Unmanned aerial vehicle route planning method for hydropower engineering

Abstract

The invention provides an unmanned aerial vehicle route planning method for hydropower engineering, and relates to the technical field of hydropower engineering safety monitoring. The method comprises the steps of generating a three-dimensional terrain grid model corresponding to a target operation area, performing voxelization processing to obtain an initial voxel map, and determining a first candidate waypoint by combining the initial voxel map; based on the average normal vector distribution of the initial voxel map, segmenting and refining the voxel units to generate a fine voxel map, and determining second candidate waypoints in combination with the fine voxel map; performing collision detection on all candidate waypoints to obtain a target waypoint set; and generating an unmanned aerial vehicle operation route according to the target waypoint set. According to the scheme, the multi-level voxel map is constructed and optimized, candidate waypoint screening and space collision detection are combined, and unmanned aerial vehicle patrol monitoring route planning which adapts to complex terrains and is complete in coverage and high in feasibility is achieved at low calculation cost.

Description

Technical Field

[0001] The present disclosure relates to the technical field of hydropower project safety monitoring, and in particular to a method for planning unmanned aerial vehicle (UAV) routes for hydropower projects. Background Art

[0002] As vital infrastructure for public well-being, water conservancy and hydropower projects play a key role in social development. With growing societal demands, higher requirements are being placed on the construction, operation, and management of these projects, particularly in terms of project inspection and maintenance. For a long time, monitoring and inspection of hydraulic structures and reservoir bank slopes during hydropower station construction and operation has faced challenges such as heavy workloads, the difficulty of manual inspections, and numerous loopholes. In complex terrain, traditional inspection and monitoring methods are no longer able to meet the operational requirements of modern hydropower projects due to operational difficulties, low accuracy, limited measurement results, high costs, and high risks. The introduction of drone technology offers a new solution for the inspection and monitoring of hydropower projects. Its advantages include a wide inspection range, high automation, and strong flexibility, effectively reducing the safety risks, labor costs, and time associated with manual inspections. Therefore, the refined flight path design for drone inspection and monitoring tasks in hydropower projects has become a critical issue that needs to be addressed to further improve the efficiency and quality of hydropower project operation and management.

[0003] Conventional route planning methods struggle to adapt to the complex and ever-changing terrain and environmental characteristics of hydropower projects using drones. Due to interference from factors such as occlusion, traditional methods often fail to fully capture information about the target area. In recent years, route planning methods based on geometric learning algorithms, by integrating geometric distances with environmental threat models, have demonstrated applicability in complex environments. However, this method relies heavily on the accuracy and completeness of the flight area model and is computationally complex, limiting its widespread application in practical engineering. Meanwhile, route planning methods based on optimal photogrammetry can significantly improve the quality of 3D model reconstruction in heavily occluded areas and enhance data acquisition efficiency by optimizing viewpoint selection. However, this method is computationally complex, and the increased number of waypoints increases data acquisition time. Furthermore, it places high demands on hardware performance. Furthermore, while simplifying the slope digital elevation model (DEM) and extracting a slope datum can improve the efficiency of optimal photogrammetry of slope terrain, the computational complexity remains high and the quality requirements for the DEM are extremely stringent.

[0004] In summary, relevant technologies have many shortcomings in efficiently and accurately planning routes in the complex environments of hydropower projects, making it difficult to fully meet actual needs. Therefore, it is urgent to propose a route planning method for hydropower project drone inspection and monitoring tasks, aiming to generate detailed inspection and monitoring routes that can obtain complete information at a low computational cost.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the Invention

[0006] The purpose of the embodiments of the present disclosure is to provide a drone route planning method for hydropower projects, a drone route planning device for hydropower projects, an electronic device, and a computer-readable storage medium, so as to achieve drone inspection and monitoring route planning that adapts to complex terrain, has complete coverage, and is highly feasible at a low computational cost.

[0007] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by practice of the present disclosure.

[0008] According to a first aspect of an embodiment of the present disclosure, a method for planning a route for a drone in a hydropower project is provided, comprising: Determine multiple target operation areas to be inspected and monitored, and generate a three-dimensional terrain grid model corresponding to each target operation area; voxelize the three-dimensional terrain grid model to construct an initial voxel map, and determine a first candidate waypoint based on the initial voxel map; Based on the average normal vector distribution of the initial voxel map, segment and refine each voxel unit in the initial voxel map to generate a refined voxel map, and determine a second candidate waypoint in combination with the refined voxel map; performing collision detection on the first candidate waypoint and the second candidate waypoint in combination with the refined voxel map to obtain a target waypoint set; An unmanned aerial vehicle (UAV) operation route is generated according to the target waypoint set, so as to control the UAV to inspect and test the corresponding target operation areas through the UAV operation route.

[0009] In some example embodiments of the present disclosure, based on the aforementioned solution, voxelizing the three-dimensional terrain grid model to construct an initial voxel map includes: Determine the initial voxel size of the basic voxel unit; Determining a minimum bounding box corresponding to the three-dimensional terrain grid model, and evenly dividing the minimum bounding box according to the initial voxel size to obtain a basic voxel map; A mapping relationship between the three-dimensional terrain grid model and each of the basic voxel units in the basic voxel map is constructed, and the three-dimensional terrain grid model is mapped to the basic voxel map based on the mapping relationship to obtain an initial voxel map.

[0010] In some example embodiments of the present disclosure, based on the aforementioned solution, determining the initial voxel size of the basic voxel unit includes: Obtain the flight parameters corresponding to the UAV, including scale parameters, pixel size, horizontal pixel number of the sensor, vertical pixel number of the sensor, heading overlap, and lateral overlap; An initial voxel size of the basic voxel unit is determined according to the flight parameters.

[0011] In some example embodiments of the present disclosure, based on the aforementioned solution, constructing a mapping relationship between the three-dimensional terrain grid model and each of the basic voxel units in the basic voxel map includes: Obtaining the center coordinates of each facet in the three-dimensional terrain grid model, and obtaining the minimum value of the vertex coordinates corresponding to the three-dimensional terrain grid model; Determine the index information of each of the model facets in the basic voxel map according to the facet center coordinates and the minimum value of the vertex coordinates; The index information is used as a mapping relationship between the three-dimensional terrain grid model and each of the basic voxel units in the basic voxel map.

[0012] In some example embodiments of the present disclosure, based on the aforementioned solution, determining the first candidate waypoint in combination with the initial voxel map includes: Obtaining a scale parameter corresponding to the three-dimensional terrain grid model and a lens focal length corresponding to the drone, and determining a moving distance according to the scale parameter and the lens focal length; Mark a target voxel unit at the top position in the initial voxel map, and move the voxel center coordinates of the target voxel unit in a target direction according to the moving distance to obtain a first candidate waypoint, wherein the target direction is the opposite direction of the observation direction at the first candidate waypoint.

[0013] In some example embodiments of the present disclosure, based on the aforementioned solution, segmenting and refining each voxel unit in the initial voxel map based on the average normal vector distribution of the initial voxel map to generate a refined voxel map includes: Based on the normal vectors of the model patches contained in each basic voxel unit in the initial voxel map, calculating the average normal vector corresponding to each basic voxel unit; screening voxel units to be refined in the initial voxel map according to the average normal vector; Determining a segmentation voxel size of the to-be-refined voxel unit according to the initial voxel size of the basic voxel unit, wherein the segmentation voxel size is half of the initial voxel size, and segmenting the to-be-refined voxel unit according to the segmentation voxel size to obtain a new basic voxel unit; The above steps are repeated until all basic voxel units in the initial voxel map meet the segmentation restriction conditions, thereby obtaining a refined voxel map.

