Environment scanning method and device for three-dimensional reconstruction, electronic device and medium

Abstract

The present application relates to the field of three-dimensional reconstruction, and particularly relates to an environment scanning method and device for three-dimensional reconstruction, electronic equipment and medium, the method comprising: obtaining a current effective front determined according to a depth map and a reason for insufficient observation thereof; clustering the current effective front to obtain a current front cluster satisfying a normal vector consistency condition and a spatial distance constraint; generating a target scanning waypoint corresponding to the current front cluster and a best yaw angle corresponding to the target scanning waypoint according to the reason for insufficient observation corresponding to the current effective front in the current front cluster, so as to control a flying vehicle to scan a target environment based on the target scanning waypoint and the corresponding best yaw angle. The present application can distinguish the reasons for insufficient observation of voxels, and combines the normal vector during clustering, thereby improving the overall scanning efficiency and observation quality.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional reconstruction, and more specifically, to environmental scanning methods, apparatus, electronic devices, and media for three-dimensional reconstruction. Background Technology

[0002] 3D environmental data is a core infrastructure for digital transformation. In many industries such as power line inspection and building surveying, high-quality 3D models are a prerequisite for realizing digital twins and intelligent operation and maintenance. With the development of AI technologies such as embodied intelligence, 3D spatial data has become a key bridge connecting the physical and digital worlds, and the efficient and automated acquisition of large-scale 3D data has become an important "digital infrastructure" requirement.

[0003] Unmanned aerial vehicles (UAVs) equipped with lidar are crucial platforms for acquiring 3D data. The quality of their reconstruction depends on scan coverage and observation quality, requiring them to autonomously perform target identification, waypoint generation, and global route planning. However, current technologies cannot effectively cover all scanned surfaces during waypoint generation, reducing overall scanning efficiency and observation quality. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an environmental scanning method, apparatus, electronic device and medium for 3D reconstruction, which can distinguish the reasons for insufficient observation of voxels and incorporate normal vectors during clustering to improve overall scanning efficiency and observation quality.

[0005] In a first aspect, embodiments of this application provide an environmental scanning method for three-dimensional reconstruction, the method comprising: The system obtains the current effective frontier determined by the depth map within the current planning period, the reasons for insufficient observation corresponding to the current effective frontier, and the first historical effective frontier within the first historical frontier cluster in the target cluster list that overlaps with the current effective frontier; the depth map contains the first distance between each scanned pixel in the target environment and the nearest obstacle on the corresponding scan ray; the frontier refers to the voxel that needs to be scanned by the aircraft; Clustering is performed on the current effective frontier and the first historical effective frontier to obtain current frontier clusters that satisfy the normal vector consistency condition and spatial distance constraint; the normal vector consistency condition is used to constrain the angle between the normal vectors of two frontiers in the same frontier cluster; Based on the reasons for insufficient observation corresponding to the current effective frontier in the current frontier cluster and the reasons for insufficient observation corresponding to the second historical effective frontier in the second historical frontier cluster in the target cluster list, the target scanning waypoints with the highest comprehensive scores and the optimal yaw angles corresponding to each target scanning waypoint are generated for each current frontier cluster and the second historical frontier cluster. The aircraft is then controlled to scan the target environment based on the target scanning waypoints and the corresponding optimal yaw angles. The comprehensive score corresponding to the target scanning waypoint is obtained by comprehensively scoring the target scanning waypoint from the dimensions of visibility, distance, and orientation consistency.

[0006] In one possible implementation, obtaining the current effective frontier determined from the depth map within the current planning period and the corresponding reasons for insufficient observation includes: Within the current planning cycle, based on the first distance between each scanned pixel in the depth map and the nearest obstacle on the corresponding scanning ray, the field of view boundary point and gap point of the aircraft in the target environment are determined; the gap point refers to the pixel used to indicate the occlusion boundary or the scanning blind zone of the aircraft. Based on the gap points and the field of view boundary points, determine the current effective frontier in the set of voxels to be updated and the reasons for insufficient observation corresponding to each current effective frontier.

