than a preset increment threshold, the expanded new convex polyhedron... As the final constructed component containing smooth path segments convex polyhedron Otherwise, the expanded new convex polyhedron As the new current convex polyhedron Then return to S2 above and continue execution until the final constructed smooth path segment is obtained. convex polyhedron .
6. The inspection path generation method based on low-altitude flight according to claim 5, characterized in that, The convex polyhedron Expressed using a system of linear inequalities: ； ， ； in, Represents any coordinate point in three-dimensional space. Indicates the first Normal vectors of the hyperplane, Indicates the first The offset of each hyperplane, It is the transpose symbol. , The convex polyhedron constitutes the structure of the polyhedron. The number of hyperplanes.
7. The inspection path generation method based on low-altitude flight according to claim 4, characterized in that, Among the plurality of ordered smooth path segments, there is an overlap between two adjacent smooth path segments.
8. A patrol path generation device based on low-altitude flight, comprising a processor, a communication interface, a memory, and a communication bus, characterized in that, The processor, the communication interface, and the memory communicate with each other via the communication bus; The memory is used to store computer programs; When the processor executes a program stored in the memory, it implements the steps of the method described in any one of claims 1-7.
9. A low-altitude flight device, characterized in that, The low-altitude flight equipment pre-stores inspection paths and a set of linear inequalities for the underground space to be inspected, generated using any one of the methods described in claims 1-7. During the inspection of the underground space, the low-altitude flight equipment uses the set of linear inequalities of the smooth path segment to which its current position belongs as constraints for each obstacle point it currently detects. If a preset number of obstacle points among the currently detected obstacle points satisfy the constraints, it indicates that the obstacle has intruded into the virtual safety corridor, and obstacle avoidance is performed. Otherwise, it indicates that the obstacle has not intruded into the virtual safety corridor, and the inspection continues according to the inspection path.