[0014] In some example embodiments of the present disclosure, based on the aforementioned solution, screening voxel units to be refined in the initial voxel map according to the average normal vector includes: Calculating a first angle between the normal vector of each model facet in the basic voxel unit and the average normal vector; Determine the number of facets in the basic voxel unit whose first angle is greater than or equal to a preset angle threshold; If the ratio of the number of facets to the total number of facets in the basic voxel unit exceeds a preset ratio threshold, the basic voxel unit is determined to be a voxel unit to be refined.

[0015] In some example embodiments of the present disclosure, based on the aforementioned solution, the segmentation restriction condition is that the segmentation voxel size is smaller than one eighth of the initial voxel size.

[0016] In some example embodiments of the present disclosure, based on the aforementioned solution, determining the second candidate waypoint in combination with the refined voxel map includes: Calculating an average normal vector of each fine voxel unit of the fine voxel map, and determining a second angle between the average normal vector and a direction opposite to the observation direction in combination with an observation direction of the first candidate waypoint; determining the observability of each of the fine voxel units at each of the first candidate waypoints according to the second angle; Recording all observable fine voxel units into a visible voxel set through the observability, and recording all unobservable fine voxel units into an invisible voxel set; If the invisible voxel set corresponding to any first candidate waypoint is not an empty set, the voxel center of the fine voxel unit is offset by a moving distance along the average normal vector direction of the unobservable fine voxel unit to generate a new waypoint as a second candidate waypoint for supplementing the first candidate waypoint.

[0017] In some example embodiments of the present disclosure, based on the aforementioned solution, performing collision detection on the first candidate waypoint and the second candidate waypoint in combination with the refined voxel map to obtain a target waypoint set includes: Constructing a directed bounding box with the first candidate waypoint and the second candidate waypoint as centers; If the directed bounding box corresponding to any candidate waypoint intersects or contains a fine voxel unit in the fine voxel map, the candidate waypoint is determined to have a collision risk, and the first candidate waypoint and the second candidate waypoint are eliminated to obtain a target waypoint set.

[0018] According to a second aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the drone route planning method for hydropower engineering in the first aspect is implemented.

[0019] According to a third aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for planning a drone route for hydropower projects in the first aspect is implemented.

[0020] The technical solutions provided by the embodiments of the present disclosure may have the following beneficial effects: The UAV route planning method for hydropower engineering in the example embodiment of the present disclosure, on the one hand, performs spatial modeling based on a three-dimensional terrain grid model and establishes a structured voxel unit system in combination with voxelization processing, thereby achieving spatial discretization and index organization of the target operation area while ensuring the accuracy of spatial terrain feature expression, facilitating the rapid execution of subsequent waypoint extraction, visibility judgment and spatial analysis; further, through the voxel segmentation mechanism driven by the average normal vector, the voxel granularity can be adaptively adjusted to adapt to landforms or structural boundaries of different complexities, thereby taking into account both structural fidelity and computing resource control when constructing a waypoint set, reducing the traditional reliance on precise global modeling methods. On the other hand, by introducing visibility analysis and spatial collision detection mechanism, dynamic compensation and validity verification of waypoint selection strategy are realized. The first candidate waypoint and the second candidate waypoint are complementary based on the logical relationship between observable voxels and unobservable voxels. On the basis of achieving the integrity of terrain information acquisition, redundant waypoints with flight safety risks are eliminated through spatial interaction with the fine voxel map, thereby improving the effective density and executability of the overall waypoint set, improving the actual airworthiness and operational stability of the final route generation, and avoiding common problems in related technologies caused by misjudgment of occlusion, flight path crossing obstacles or insufficient coverage of the monitoring area.

[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0023] Figure 1 The present invention schematically illustrates a flow chart of a method for planning a route for a hydropower project using a drone according to some embodiments of the present disclosure.

[0024] Figure 2 The figure schematically shows a flow chart of constructing an initial voxel map according to some embodiments of the present disclosure.

[0025] Figure 3 A schematic diagram of determining a first candidate waypoint in combination with an initial voxel map according to some embodiments of the present disclosure is schematically shown.

[0026] Figure 4 The flowchart of constructing a refined voxel map based on an initial voxel map according to some embodiments of the present disclosure is schematically shown.

[0027] Figure 5 A schematic diagram of segmenting and refining a voxel unit to be refined according to some embodiments of the present disclosure is schematically shown.

[0028] Figure 6 A schematic diagram schematically illustrates the process of segmenting and refining an initial voxel map to obtain a fine voxel map according to some embodiments of the present disclosure.

[0029] Figure 7 A schematic diagram of a process for determining a second candidate waypoint according to some embodiments of the present disclosure is schematically shown.

[0030] Figure 8 The following schematically illustrates a flow chart of determining a target waypoint set according to some embodiments of the present disclosure.

[0031] In the drawings, the same or corresponding reference numerals denote the same or corresponding parts. DETAILED DESCRIPTION

[0032] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with this specification. Rather, they are merely examples of apparatus and methods consistent with certain aspects of this specification, as detailed in the appended claims.

[0033] It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information without departing from the scope of this specification. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to determining."

[0034] Furthermore, the drawings are schematic illustrations only and are not necessarily drawn to scale. The block diagrams shown in the drawings are merely functional entities and do not necessarily correspond to physically separate entities. In other words, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0035] In this example embodiment, a drone route planning method for hydropower projects is first provided. This drone route planning method for hydropower projects can be applied to terminal devices or servers. This example embodiment does not make any special limitations on this, and the following explanation will be given using the server executing this method as an example. Figure 1 The following schematically illustrates a flow chart of a method for planning a route for a hydropower project using a drone according to some embodiments of the present disclosure. Figure 1 As shown, the UAV route planning method for hydropower engineering may include the following steps: Step S110, determining multiple target operation areas to be inspected and monitored, and generating a three-dimensional terrain grid model corresponding to each target operation area; Step S120, voxelizing the three-dimensional terrain grid model to construct an initial voxel map, and determining a first candidate waypoint in combination with the initial voxel map; Step S130 , segmenting and refining each voxel unit in the initial voxel map based on the average normal vector distribution of the initial voxel map to generate a refined voxel map, and determining a second candidate waypoint in combination with the refined voxel map; Step S140, performing collision detection on the first candidate waypoint and the second candidate waypoint in combination with the fine voxel map to obtain a target waypoint set; Step S150 , generating a UAV operation route according to the target waypoint set, so as to control the UAV to inspect the corresponding target operation areas through the UAV operation route.

[0036] According to the UAV route planning method for hydropower engineering in this example embodiment, on the one hand, by performing spatial modeling based on a three-dimensional terrain grid model and combining voxelization processing to establish a structured voxel unit system, while ensuring the accuracy of spatial terrain feature expression, the spatial discretization and index organization of the target operation area are achieved, which facilitates the rapid execution of subsequent waypoint extraction, visibility judgment and spatial analysis; further, through the voxel segmentation mechanism driven by the average normal vector, the voxel granularity can be adaptively adjusted to adapt to landforms or structural boundaries of different complexities, thereby taking into account both structural fidelity and computing resource control when constructing a waypoint set, reducing the traditional reliance on precise global modeling methods. On the other hand, by introducing visibility analysis and spatial collision detection mechanism, dynamic compensation and validity verification of waypoint selection strategy are realized. The first candidate waypoint and the second candidate waypoint are complementary based on the logical relationship between observable voxels and unobservable voxels. On the basis of achieving the integrity of terrain information acquisition, redundant waypoints with flight safety risks are eliminated through spatial interaction with the fine voxel map, thereby improving the effective density and executability of the overall waypoint set, improving the actual airworthiness and operational stability of the final route generation, and avoiding common problems in related technologies caused by misjudgment of occlusion, flight path crossing obstacles or insufficient coverage of the monitoring area.

[0037] The following further describes the drone route planning method for hydropower projects in this example embodiment.

[0038] In step S110 , a plurality of target operating areas to be inspected and monitored are determined, and a three-dimensional terrain grid model corresponding to each of the target operating areas is generated.