[0007] In one possible implementation, the field-of-view boundary point and gap point of the aircraft in the target environment are determined according to the following steps: Based on the ratio of the first distance between each scanned pixel and its corresponding neighboring pixel in the depth map and the first distance between each scanned pixel and its corresponding neighboring pixel on the corresponding scanning ray and the nearest obstacle, gap points are determined from all scanned pixels. Based on the comparison between the first distance corresponding to each scanned pixel and the preset effective distance threshold, each scanned pixel is divided into an effective region and an invalid region; pixels in the effective region have real obstacle echoes on the corresponding scanning ray; The boundary point between the effective region and the invalid region is defined as the field of view boundary point.

[0008] In one possible implementation, determining the current effective frontier in the set of voxels to be updated and the reasons for insufficient observation corresponding to each current effective frontier, based on the gap point and the field of view boundary point, includes: For each first candidate frontier in the set of voxels to be updated that is not marked as sufficiently observed, if the first candidate frontier does not meet the close-range strong trust determination condition and the conventional sufficient observation determination condition, then the first candidate frontier is determined as the second candidate frontier; the close-range strong trust determination condition is that the second distance from the voxel to the scanning sensor in the aircraft is less than a preset forced trust distance threshold and is not a field of view boundary point; the conventional sufficient observation determination condition is that the second distance from the voxel to the scanning sensor in the aircraft is less than a preset trust distance and is not a gap point or a field of view boundary point; For each second candidate front, if the voxel observation quality of the second candidate front at a preset pixel distance in all preset directions of the corresponding normal vector is not sufficiently observed, then the second candidate front is determined as the third candidate front. For each third candidate frontier, if there is at least one voxel in the neighborhood voxel of the third candidate frontier that is in a sufficiently observed state, then the third candidate frontier is determined as the current valid frontier; based on the numerical comparison between the second distance corresponding to the current valid frontier and the preset trust distance, the reason for insufficient observation corresponding to the current valid frontier is determined.

[0009] In one possible implementation, the method further includes: If the first candidate front meets the close-range strong trust determination condition or the conventional sufficient observation determination condition, then determine whether the number of observed angle intervals of the first candidate front is greater than or equal to a preset coverage threshold. If the number of observed angles of the first candidate frontier is less than the preset coverage threshold, then the first candidate frontier is determined as the current effective frontier.

[0010] In one possible implementation, the corresponding target scan waypoint and corresponding optimal yaw angle are generated for any leading-edge cluster according to the following steps: Based on the reasons for insufficient observation corresponding to the effective front in the current front cluster, candidate scan waypoints and the optimal yaw angles corresponding to each candidate scan waypoint are generated for the current front cluster. Each candidate waypoint is comprehensively scored across multiple evaluation dimensions to obtain a comprehensive score. These evaluation dimensions include visibility, distance, and orientation consistency. Visibility measures the potential exploration value of a candidate waypoint in an unknown environment. Distance measures the spatial distance cost between the candidate waypoint and the aircraft's current position. Orientation consistency measures the consistency between the direction from the aircraft's current position to the candidate waypoint and the aircraft's current direction of motion. The candidate scan waypoint with the highest overall score is determined as the target scan waypoint corresponding to the current leading-edge cluster.

[0011] In one possible implementation, controlling the aircraft to scan the target environment based on the target waypoints and the corresponding optimal yaw angle includes: Based on the topological road network, a flight cost matrix is ​​constructed; the topological road network contains multiple environmental nodes and edges between reachable pairs of environmental nodes; the weight value of each edge corresponds to the actual path length between the corresponding pairs of environmental nodes; the environmental nodes are obtained by dividing the scanned environmental regions in the target environment; the flight cost matrix contains the flight cost between the scanned waypoints of the target. The flight order between all target scan waypoints is optimized based on the flight cost matrix to obtain the latest global scan route. The aircraft is then controlled to scan the target environment based on the latest global scan route and the optimal yaw angle corresponding to each target scan waypoint.

[0012] Secondly, embodiments of this application also provide an environmental scanning device for three-dimensional reconstruction, the device comprising: The acquisition module is used to acquire the current effective frontier determined by the depth map within the current planning period and the reasons for insufficient observation corresponding to the current effective frontier, as well as the first historical effective frontier in the first historical frontier cluster in the target cluster list that overlaps with the current effective frontier; the depth map contains the first distance between each scanned pixel in the target environment and the nearest obstacle on the corresponding scan ray; the frontier refers to the voxel that needs to be scanned by the aircraft; The clustering module is used to cluster the current effective frontier and the first historical effective frontier to obtain current frontier clusters that satisfy the normal vector consistency condition and spatial distance constraint; the normal vector consistency condition is used to constrain the angle between the normal vectors of two frontiers in the same frontier cluster; The generation module is used to generate target scanning waypoints with the highest comprehensive scores and corresponding optimal yaw angles for each current frontier cluster and the second historical frontier cluster, based on the insufficient observation reasons corresponding to the current effective frontier in the current frontier cluster and the insufficient observation reasons corresponding to the second historical effective frontier in the target cluster list. This allows the aircraft to scan the target environment based on the target scanning waypoints and their corresponding optimal yaw angles. The comprehensive score corresponding to the target scanning waypoint is obtained by comprehensively scoring the target scanning waypoint from the dimensions of visibility, distance, and orientation consistency.

[0013] In one possible implementation, the acquisition module is specifically used to determine, within the current planning cycle, the field-of-view boundary points and gap points of the aircraft in the target environment based on the first distance between each scanned pixel in the depth map and the nearest obstacle on the corresponding scan ray; the gap points refer to pixels used to indicate occlusion boundaries or scan blind spots of the aircraft; and based on the gap points and the field-of-view boundary points, determine the current effective frontier in the voxel set to be updated and the reasons for insufficient observation corresponding to each current effective frontier.

[0014] In one possible implementation, the acquisition module is specifically configured to determine the field-of-view boundary points and gap points of the aircraft in the target environment according to the following steps: determining gap points from all scanned pixels based on the ratio of a first distance between each scanned pixel and its corresponding adjacent pixel in the depth map and the first distance between each scanned pixel and its corresponding adjacent pixel and the nearest obstacle on the corresponding scanning ray; dividing each scanned pixel into an effective region and an invalid region based on a numerical comparison between the first distance corresponding to each scanned pixel and a preset effective distance threshold; ensuring that the pixels in the effective region have real obstacle echoes on the corresponding scanning ray; and determining the boundary point between the effective region and the invalid region as the field-of-view boundary point.

[0015] In one possible implementation, the acquisition module is specifically configured to, for each first candidate frontier not marked as sufficiently observed in the voxel set to be updated, if the first candidate frontier does not meet the close-range strong trust determination condition and the conventional sufficient observation determination condition, then the first candidate frontier is determined as a second candidate frontier; the close-range strong trust determination condition is that the second distance from the voxel to the scanning sensor in the aircraft is less than a preset forced trust distance threshold and is not a field of view boundary point; the conventional sufficient observation determination condition is that the second distance from the voxel to the scanning sensor in the aircraft is less than a preset trust distance and is not a gap point or a field of view boundary point; for each second candidate frontier, if the voxel observation quality at a preset pixel distance in all preset directions of the corresponding normal vector of the second candidate frontier is not sufficiently observed, then the second candidate frontier is determined as a third candidate frontier; for each third candidate frontier, if there is at least one voxel in the neighborhood voxels of the third candidate frontier with a sufficiently observed state, then the third candidate frontier is determined as the current effective frontier; and the reason for insufficient observation corresponding to the current effective frontier is determined based on the numerical comparison result between the second distance corresponding to the current effective frontier and the preset trust distance.

[0016] In one possible implementation, the acquisition module is further configured to determine whether the number of observed angle intervals of the first candidate frontier is greater than or equal to a preset coverage threshold if the first candidate frontier satisfies the close-range strong trust determination condition or the conventional sufficient observation determination condition; and if the number of observed angles of the first candidate frontier is less than the preset coverage threshold, then the first candidate frontier is determined as the current valid frontier.