[0039] In an exemplary embodiment of the present disclosure, the target operating area refers to inspection and monitoring objects such as operating facilities, slope areas, flood discharge channels, and hydraulic structure surfaces that require special attention in water conservancy and hydropower projects. For example, the target operating area can be obtained by dividing the monitored targets or areas of the hydropower project based on the performance of the drone equipment used. The target operating area can also be determined through engineering drawings, remote sensing images, Geographic Information System (GIS) platforms, or manual selection. This exemplary embodiment does not specifically limit the method for determining the target operating area. In addition, relevant inspection and monitoring parameters such as ground resolution, altitude, image overlap, and other data can also be specified.

[0040] A 3D terrain mesh model is a collection of model data representing the surface geometry in the form of polygonal patches (usually triangular patches), obtained through spatial data collection and 3D reconstruction of the target operating area. A 3D terrain mesh model can be used to represent spatial entities such as topographic relief, hydraulic structure outlines, and slope sections found in typical hydropower project environments, with high structural fidelity and surface continuity.

[0041] Aerial photogrammetry can be used to generate a 3D terrain mesh model. For example, this process involves acquiring multi-view images from an unmanned aerial vehicle (UAV) equipped with an imaging sensor. The camera pose is then restored using algorithms based on feature point matching and sparse reconstruction (e.g., Structure from Motion). This is then combined with data such as multi-view stereo (MVS) or lidar point clouds to form a dense point set. This dense point set is then subjected to surface reconstruction, such as through Poisson reconstruction, Delaunay triangulation, or Alpha Shape methods, to generate a structurally coherent and topologically complete 3D terrain mesh model. The resulting 3D terrain mesh model must have sufficient facet density and terrain restoration to support subsequent voxel partitioning and visibility analysis. Furthermore, the mesh structure must maintain topological integrity, avoiding issues such as non-manifold boundaries or dangling faces that affect spatial indexing quality.

[0042] In step S120, the three-dimensional terrain grid model is voxelized to construct an initial voxel map, and a first candidate waypoint is determined in combination with the initial voxel map.

[0043] In an example embodiment of the present disclosure, voxelization refers to the process of performing discrete division operations on the continuous surface space described by a three-dimensional terrain grid model, with the resulting division units being voxel units. Voxel, short for three-dimensional pixel, represents a spatial unit of fixed size and regular shape (usually a cube) in three-dimensional space. The essence of voxelization is to project the target space onto a voxel grid, forming a regular voxel structure, thereby achieving a structural mapping between the terrain model and the spatial coordinate system, facilitating subsequent spatial calculations, direction determination, and visibility analysis.

[0044] The initial voxel map is the three-dimensional structured spatial representation data formed by establishing a spatial index structure and mapping relationships after voxelization. The initial voxel map is based on a set of position indices in a three-dimensional coordinate system. Each index corresponds to a voxel unit. Each voxel unit records the set of triangles it contains, as well as attribute information such as the normal information, center location, and area statistics of the facets.

[0045] The initial voxel map can support fast voxel-oriented statistical calculations. For example, it can be used to calculate the average normal vector or patch density. It can also serve as the basis for waypoint extraction and occlusion analysis at a coarse-grained level. It can also provide the initial data structure for subsequent fine-grained voxel map construction. In terms of spatial expression, the initial voxel map can be organized using a sparse voxel octree (SVO) or a sparse index hash table to save storage resources and improve access efficiency. This embodiment does not specifically limit the organization of the initial voxel map.

[0046] The first candidate waypoint refers to an initial level waypoint set generated based on the initial voxel map and having three-dimensional spatial position and orientation attributes, and is mainly used to cover the surface or overhead area of the easily observable area. For example, after the initial voxel map is established, according to the candidate waypoint layout logic, the patch-covered voxels at the top of the voxel column can be preferentially extracted as the area with a good overhead angle, and at the center position of each such voxel, the voxels are offset along the negative observation direction according to the set movement distance to generate the first candidate waypoint with the observation attitude attribute. Of course, the first candidate waypoint can also be determined by other methods, and this embodiment does not specifically limit this.

[0047] In step S130 , based on the average normal vector distribution of the initial voxel map, each voxel unit in the initial voxel map is segmented and refined to generate a refined voxel map, and the second candidate waypoint is determined in combination with the refined voxel map.

[0048] In an example embodiment of the present disclosure, the average normal vector refers to a direction vector obtained by weighted averaging the unit normal vectors of multiple triangular facets contained in each voxel unit. The average normal vector can reflect the overall geometric orientation of the terrain or building surface covered in the voxel unit.

[0049] Segmentation refinement is the process of rescaling voxels with complex geometry and drastically varying surface orientations based on the initial voxel map to improve voxel resolution and geometric adaptability. For example, segmentation refinement can be triggered based on a mean normal deviation criterion. If a voxel contains a significant number of triangles with an angle exceeding a preset threshold (e.g., 30°-45°) from the mean normal, and the proportion of these triangles exceeds a threshold (e.g., 10%-20%), the voxel is considered to have inconsistent normal distribution and requires segmentation refinement. Refinement can be performed by splitting the voxel in half in the X, Y, and Z directions, generating eight new voxel cells of equal volume. Each new voxel inherits a portion of the original voxel's facets, redistributed based on the location of the facet's centroid. This refinement process can be recursively executed until the normal distribution of each voxel cell meets the consistency criteria, or the voxel size falls below a set minimum voxel size (e.g., 1 / 8 of the initial voxel size). The purpose of segmentation refinement is to improve voxel resolution in areas of high geometric complexity and provide a more refined data basis for subsequent visibility analysis.

[0050] A refined voxel map refers to an enhanced version of the voxel map structure constructed based on the segmentation and refinement of the initial voxel map, and is used to express the spatial patch distribution, normal changes, and occlusion relationships of complex areas at a higher resolution. The refined voxel map can retain the basic index structure of the initial voxel map, but the voxel size is reduced due to the segmentation operation, and the number of patches included is also relatively reduced, so that the normal distribution of each voxel unit is more consistent and the geometric expression is closer to the actual surface morphology. The refined voxel map can be used as input for second candidate waypoint generation, collision detection, and path feasibility analysis. This embodiment does not specifically limit the use of the refined voxel map.

[0051] The second candidate waypoint refers to a set of supplementary waypoints added after the first candidate waypoint is obtained to compensate for observation blind spots caused by occlusion, irregular structures, or normal directionality. Its generation logic relies on the fine voxel map and visibility analysis mechanism. For example, with the observation direction of each first candidate waypoint as a reference, visibility judgment is performed on all voxels in the fine voxel map. Whether the voxel is in an observable state is determined based on whether the angle between the average normal vector and the observation direction is less than a set threshold. If there are unobservable voxel units, for each invisible voxel, a waypoint offset is generated based on its average normal vector direction. The voxel center is often used as the reference, and a set distance is offset along the average normal vector direction as the waypoint position. The offset direction is used as the waypoint observation direction to form the second candidate waypoint. The second candidate waypoint forms an observation complementary relationship with the first candidate waypoint to ensure full coverage and information integrity in complex terrain areas. In actual deployment, the valid point set is screened through collision detection and incorporated into the final route generation process.

[0052] It should be noted that the "first" and "second" in "first candidate waypoint" and "second candidate waypoint" are only used to distinguish candidate waypoints determined at different stages and do not have any special meaning and should not cause any special limitation to this exemplary embodiment.

[0053] In step S140 , collision detection is performed on the first candidate waypoint and the second candidate waypoint in combination with the fine voxel map to obtain a target waypoint set.

[0054] In an exemplary embodiment of the present disclosure, collision detection refers to the process of analyzing the spatial relationship between candidate waypoints and voxel cells representing obstacles or terrain surfaces in a refined voxel map to determine whether the waypoint poses a risk of contacting or crossing an obstacle during actual flight. This process is primarily used to screen out waypoints with potential flight conflict risks, thereby improving the safety and feasibility of the generated path. For example, collision detection can be implemented by expressing waypoints using oriented bounding boxes (OBBs), or by constructing constraints and solving equations. This embodiment does not specifically limit the specific implementation method of collision detection.

[0055] The target waypoint set refers to the set of candidate waypoints with actual flight feasibility that remain after collision detection. It is the direct input point set for subsequent UAV operation route generation. The target waypoint set is obtained by screening and merging the first and second candidate waypoints. It has two basic elements: complete spatial position (XYZ coordinates) and observation direction (unit vector), and each waypoint has passed spatial safety verification. The spatial distribution of the target waypoint set fully covers the target operation area, avoiding data collection blind spots, and the attitude information is clear, which is conducive to maintaining image overlap and shooting angle consistency during mission execution. The waypoints in the target waypoint set have flight safety guarantees and can be directly used as path planning nodes for continuous connection, reducing flight control failures or mission interruptions caused by path crossing or altitude jumps.

[0056] In step S150, a UAV operation route is generated according to the target waypoint set, so as to control the UAV to inspect the corresponding target operation areas through the UAV operation route.

[0057] In one exemplary embodiment of the present disclosure, a drone operational route refers to a sequence of three-dimensional paths generated from a set of target waypoints, characterized by spatial continuity, directional constraints, and flight feasibility, serving as the flight trajectory that the drone must follow during actual mission execution. A drone operational route not only includes the sequence of waypoint connections but also includes additional attributes such as flight direction, steering nodes, flight speed settings, and mission trigger points. It can be output in a standard format recognizable by drone mission planning systems, such as KML (Keyhole Markup Language) or KMZ files, or in a path encoding format adapted for a specific flight control system.

[0058] During the route generation process, the target waypoint set can be sorted and connected by paths, for example, through shortest path traversal, approximate Traveling Salesman Problem (TSP) optimization, and minimum curvature path fitting. To improve the stability and efficiency of path execution, an "S-shaped" path structure can also be generated. That is, using the minimum enclosing rectangle of the spatial distribution of the target waypoints as a reference, approximately parallel segments are generated along its long side, and connected by return trips to form a continuous track. This embodiment does not specifically limit the method for generating the drone operation route in combination with the target waypoint set.

[0059] Next, steps S110 to S150 will be described in detail.

[0060] In an exemplary embodiment of the present disclosure, Figure 2 The steps in the voxelization process are used to voxelize the 3D terrain grid model and construct the initial voxel map. Figure 2 Specifically, it may include: Step S210, determining the initial voxel size of the basic voxel unit; Step S220, determining a minimum bounding box corresponding to the three-dimensional terrain grid model, and evenly dividing the minimum bounding box according to the initial voxel size to obtain a basic voxel map; Step S230 : constructing a mapping relationship between the three-dimensional terrain grid model and each of the basic voxel units in the basic voxel map, and mapping the three-dimensional terrain grid model to the basic voxel map based on the mapping relationship to obtain an initial voxel map.

[0061] The initial voxel size refers to the side length of the cube that makes up the discrete grid cells in three-dimensional space. The size of the initial voxel size determines the resolution and spatial coverage of the voxel grid. In drone inspection and monitoring applications, the voxel size can be determined based on the image sensor parameters used and the mapping requirements of the mission. For example, the initial voxel size can be calculated based on the drone's altitude, mapping scale, pixel resolution (ground sampling distance), and the number of horizontal and vertical pixels in the sensor.

[0062] After the initial voxel size is determined, the minimum bounding box corresponding to the three-dimensional terrain grid model can be constructed, and the bounding box can be evenly divided based on the initial voxel size to obtain a basic voxel map. The minimum bounding box of the three-dimensional terrain grid model refers to the minimum rectangular volume area required to completely enclose all vertex coordinates of the model in the world coordinate system. Its three-dimensional range can be determined by scanning all vertex coordinates in the model and obtaining the minimum and maximum values in the X, Y, and Z directions respectively. The division process can be based on the orthogonal coordinate axis direction, and the minimum bounding box is divided into equal intervals along the X, Y, and Z dimensions according to the initial voxel size, and finally a three-dimensional voxel grid structure composed of several basic voxel units of the initial voxel size is formed. Each voxel unit has a unique three-dimensional index number in space, which can be represented by its sequence number triplet in each direction, which is used for the subsequent voxel attribution and index management of the facet.

[0063] A mapping relationship is a spatial correspondence established between the facets of a 3D terrain mesh model and the basic voxel units in the underlying voxel map. This relationship can be used to determine the voxel unit to which each triangular facet belongs and supports attribute statistics and geometric analysis of the internal spatial structure of each voxel unit. A mapping relationship can be established between the 3D terrain mesh model and the basic voxel units in the underlying voxel map. For example, each triangular facet in the 3D terrain mesh model can be sequentially traversed, the coordinates of its three vertices can be extracted, and the coordinates of its centroid can be calculated. The minimum coordinate value of all vertices in the 3D terrain mesh model in space (in three directions) is also obtained as the reference point for the voxel index origin. Next, based on the offset of the facet centroid's coordinates relative to the origin in space and the initial voxel size, the corresponding index position in the X, Y, and Z directions is calculated. By dividing the coordinate offset in each direction by the voxel side length and rounding the offset, the voxel unit number to which the facet belongs is obtained. This index information is then mapped to the facet index to complete the mapping of the facet to the voxel unit.

[0064] In an optional implementation, the face ownership information can be implemented using a hash table or a sparse three-dimensional array, where the key is a voxel index triple and the value is a pointer to a face index set. The mapping relationship formed by the face ownership information constitutes the data body of the initial voxel map, supporting subsequent normal vector statistics, area accumulation, boundary state judgment and visibility analysis operations for the facets contained in each voxel. Since the model facets may span multiple voxel boundaries, an optional implementation method can adopt a strategy of sharing ownership between the facet and multiple voxels, and give priority to the main voxel when the facet center of gravity offset is not large, thereby reducing the redundant storage burden.

[0065] By using the initial voxel map as a conversion carrier from a continuous geometric model to a discrete spatial structure, not only can the unit division and index establishment of the spatial structure be completed, but also the organizational compression, structural reconstruction and data reuse of model information can be achieved, providing a unified spatial analysis basis for subsequent multiple stages. The establishment of the initial voxel map not only ensures the local consistency of the voxel unit in geometric expression, but also improves the structural response capability for waypoint generation.

[0066] Optionally, the initial voxel size of the basic voxel unit may be determined by the following steps, which may specifically include: The flight parameters corresponding to the UAV can be obtained, which may include scale parameters, pixel size, horizontal pixel number of the sensor, vertical pixel number of the sensor, heading overlap and lateral overlap; the initial voxel size of the basic voxel unit is determined according to the flight parameters.

[0067] Among them, flight parameters refer to the pre-set flight mission configuration parameters required to achieve specified mapping accuracy in aerial photogrammetry or remote sensing data collection tasks. For example, flight parameters may include but are not limited to scale parameters, pixel size, horizontal pixel number of the sensor, vertical pixel number of the sensor, heading overlap and lateral overlap.

[0068] The scale parameter reflects the proportional relationship between actual ground distance and image projection distance and is a key metric for controlling spatial restoration accuracy during flight missions. For example, a 1:1000 image scale means that 1 mm on the image corresponds to 1000 mm on the ground. Pixel size refers to the physical size of a single sensor pixel in the image plane, typically measured in microns, such as 3.6 μm. The horizontal and vertical pixel counts of the sensor together determine the image coverage of a single image in the camera coordinate system—the width and height of the information recorded in the image in the X and Y directions. Heading overlap and lateral overlap refer to the overlap ratios between two adjacent images in the flight direction and lateral direction, respectively. These overlaps are typically expressed as percentages and are used to ensure image stitching continuity and stereo redundancy. For example, a common configuration is 75% heading overlap and 65% lateral overlap. These two values ensure that each region is reproducibly observed across multiple images, enhancing 3D reconstruction stability.