[0017] In one possible implementation, the generation module is specifically configured to generate a corresponding target scan waypoint and a corresponding optimal yaw angle for any leading-edge cluster according to the following steps: Based on the reasons for insufficient observation corresponding to the effective front in the current front cluster, candidate scan waypoints and the optimal yaw angles corresponding to each candidate scan waypoint are generated for the current front cluster. Each candidate waypoint is comprehensively scored across multiple evaluation dimensions to obtain a comprehensive score. These evaluation dimensions include visibility, distance, and orientation consistency. Visibility measures the potential exploration value of a candidate waypoint in an unknown environment. Distance measures the spatial distance cost between the candidate waypoint and the aircraft's current position. Orientation consistency measures the consistency between the direction from the aircraft's current position to the candidate waypoint and the aircraft's current direction of motion. The candidate scan waypoint with the highest overall score is determined as the target scan waypoint corresponding to the current leading-edge cluster.

[0018] In one possible implementation, the generation module is specifically used to construct a flight cost matrix based on a topological road network; the topological road network includes multiple environmental nodes and edges between reachable pairs of environmental nodes; the weight value corresponding to each edge is the actual path length between the corresponding pairs of environmental nodes; the environmental nodes are obtained by dividing the scanned environmental areas in the target environment; the flight cost matrix includes the flight cost between target scanning waypoints; the flight order between all target scanning waypoints is optimized based on the flight cost matrix to obtain the latest global scanning route, so as to control the aircraft to scan the target environment based on the latest global scanning route and the optimal yaw angle corresponding to each target scanning waypoint.

[0019] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the environment scanning method for three-dimensional reconstruction as described in any of the first aspects.

[0020] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the environment scanning method for three-dimensional reconstruction as described in any of the first aspects.

[0021] This application provides an environment scanning method, apparatus, electronic device, and medium for 3D reconstruction. The method includes: acquiring the current effective frontier and its reasons for insufficient observation determined based on a depth map; clustering the current effective frontier to obtain current frontier clusters that satisfy the normal vector consistency condition and spatial distance constraint; generating corresponding target scanning waypoints and optimal yaw angles for the current frontier clusters based on the reasons for insufficient observation corresponding to the current effective frontiers, so as to control the aircraft to scan the target environment based on the target scanning waypoints and the corresponding optimal yaw angles. This application can distinguish the reasons for insufficient observation of voxels and incorporates normal vectors during clustering, thereby improving the overall scanning efficiency and observation quality. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of an environmental scanning method for 3D reconstruction provided in an embodiment of this application is shown; Figure 2 A flowchart illustrating the acquisition of the effective frontier provided in an embodiment of this application is shown; Figure 3 A flowchart illustrating the determination of field boundary points and gap points provided in an embodiment of this application is shown. Figure 4 A flowchart illustrating the determination of the current effective frontier and the reasons for its insufficient observation, provided in an embodiment of this application, is shown. Figure 5 A flowchart illustrating the generation process of target scanning waypoints and their optimal yaw angles provided in an embodiment of this application is shown. Figure 6 A flowchart illustrating the process of controlling an aircraft to scan the target environment, as provided in an embodiment of this application, is shown. Figure 7 A schematic diagram illustrating the observation quality provided in an embodiment of this application is shown; Figure 8 A schematic diagram of the multi-angle coverage determination mechanism provided in an embodiment of this application is shown; Figure 9 This paper illustrates a schematic diagram of the gap point and field of view boundary provided in an embodiment of this application. Figure 10 A schematic diagram of a depth map provided in an embodiment of this application is shown; Figure 11 A comparison diagram of normal vector-driven clustering and pure distance clustering provided in the embodiments of this application is shown; Figure 12 This illustration shows a schematic diagram of bar-shaped sampling waypoint generation provided in an embodiment of this application; Figure 13 This paper illustrates a schematic diagram of the evaluation dimensions for candidate scan waypoints provided in an embodiment of this application. Figure 14 A schematic diagram of the topology network provided in an embodiment of this application is shown; Figure 15 This illustration shows a schematic diagram of TSP global route optimization provided in an embodiment of this application; Figure 16 This paper shows a schematic diagram of the structure of an environmental scanning device for three-dimensional reconstruction provided in an embodiment of this application; Figure 17 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0025] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0026] To enable those skilled in the art to utilize the content of this application, and in conjunction with the specific application scenario of "3D reconstruction," the following implementation methods are provided. For those skilled in the art, the general principles defined herein can be applied to other embodiments and application scenarios without departing from the spirit and scope of this application. Although this application is primarily described in the "3D reconstruction field," it should be understood that this is merely an exemplary embodiment.

[0027] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0028] 3D environmental data is a core infrastructure for digital transformation. In numerous industry scenarios such as power line inspection, factory surveying, building mapping, security monitoring, and mining, high-quality 3D environmental models are prerequisites for realizing digital twins, intelligent operation and maintenance, and automated decision-making. With the rapid development of artificial intelligence technologies such as embodied intelligence and large language models, 3D spatial data is becoming a crucial bridge connecting the physical and digital worlds—embodied intelligent agents need to perceive, understand, and act in a 3D environment, and the spatial reasoning capabilities of large models also rely on high-quality 3D scene representations. Therefore, efficiently and automatically acquiring 3D data of large-scale environments has become a crucial "digital infrastructure" requirement supporting the implementation of these technologies.

[0029] Unmanned aerial vehicles (UAVs) equipped with lidar are an important platform for acquiring large-scale 3D environmental data. The quality of 3D reconstruction is highly dependent on the coverage and observation quality of the original scan data—the UAV needs to fully scan various surfaces in the environment from appropriate distances and angles. This requires the UAV to autonomously decide: (1) which surface areas need to be scanned (target identification); (2) from which location to scan most effectively (waypoint generation); and (3) in what order to visit waypoints most efficiently (global route planning). However, existing technologies still have the following main shortcomings in the above three stages: (i) The coarse-grained target identification fails to distinguish between different causes of insufficient observation. Existing autonomous exploration methods typically divide the environment into two simple states: "known" and "unknown," identifying scanned targets by detecting the boundary (front edge) between known and unknown regions. However, this binary classification cannot distinguish the specific reasons for insufficient observation—whether it is due to excessive observation distance leading to sparse point clouds, or due to poor observation perspective (such as occlusion boundaries or sensor field of view edges) resulting in incomplete data. Different causes of insufficient observation require different supplementary scanning strategies (the former requires closer proximity, the latter requires changing the angle), and coarse-grained classification cannot provide precise guidance for waypoint generation.

[0030] (ii) Front clustering based solely on spatial distance ignores the directionality of scanning requirements. Existing methods typically employ Euclidean distance-based clustering algorithms to group fronts, classifying spatially adjacent fronts as belonging to the same scanning task. However, distance-based clustering may mix fronts facing different directions in the same group. For example, at the corner of an L-shaped corridor, the front normal vector of one wall faces east, while the front normal vector of another wall faces north. These two groups of fronts are spatially adjacent, but their scanning requirements are perpendicular to each other. Simple distance clustering would merge them into the same group, resulting in the cluster's principal normal vector becoming a compromise average of the two directions. Waypoints generated accordingly cannot effectively scan the eastern wall or the northern wall, reducing observation quality and scanning efficiency.

[0031] (iii) Waypoint evaluation methods are simplistic and lack comprehensive consideration of flight efficiency. Existing waypoint evaluation methods typically only consider information gain (i.e., the area of ​​the unknown region that can be observed from the waypoint), while ignoring factors that affect actual data acquisition efficiency, such as flight distance and consistency of motion direction. This may lead to the aircraft frequently selecting waypoints with the highest information gain but which are far away, resulting in a large number of unnecessary round trips and reducing overall data acquisition efficiency.