[0069] For example, the initial voxel size can be determined by the following relationship: ; in, can represent the initial voxel size, Can represent scale parameters, It can represent the pixel size, It can represent the number of horizontal pixels of the sensor, It can represent the number of vertical pixels of the sensor, It can indicate the degree of heading overlap, It can represent the degree of lateral overlap.

[0070] In an optional implementation, in order to ensure that the spatial structure after voxel division can accommodate the details of the terrain surface without being over-subdivided to cause storage expansion or a sharp increase in computing pressure, a minimum limit scale or maximum voxel capacity control strategy can be introduced in the actual setting. For example, based on the terrain slope, model density or image feature point distribution density, an adaptive voxel size setting mechanism is adopted for different areas, using smaller voxel sizes in complex areas and larger voxel sizes in flat areas, forming a hierarchical and controllable voxel size strategy.

[0071] By quantitatively converting flight parameters into three-dimensional space division scales, the transition from mapping indicators to model structure construction indicators is achieved, providing a theoretical basis and operational support for the subsequent execution of a voxel modeling process with reasonable accuracy and controllable calculations, and effectively improving the consistency and traceability between voxel units and image observation parameters.

[0072] Optionally, the mapping relationship between the three-dimensional terrain grid model and each basic voxel unit in the basic voxel map can be constructed by the following steps, which may specifically include: The center coordinates of the facets corresponding to each model facet in the three-dimensional terrain grid model can be obtained, as well as the minimum vertex coordinates corresponding to the three-dimensional terrain grid model; the index information of each model facet in the basic voxel map is determined according to the facet center coordinates and the minimum vertex coordinates; the index information is used as the mapping relationship between the three-dimensional terrain grid model and each basic voxel unit in the basic voxel map.

[0073] Among them, the three-dimensional terrain grid model is usually composed of several model patches, which are generally triangular patches. Each triangular patch consists of three vertex coordinates. In geometric processing, the patch center coordinates of the triangular patch can be represented by its center of gravity position. The calculation method is to find the average value of the X, Y, and Z coordinates of the three vertices of the triangular patch. The calculation method is not described here. The patch center coordinates can be used to represent the overall position of the patch in three-dimensional space, and can also be used as the main reference point for the patch to participate in voxel division and spatial positioning. In practical applications, the patch center coordinates can be uniformly represented in the world coordinate system to ensure its consistency with the basic voxel map in spatial scale and reference origin.

[0074] The minimum vertex coordinates corresponding to the 3D terrain grid model can be obtained and used as the origin reference value in the voxel index calculation process. When establishing a basic voxel map, the minimum bounding box of the entire terrain model is used as the voxel grid boundary, where the starting coordinates of the bounding box (i.e., the minimum values in the X, Y, and Z directions) are used as the origin of the voxel coordinate system. The minimum vertex coordinates can be obtained by traversing and calculating all vertex coordinates. The minimum vertex coordinates determine the reference origin of the voxel grid in 3D space, which is used to establish the offset relationship from model space to voxel space and is the basis for achieving mapping alignment.

[0075] For example, the following relationship can be used to determine the index information of each model face in the basic voxel map based on the minimum value of the face center coordinates and vertex coordinates: ; in,( , , ) can represent the center coordinates of each facet in the three-dimensional terrain grid model. , , ) can represent the index information of each model patch in the basic voxel map, ( , , ) can represent the minimum vertex coordinates corresponding to the three-dimensional terrain grid model, It can represent the initial voxel size.

[0076] Index information can be registered as a mapping between the 3D terrain mesh model and each basic voxel unit in the base voxel map. This means that a facet set is created for each voxel unit to represent the model surface structure covered by that voxel. To improve storage and access efficiency, a sparse hash structure or nested dictionary structure can be used to organize data, where the key is an index triple and the value is a facet index list or a facet attribute structure pointer. To avoid missing faces when they cross voxel boundaries, a mechanism can be set up to co-own multiple voxels, or a master voxel strategy can be used to only own the voxel containing the center of gravity.

[0077] This mapping relationship not only establishes a spatial link between the model patch and the voxel unit, but also aggregates the geometric information contained in each voxel unit, including normal vectors, area statistics, boundary status, etc., providing the necessary data support for subsequent steps such as normal vector mean calculation, voxel refinement judgment, visibility analysis, and waypoint offset direction determination. In this way, a precise, unified, and computable geometric mapping system is constructed between the 3D terrain grid model and the voxel structure, improving the efficiency and accuracy of waypoint calculation.

[0078] Optionally, the following steps may be performed to determine the first candidate waypoint in combination with the initial voxel map, which may specifically include: The scale parameters corresponding to the three-dimensional terrain grid model and the lens focal length corresponding to the drone can be obtained, and the moving distance can be determined according to the scale parameters and the lens focal length; the target voxel unit at the top position in the initial voxel map is marked, and the voxel center coordinates of the target voxel unit are moved in the target direction according to the moving distance to obtain the first candidate waypoint, where the target direction is the opposite direction of the observation direction at the first candidate waypoint.

[0079] Among them, the focal length of the drone lens refers to the distance from the camera's main optical axis to the imaging plane, generally measured in millimeters. It is a determining factor affecting the field of view angle and ground coverage scale. Under orthophotography conditions, the flight altitude or image ground projection length can be derived based on basic photogrammetry relationships, scale parameters, and lens focal length. The appropriate spatial scale for waypoint offset, i.e., the travel distance, can then be calculated back. For example, the travel distance can be determined based on scale parameters and lens focal length using the following relationship: ; in, It can represent the focal length of the camera lens on the drone. Can represent scale parameters, It can be understood that this relationship assumes that the main visual axis is perpendicular to the ground during imaging, which is applicable to most structural direct observation tasks and can also be approximately applicable in areas with mild terrain height differences.

[0080] The top voxel unit is the highest voxel in the same vertical projection column, located near the positive Z axis, above which no voxels containing valid model patches exist. This can be achieved by sorting the Z coordinates of all voxel units with model patch mappings at a fixed cross-sectional position, and taking the one with the largest index value as the top voxel. Top voxel units can correspond to spatial locations such as raised areas on the terrain surface, the top of a dam, or the roof of a building. These voxels offer natural observation advantages such as unobstructed viewing, a wide unobstructed viewing angle, and high acquisition efficiency, making them suitable as priority locations for the first round of waypoint deployment.

[0081] After determining the top voxel unit, the center coordinates of the top voxel unit can be extracted as the position reference point of the candidate waypoint, and the position can be offset along the target direction based on the previously calculated moving distance. The voxel center coordinates can be calculated by multiplying the index position of the voxel in the three-dimensional grid by the voxel size, plus an offset of half the voxel length to ensure that the waypoint is positioned on the central axis of the voxel space. The offset direction is the opposite direction of the observation direction at the waypoint. For example, the offset direction can be the negative direction of the Z axis in a vertical observation task, or it can be set to the opposite direction of the voxel average normal vector according to the specific task. The specific setting can be customized according to the actual application situation, and this embodiment does not make special restrictions on this.

[0082] refer to Figure 3 As shown, the target voxel unit 310 at the top position in the initial voxel map can be marked, and the voxel center coordinates of the target voxel unit 310 are moved toward the target direction according to the moving distance to obtain the first candidate waypoint 320.

[0083] By extracting the top voxel unit based on the initial voxel map and combining the scale parameters and lens focal length to generate the first candidate waypoint with directional attributes, the rationality and visibility of waypoint selection can be significantly improved while ensuring the efficiency of mission area coverage. By selecting the top space area as the priority deployment location, terrain obstruction and building interference can be effectively avoided, and the degree of unobstructed image acquisition and the integrity of the field of view can be improved; combining the moving distance calculated by inverse calculation of the lens focal length and scale to offset the waypoint helps to ensure that the distance between the waypoint and the observation surface is controlled within the optimal range for mapping, thereby improving the clarity and geometric restoration accuracy of the captured image; in addition, setting the waypoint orientation to the observation direction relative to the surface normal direction can enhance the consistency and directional integrity of data acquisition, providing structural support for the continuity of subsequent routes and attitude planning.