[0032] (iv) Lack of or oversimplification of global route planning. Some existing methods employ a greedy strategy (selecting the nearest or most information-gain waypoint each time), lacking global optimization. This local optimum strategy easily leads to the aircraft repeatedly shuttling between explored and unexplored areas, resulting in a large amount of redundant flight. Other methods, while introducing TSP solutions, use Euclidean distance as an approximation of the cost between waypoints, ignoring the impact of obstacles on the flight path in the actual environment, resulting in a large deviation between the planned route and the actual executable route.

[0033] (v) Map representation methods have high memory overhead and are difficult to scale to large-scale environments. Existing mainstream methods generally use voxel grids or octrees as environment representations, and memory usage increases cubically with the environment volume. For large-scale environments (such as factory areas or building complexes with a square span of several hundred meters), the memory usage of voxel grids can reach several GB, which poses a serious resource bottleneck for unmanned aerial vehicles deployed on embedded platforms.

[0034] In summary, existing technologies have shortcomings in terms of target recognition accuracy, waypoint generation quality, global route optimization, and large-scale environmental adaptability, which restricts the application of unmanned aerial vehicles in high-quality 3D reconstruction in fields such as inspection, surveying, and security.

[0035] In view of the above, the embodiments of this application aim to solve the following main technical problems existing in the prior art: (1) Existing scanning target identification methods are coarse-grained and cannot distinguish between two different reasons for data loss: insufficient observation distance and poor observation angle. Therefore, they cannot provide accurate scanning requirement information for waypoint generation. (2) Existing front clustering methods are based only on spatial distance and ignore the normal vector information of the front, which leads to the incorrect merging of fronts in different scanning directions; (3) Existing waypoint evaluation methods only consider information gain and lack a comprehensive evaluation of flight distance and motion continuity; (4) Existing global route planning methods are lacking or use Euclidean distance approximation, and do not consider the impact of obstacles on flight costs in the actual environment; (5) Existing map representation methods have high memory overhead and are difficult to support autonomous scanning of large-scale environments.

[0036] The following is a detailed description of an environmental scanning method for 3D reconstruction provided by an embodiment of this application.

[0037] Reference Figure 1 The diagram shown is a flowchart illustrating an environmental scanning method for 3D reconstruction provided in an embodiment of this application. The exemplary steps of this embodiment are described below: S101. Obtain the current effective frontier determined by the depth map within the current planning period and the reasons for insufficient observation corresponding to the current effective frontier, as well as the first historical effective frontier in the first historical frontier cluster that overlaps with the current effective frontier in the target cluster list.

[0038] In this embodiment, scanning waypoints and their corresponding optimal yaw angles are regenerated for the aircraft according to a preset planning cycle to control the aircraft to scan the target environment. The scanned environmental data is then used to perform 3D reconstruction of the target environment. The target environment refers to the scene area requiring 3D reconstruction. The leading edge refers to the voxels that the aircraft needs to scan; insufficient observation reasons refer to the reasons why the corresponding voxels have not been fully observed and require further scanning.

[0039] The depth map includes the first distance between each scanned pixel in the target environment and the nearest obstacle along its corresponding scan ray. A scanned pixel refers to a pixel in the target environment that has been scanned by the aircraft. The corresponding scan ray is the ray emitted by the aircraft's scanning sensor when scanning that pixel.

[0040] In addition, this application's implementation also maintains a target clustering list, which stores frontier clusters considered unreachable by the aircraft during historical planning cycles. The first historical frontier cluster refers to the frontier cluster in the target clustering list that overlaps with the current effective frontier. The first historical effective frontier refers to the historical effective frontier within the first historical frontier cluster.

[0041] Specifically, refer to Figure 2 The diagram shown is a flowchart illustrating the acquisition of the effective frontier according to an embodiment of this application. Figure 2 As shown, the current effective frontier determined based on the depth map within the current planning period and the corresponding reasons for insufficient observation are obtained, including: S201. Within the current planning cycle, based on the first distance between each scanned pixel in the depth map and the nearest obstacle on the corresponding scanning ray, determine the field-of-view boundary point and gap point of the aircraft in the target environment.

[0042] In this application embodiment, the field of view boundary point is the boundary point of the scanned environmental area in the target environment, caused by the scanning range limitation of the scanning sensor in the aircraft. The gap point refers to the pixel point used to indicate the occlusion boundary or the scanning blind zone of the aircraft. It is caused by the geometric relationship (occlusion) of objects in the target scene and exists inside the scanned environmental area in the target environment. It indicates that there is an occlusion boundary at the gap point, and there may be an unscanned area behind it.

[0043] Here, this embodiment of the application maintains a depth map to store the first distances between each scanned pixel in the target environment and the nearest obstacle on the corresponding scan ray. During initialization, all initial element values ​​in the depth map are set to a negative value close to zero (indicating no point cloud echo on the corresponding scan ray). After point cloud projection, the element values ​​of pixels with obstacle echoes are written as positive values ​​(i.e., the distance of the pixel to the nearest obstacle on the corresponding scan ray). Furthermore, the positional relationships of the pixels in the depth map are the same as the positional relationships of the corresponding pixels in the target environment.

[0044] Specifically, when the scanning sensor detects new 3D point cloud data in the target environment, it transforms the new 3D point cloud data into the sensor's local coordinate system and projects it onto a 2D depth map formed by a spherical coordinate system, so as to store the first distance between each new pixel and the nearest obstacle on the corresponding scanning ray as the value of the corresponding element.

[0045] Furthermore, referring to Figure 3 The diagram shown is a flowchart illustrating the determination of field boundary points and gap points according to an embodiment of this application; as follows: Figure 3 As shown, the field-of-view boundary points and gap points of the aircraft in the target environment are determined through the following steps: S301. Based on the ratio of the first distance between each scanned pixel and its corresponding adjacent pixel in the depth map and the first distance between each scanned pixel and the nearest obstacle on the corresponding scanning ray, determine the gap point from all scanned pixels.

[0046] In this embodiment of the application, each scanned pixel in the depth map has 8 neighboring pixels (i.e. (All adjacent pixels within the window); For each scanned pixel, if the ratio of the first distance between the scanned pixel and at least one corresponding adjacent pixel is less than a preset gap ratio, or if the first distance between at least one adjacent pixel corresponding to the scanned pixel and the nearest obstacle on the corresponding scan ray is an initial value (indicating that there is no obstacle echo in this direction, meaning that there may be an occlusion boundary or sensor blind spot), then the scanned pixel is a gap point. That is, a gap point is a location where the first distance in the depth map changes abruptly. Specifically, the ratio of the first distance between any scanned pixel and any corresponding adjacent pixel is determined by the following formula: ; in, It is the ratio of the first distance between the scanned pixel and its corresponding neighboring pixel. This represents the first distance between the scanned pixel and the nearest obstacle on the corresponding scan ray. This represents the first distance between the nearest obstacle and the adjacent pixel of the scanned pixel on the corresponding scan ray.

[0047] Furthermore, after "determining gap points from all scanned pixels based on the ratio of the first distance between each scanned pixel and its corresponding neighboring pixel in the depth map and the first distance between each scanned pixel and its corresponding neighboring pixel on the corresponding scanning ray and the nearest obstacle", the method further includes: marking the gap points in the depth map; after marking, performing a morphological dilation operation on the marked gap points in the depth map, and also determining the neighboring pixels of the gap points in the depth map after the morphological dilation operation as gap points, so as to expand the coverage of the gap point marking, so that the observation data near the occlusion boundary is also included in the judgment range of poor viewpoint, and avoids low-quality observations at the boundary being misjudged as sufficient observations.

[0048] S302. Based on the numerical comparison between the first distance corresponding to each scanned pixel and the preset effective distance threshold, each scanned pixel is divided into an effective region and an invalid region.