[0084] In an exemplary embodiment of the present disclosure, Figure 4 The steps in the above code are to segment and refine each voxel unit in the initial voxel map based on the average normal vector distribution of the initial voxel map, and generate a fine voxel map. Figure 4 Specifically, it may include: Step S410, calculating an average normal vector corresponding to each basic voxel unit based on the normal vector of the model face included in each basic voxel unit in the initial voxel map; Step S420 , screening voxel units to be refined in the initial voxel map according to the average normal vector; Step S430 , determining a segmentation voxel size of the voxel unit to be refined based on the initial voxel size of the basic voxel unit, wherein the segmentation voxel size is half of the initial voxel size, and segmenting the voxel unit to be refined based on the segmentation voxel size to obtain a new basic voxel unit; Step S440 , repeatedly executing steps S410 to S430 until all basic voxel units in the initial voxel map meet the segmentation restriction conditions, thereby obtaining a refined voxel map.

[0085] The basic voxel unit is a spatially discrete cubic unit, which may contain multiple triangular facets. Each facet has a spatial normal vector that represents its surface orientation. In order to express the main orientation characteristics of the entire voxel unit, the average normal vector can be solved by vector weighted averaging. For example, the average normal vector corresponding to each basic voxel unit can be determined by the following relationship: ; in, It can represent the average normal vector corresponding to the basic voxel unit, It can represent the surface of the k-th model patch in the basic voxel unit, It can represent the normal vector corresponding to the face of the k-th model patch in the basic voxel unit, It can represent the total number of model faces within the basic voxel unit.

[0086] After obtaining the average normal vector, the voxel units to be refined can be screened in the initial voxel map based on the average normal vector. The screening criteria can be based on the normal consistency principle. That is, if there is a significant deviation between the normal vector of the facet within the voxel and the average normal vector, the internal surface of the voxel is considered to have changed dramatically and is not suitable for representation at the current voxel granularity. In the specific implementation process, for example, the angle between the normal vector of each facet within the voxel and the average normal vector can be calculated. If the angle is greater than a set threshold (such as 30°), it is included in the deviation facet statistics. If the proportion of the number of deviation facets to the total number of facets in the voxel exceeds a certain proportion threshold (such as 10% to 20%), the voxel unit is considered to require segmentation and refinement and is determined to be a voxel unit to be refined. Of course, when screening voxel units to be refined, auxiliary geometric indicators such as the facet normal standard deviation, point density change rate, or curvature distribution can also be evaluated to form a multi-factor refinement criterion to further improve the accuracy of screening voxel units to be refined.

[0087] After the voxel units to be refined are selected, the size of their split voxel units can be determined by the initial voxel size of the basic voxel unit. The split voxel size can be half of the initial voxel size, that is, the side length of the original voxel is halved along the X, Y, and Z directions, thereby forming 8 new sub-voxel units, each with a volume of 1 / 8 of the original voxel, as the new basic voxel unit. Figure 5 As shown, the basic voxel unit 510 at the initial voxel size can be segmented and refined by splitting the voxel size to obtain 8 new basic voxel units 520. This voxel segmentation process can synchronously update the spatial index structure and recalculate the set of patches covered by each new voxel unit. For example, this can be achieved by traversing all patches in the original voxel, recalculating the centroid coordinates of each patch to its belonging position in the new index framework, and constructing a new patch mapping relationship.

[0088] After the segmentation operation of the initially screened voxel units to be refined is completed, the above steps are repeated, that is, the average normal vector calculation, deviation patch screening and segmentation judgment are performed again on the newly generated new basic voxel units to form a recursive hierarchical refinement mechanism. This process can continue until all basic voxel units in the initial voxel map meet the segmentation restriction conditions and a fine voxel map is obtained. For example, the segmentation restriction condition can be that the segmentation voxel size is less than one-eighth of the initial voxel size. Of course, it can be understood that in some scenarios, the segmentation restriction condition can also be that the segmentation voxel size is less than one-sixteenth of the initial voxel size. This embodiment does not specifically limit the specific content of the segmentation restriction condition. Reference Figure 6 As shown, voxel units to be refined can be screened based on the average normal vector distribution of the initial voxel map 610 , and each voxel unit to be refined in the initial voxel map 610 can be segmented and refined to generate a refined voxel map 620 .

[0089] The final fine voxel map is composed of multi-level voxel units, which has higher geometric adaptability and orientation expression capabilities at the structural level. It can accurately capture the spatial characteristics of areas such as terrain mutation areas, building corner structures or steep slope surfaces, and effectively support the local offset calculation, occlusion area identification and flight visibility judgment of subsequent candidate waypoint layout.

[0090] Optionally, the following steps may be performed to select voxel units to be refined in the initial voxel map according to the average normal vector, which may specifically include: The first angle between the normal vector of each model facet in the basic voxel unit and the average normal vector can be calculated; the number of faces in the basic voxel unit whose first angle is greater than or equal to a preset angle threshold is determined; if the ratio of the number of faces to the total number of faces in the basic voxel unit exceeds the preset ratio threshold, the basic voxel unit is determined to be a voxel unit to be refined.

[0091] Among them, the basic voxel unit is the minimum analysis unit in the initial voxel map, which contains several model patches. Each patch corresponds to a unit normal vector, which represents the orientation direction of the triangle in three-dimensional space. The average normal vector is the comprehensive orientation result obtained by normalizing the unit normal vectors of all patches in the voxel according to the patch area or other geometric weights. In order to evaluate the spatial difference between the normal vector of a single patch and the average normal vector of the voxel, the first angle between the normal vector of each model patch in the basic voxel unit and the average normal vector can be calculated by the following relationship: ; in, Can represent the average normal vector of the basic voxel unit and the kth face normal vector in the base voxel unit The first angle between them.

[0092] After completing the calculation of the first angle, the number of facets in the basic voxel unit whose first angle is greater than or equal to the preset angle threshold can be determined. For example, the normal angle deviation results of all facets in the current voxel can be traversed, and the number of facets whose angle is greater than or equal to the preset angle threshold can be counted, and the proportional relationship between the number of facets and the total number of facets in the voxel can be recorded. This statistical operation can be completed online using an accumulator, or all angles can be pre-calculated and then uniformly counted. This example embodiment does not impose any special restrictions on this. To enhance the stability of the evaluation, when the total number of facets is small (such as less than 3 facets), it can be set to not perform a detailed judgment to avoid misjudgment due to fluctuations in small samples.

[0093] After the proportion of patches that deviate from the threshold is counted, a determination is made as to whether the refinement conditions are met. For example, if the proportion of the number of patches in a basic voxel unit to the total number of patches in the basic voxel unit exceeds a preset ratio threshold, such as 10%, 15%, or 20%, then the basic voxel unit can be marked as a voxel unit to be refined. The selection of the preset ratio threshold can comprehensively consider the sensitivity of the algorithm and the tolerance of patch noise, and can be in the range of 10% to 30%. Of course, parameter regression optimization can also be performed based on an empirical data set to obtain a more adaptable threshold configuration. This example embodiment does not specifically limit the setting of the preset ratio threshold.

[0094] By screening the refined voxel units, not only can the locations with corners, mutations or drastic curvature changes in the model be accurately identified spatially, but the resolution of the voxel map can also be adaptively controlled structurally, providing basic support for enhancing local geometric expression capabilities for the subsequent hierarchical construction of fine voxel maps, waypoint position optimization and direction control strategies, thereby improving the calculation efficiency and accuracy of waypoint positions.

[0095] In an exemplary embodiment of the present disclosure, Figure 7 The steps in the above are used to determine the second candidate waypoint by combining the refined voxel map. Figure 7 Specifically, it may include: Step S710, calculating the average normal vector of each fine voxel unit of the fine voxel map, and determining a second angle between the average normal vector and the direction opposite to the observation direction in combination with the observation direction of the first candidate waypoint; Step S720: determining the observability of each of the fine voxel units at each of the first candidate waypoints according to the second angle; Step S730, recording all observable fine voxel units into a visible voxel set through the observability, and recording all unobservable fine voxel units into an invisible voxel set; Step S740: If the invisible voxel set corresponding to any first candidate waypoint is not an empty set, the voxel center of the fine voxel unit is offset by a moving distance along the average normal vector direction of the unobservable fine voxel unit to generate a new waypoint as a second candidate waypoint for supplementing the first candidate waypoint.

[0096] Among them, the fine voxel unit refers to a voxel unit with higher spatial resolution obtained on the basis of the refinement of the initial voxel unit, and each voxel unit contains multiple triangular facets. The average normal vector is the weighted vector and normalized result of the unit normal vectors of all facets in the voxel, reflecting the main orientation of the terrain surface represented by the voxel. After obtaining the observation direction of the first candidate waypoint, its reverse direction vector can be taken to represent its projection direction from the waypoint to the ground, thereby determining the second angle between the average normal vector and the direction opposite to the observation direction. For example, the second angle can be calculated using the following relationship: ; in, It can represent the second angle between the average normal vector of the fine voxel unit and the opposite direction of the observation direction of the first candidate waypoint, It can indicate the viewing direction of the first candidate waypoint, It can represent the average normal vector of the fine voxel unit.

[0097] Observability refers to whether the voxel surface can be directly observed by the current waypoint under a given posture. If the second angle between the average normal vector of the fine voxel unit and the observation direction of the first candidate waypoint is less than or equal to the preset angle threshold (for example, the preset angle threshold is 45°), and there is no other voxel blocking the direction, then it can be considered that the fine voxel unit can be observed at the current first candidate waypoint; otherwise, it is unobservable. In an alternative approach, a visual field analysis method based on ray casting can also be introduced to trace rays from the waypoint in the direction of the voxel. If the ray path does not pass through other voxels, it is further confirmed that the voxel is in an unobstructed and observable state. This embodiment is not limited to this.

[0098] After the observability judgment is completed, all observable fine voxel units can be recorded into the visible voxel set based on the observability result, and all unobservable fine voxel units can be recorded into the invisible voxel set. If the invisible voxel set corresponding to any first candidate waypoint is not an empty set, the voxel center of the fine voxel unit is offset by the moving distance along the average normal vector direction of the unobservable fine voxel unit to generate a new waypoint as the second candidate waypoint to supplement the first candidate waypoint. The moving distance can be calculated by referring to the calculation method for determining the moving distance of the first candidate waypoint, and will not be further explained here.

[0099] By determining the second candidate waypoint through a fine voxel map, special points can be added to the voxel areas that are not covered due to orientation mismatch, occlusion, or blind spots in the field of view of the first candidate waypoint. This ensures that the three-dimensional terrain model can obtain complete, continuous, and unobstructed image data during the actual mission execution, effectively avoiding the monitoring blind spot problem caused by sparse waypoints in traditional methods, improving the integrity of candidate waypoints and comprehensive coverage of the target operation area, thereby ensuring the accuracy and effectiveness of the generated UAV operation route.

[0100] In an exemplary embodiment of the present disclosure, collision detection can be performed on the first candidate waypoint and the second candidate waypoint in combination with the fine voxel map to obtain a target waypoint set through the following steps, which may specifically include: A directed bounding box is constructed with the first and second candidate waypoints as the center. If the directed bounding box corresponding to any candidate waypoint intersects or contains a fine voxel unit in the fine voxel map, the candidate waypoint is judged to have a collision risk and is removed from the first and second candidate waypoints to obtain the target waypoint set.

[0101] Among them, an oriented bounding box (OBB) refers to a compact bounding geometry whose boundary can be rotated according to a reference direction, compared to a traditional axis-aligned bounding box (AABB). An oriented bounding box has the advantages of small size and high boundary fit, and can more realistically express the enclosing range of a waypoint in space under a non-aligned direction. For example, the three-dimensional coordinates of the waypoint can be determined as the geometric center of the bounding box, and the main axis direction of the OBB can be constructed based on the observation direction vector of the waypoint, so that the Z axis of the bounding box is consistent with the direction of the waypoint. Then, the length, width, and height of the OBB can be set. The size can be set according to parameters such as the size of the drone body, the minimum safe distance for image acquisition, and the attitude error range. Usually, the side length range can be set between 0.5 meters and 2 meters.

[0102] After constructing the directed bounding boxes of all candidate waypoints, the spatial interaction between the directed bounding boxes and the fine voxel map can be determined to determine whether there is a collision risk for each waypoint. Specifically, the position and voxel size of each voxel unit in the fine voxel map are known, forming a regular three-dimensional grid structure. The collision detection process can be implemented through a spatial overlap test, that is, determining whether the OBB has an intersection or inclusion relationship with any voxel unit. For example, the Separating Axis Theorem (SAT) can be used to quickly determine the intersection of the OBB and the cube voxel, or the voxel octree structure can be used to perform a neighbor voxel search and intersection test. In addition, the initial screening can be quickly completed by projecting the OBB to each coordinate axis direction and comparing whether the projection intervals overlap. After confirming the voxel area that may collide, a precise intersection test is performed. This example embodiment does not specifically limit the method of implementing collision detection in combination with the directed bounding box.

[0103] If the directed bounding box corresponding to any candidate waypoint is determined to intersect or contain a voxel unit in the refined voxel map, the waypoint is considered to have a potential collision risk. An intersection relationship refers to the geometric overlap between the OBB boundary and the voxel boundary, and a containment relationship refers to the complete embedding of the OBB within a voxel unit. Both indicate that the drone may collide with terrain or structures when executing the waypoint mission, which may cause flight control failure, image acquisition failure, or even equipment damage. Therefore, when such a determination occurs, the candidate waypoint can be removed from the candidate set and will not participate in the subsequent target waypoint sorting and route connection.

[0104] After collision risk identification and elimination operations on all first candidate waypoints and second candidate waypoints, the remaining waypoint set constitutes the target waypoint set. All waypoints in this set meet the spatial safety requirements, have good flight passability and data collection stability, and can be directly used as input nodes for the subsequent generation of drone operation routes, effectively improving the safety of drone operation routes and reducing drone operation risks.

[0105] Figure 8 The following schematically illustrates a flow chart of determining a target waypoint set according to some embodiments of the present disclosure.

[0106] refer to Figure 8 As shown in the figure, the route planning method for UAVs in hydropower projects can include three stages: initial voxel map generation stage, voxel optimization segmentation stage, and waypoint output stage. Each stage forms an orderly process based on data flow and judgment logic. Specifically: First, the voxel size calculation step is performed using the target area basemap, i.e., the 3D terrain grid model, as input. This step calculates the initial voxel unit size based on the scale parameters, sensor parameters (including pixel size, horizontal pixel number, and vertical pixel number), as well as heading overlap and lateral overlap. The determined voxel size will be used to set the spatial scale of the subsequent voxelization processing.

[0107] After the voxel size is determined, the voxelization step is entered, and the three-dimensional grid model of the target area is spatially divided according to the calculated voxel size to generate a regular grid structure. Subsequently, the intra-voxel average normal vector calculation step is executed, and the normal vector weighted averaging processing is performed on the triangular facets contained in each basic voxel unit to obtain the average normal vector reflecting its overall surface orientation.

[0108] After completing the above operations, the initial voxel map is obtained and the first candidate waypoint generation process is entered. Here, the top voxel unit can be extracted and offset along the negative viewing direction based on the voxel center to obtain the first candidate waypoint set defined by the dual elements of position and direction.

[0109] At the same time, the initial voxel map undergoes structural optimization, a process known as voxel optimization segmentation. This phase first determines whether the current voxel meets the segmentation criteria: by comparing the angle between the facet normal vector within the voxel and the average normal vector, the number of facets with an angle greater than a preset threshold (e.g., 30°) is counted to see if it exceeds a set proportion (e.g., 15%). If so, the voxel segmentation process begins, splitting the voxel in half in each three-dimensional direction to generate a new voxel unit that is half the size of the original voxel. If not, the current voxel is output as the final resolution voxel. This segmentation judgment and operation process is iteratively executed until all voxel units meet the segmentation termination criteria, including a maximum recursive depth limit (e.g., a size less than 1 / 8 of the initial voxel).

[0110] All voxel units after segmentation and refinement together constitute a fine voxel map and enter the second candidate waypoint generation process. This process analyzes the visibility relationship between the fine voxels and the first candidate waypoints, identifies unobservable areas, and offsets these unobservable voxels along the direction of the average normal vector to generate second candidate waypoints, thereby completing the area not covered by the first candidate waypoints.

[0111] After the two candidate waypoint sets are generated, they enter the collision detection phase. This step constructs an directed bounding box (OBB) centered on the waypoint and performs intersection detection with the voxel units in the fine voxel map. If there is an intersection or containment relationship, it is determined that there is a collision risk and is eliminated. The remaining waypoints constitute the final target waypoint set.

[0112] Finally, the target waypoint set can be fitted with a minimum enclosing rectangle based on its spatial distribution pattern, generating an ordered "S"-shaped route. The route direction is kept parallel to the long side of the rectangle as much as possible to reduce turns and improve flight efficiency. This route is then exported in KML or KMZ format and used to guide drones in patrol and monitoring missions.

[0113] It should be noted that although the steps of the method disclosed herein are depicted in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in that particular order, or that all steps must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one, and / or one step may be decomposed into multiple steps.

Claims

1. A UAV route planning method for hydropower projects, characterized by: include: Determine multiple target operation areas to be inspected and monitored, and generate a three-dimensional terrain grid model corresponding to each target operation area; voxelize the three-dimensional terrain grid model to construct an initial voxel map, and determine a first candidate waypoint based on the initial voxel map; Based on the average normal vector distribution of the initial voxel map, segment and refine each voxel unit in the initial voxel map to generate a refined voxel map, and determine a second candidate waypoint in combination with the refined voxel map; performing collision detection on the first candidate waypoint and the second candidate waypoint in combination with the refined voxel map to obtain a target waypoint set; An unmanned aerial vehicle (UAV) operation route is generated according to the target waypoint set, so as to control the UAV to inspect and test the corresponding target operation areas through the UAV operation route.

2. The method for UAV route planning for hydropower projects according to claim 1 is characterized in that: The voxelization of the three-dimensional terrain grid model to construct an initial voxel map includes: Determine the initial voxel size of the basic voxel unit; Determining a minimum bounding box corresponding to the three-dimensional terrain grid model, and evenly dividing the minimum bounding box according to the initial voxel size to obtain a basic voxel map; A mapping relationship between the three-dimensional terrain grid model and each of the basic voxel units in the basic voxel map is constructed, and the three-dimensional terrain grid model is mapped to the basic voxel map based on the mapping relationship to obtain an initial voxel map.

3. The method for UAV route planning for hydropower projects according to claim 2 is characterized in that: The determining of the initial voxel size of the basic voxel unit includes: Obtain the flight parameters corresponding to the UAV, including scale parameters, pixel size, horizontal pixel number of the sensor, vertical pixel number of the sensor, heading overlap, and lateral overlap; An initial voxel size of the basic voxel unit is determined according to the flight parameters.

4. The method for UAV route planning for hydropower projects according to claim 2, characterized in that: The constructing of a mapping relationship between the three-dimensional terrain grid model and each of the basic voxel units in the basic voxel map includes: Obtaining the center coordinates of each facet in the three-dimensional terrain grid model, and obtaining the minimum value of the vertex coordinates corresponding to the three-dimensional terrain grid model; Determine the index information of each of the model facets in the basic voxel map according to the facet center coordinates and the minimum value of the vertex coordinates; The index information is used as a mapping relationship between the three-dimensional terrain grid model and each of the basic voxel units in the basic voxel map.

5. The method for UAV route planning for hydropower projects according to claim 1, characterized in that: Determining a first candidate waypoint in combination with the initial voxel map includes: Obtaining a scale parameter corresponding to the three-dimensional terrain grid model and a lens focal length corresponding to the drone, and determining a moving distance according to the scale parameter and the lens focal length; Mark a target voxel unit at the top position in the initial voxel map, and move the voxel center coordinates of the target voxel unit in a target direction according to the moving distance to obtain a first candidate waypoint, wherein the target direction is the opposite direction of the observation direction at the first candidate waypoint.

6. The method for UAV route planning for hydropower projects according to claim 1, characterized in that: The step of segmenting and refining each voxel unit in the initial voxel map based on the average normal vector distribution of the initial voxel map to generate a refined voxel map includes: Based on the normal vectors of the model patches contained in each basic voxel unit in the initial voxel map, calculating the average normal vector corresponding to each basic voxel unit; screening voxel units to be refined in the initial voxel map according to the average normal vector; Determining a segmentation voxel size of the to-be-refined voxel unit according to the initial voxel size of the basic voxel unit, wherein the segmentation voxel size is half of the initial voxel size, and segmenting the to-be-refined voxel unit according to the segmentation voxel size to obtain a new basic voxel unit; The above steps are repeated until all basic voxel units in the initial voxel map meet the segmentation restriction conditions, thereby obtaining a refined voxel map.

7. The method for UAV route planning for hydropower projects according to claim 6, characterized in that: The step of screening voxel units to be refined in the initial voxel map according to the average normal vector comprises: Calculating a first angle between the normal vector of each model facet in the basic voxel unit and the average normal vector; Determine the number of facets in the basic voxel unit whose first angle is greater than or equal to a preset angle threshold; If the ratio of the number of facets to the total number of facets in the basic voxel unit exceeds a preset ratio threshold, the basic voxel unit is determined to be a voxel unit to be refined.

8. The method for UAV route planning for hydropower projects according to claim 6, characterized in that: The segmentation restriction condition is that the segmentation voxel size is smaller than one eighth of the initial voxel size.

9. The method for UAV route planning for hydropower projects according to claim 1, characterized in that: Determining a second candidate waypoint in combination with the refined voxel map includes: Calculating an average normal vector of each fine voxel unit of the fine voxel map, and determining a second angle between the average normal vector and a direction opposite to the observation direction in combination with an observation direction of the first candidate waypoint; determining the observability of each of the fine voxel units at each of the first candidate waypoints according to the second angle; Recording all observable fine voxel units into a visible voxel set through the observability, and recording all unobservable fine voxel units into an invisible voxel set; If the invisible voxel set corresponding to any first candidate waypoint is not an empty set, the voxel center of the fine voxel unit is offset by a moving distance along the average normal vector direction of the unobservable fine voxel unit to generate a new waypoint as a second candidate waypoint for supplementing the first candidate waypoint.

10. The method for UAV route planning for hydropower projects according to claim 1, characterized in that: The performing collision detection on the first candidate waypoint and the second candidate waypoint in combination with the fine voxel map to obtain a target waypoint set includes: Constructing a directed bounding box with the first candidate waypoint and the second candidate waypoint as centers; If the directed bounding box corresponding to any candidate waypoint intersects or contains a fine voxel unit in the fine voxel map, the candidate waypoint is determined to have a collision risk, and the first candidate waypoint and the second candidate waypoint are eliminated to obtain a target waypoint set.

Images