[0049] In this embodiment, pixels in the effective region exhibit real obstacle echoes on the corresponding scanning ray; pixels in the invalid region do not exhibit real obstacle echoes on the corresponding scanning ray, or are subject to interference from the aircraft itself at very close range, such as noise points generated by rotors, sensor protective covers, etc. For any scanned pixel, if the first distance corresponding to the scanned pixel is greater than a preset effective distance threshold (a small positive number close to zero, typically 0.1m), then the scanned pixel is classified into the effective region; if the first distance corresponding to the scanned pixel is less than or equal to the preset effective distance threshold, then the scanned pixel is classified into the invalid region.

[0050] S303. Determine the boundary point between the effective region and the invalid region as the field of view boundary point.

[0051] In this embodiment, the boundary between the effective and invalid regions is detected, and the first pixel at the boundary that falls within the effective region is marked as the sensor's field of view (FOV) boundary point. The observation quality at the FOV boundary is poor because these points are located at the edge of the sensor's scanning coverage area, have low point cloud density, and are easily affected by noise.

[0052] S202. Based on the gap points and field boundary points, determine the current effective frontier in the set of voxels to be updated and the reasons for insufficient observation corresponding to each current effective frontier.

[0053] In this embodiment, when the scanning sensor detects new 3D point cloud data in the target environment, a first voxel into which the new point cloud falls is created in the observation quality map. The first voxel into which the new point cloud falls, along with a second voxel in the neighborhood of the first voxel in the observation quality map that is not sufficiently observed, are added to the set to be updated. This updates the observation quality of the voxels in the set to be updated, resulting in the current effective frontier in the set to be updated and the reasons for insufficient observation corresponding to each current effective frontier. The frontier is a surface voxel in the observation quality map whose observation quality is substandard and which is adjacent to a sufficiently observed voxel. It represents the surface area in the target environment that needs further scanning and is the basis for generating scan waypoints.

[0054] The observation quality map employs an on-demand creation strategy. A voxel is added to the map (stored in a spatial hash table) only when it falls within the scanned point cloud; unobserved spaces do not occupy storage. As the scan progresses, all physical surfaces within a given scan area are sequentially added to the map and undergo observation quality evaluation. The scan task is completed when all surface voxels on the map have reached a fully observed state. In other words, the observation quality map stores the observation quality of voxels within the scanned point cloud; observation quality refers to the degree of sufficient scanning of the spatial region where the voxel resides, including both sufficient and insufficient observation. Furthermore, sufficient observation is affected by both observation distance and observation angle; therefore, the observation quality map also marks the reasons for insufficient observation (insufficient observation distance or poor observation angle) for voxels that have not been fully observed.

[0055] Specifically, refer to Figure 4 The diagram shown is a flowchart illustrating the determination of the current effective frontier and the reasons for insufficient observation provided in an embodiment of this application. Figure 4 As shown, based on the gap points and field boundary points, the current effective frontier in the set of voxels to be updated and the reasons for insufficient observation corresponding to each current effective frontier are determined, including: S401. For each first candidate frontier in the set of voxels to be updated that is not marked as sufficiently observed, if the first candidate frontier does not meet the close-range strong confidence criterion and the regular sufficient observation criterion, then the first candidate frontier is determined as the second candidate frontier.

[0056] In this embodiment, when the observation quality of a voxel is "sufficient observation," it means that the voxel has been observed from a suitable distance and angle and does not need to be scanned again. Sufficient observation is the final state of the voxel, and subsequent frames will not downgrade it to other states. Therefore, voxels that have been marked as sufficiently observed in the observation map will not be identified as frontiers. Then, voxels in the set of voxels to be updated that have not been marked as sufficiently observed are identified as first candidate frontiers. For each first candidate frontier, it is determined whether the first candidate frontier meets the close-range strong trust determination condition and the regular sufficient observation determination condition. If the first candidate frontier does not meet the close-range strong trust determination condition and the regular sufficient observation determination condition, then the first candidate frontier is identified as the second candidate frontier.

[0057] The criteria for strong trust at close range are: the second distance from the voxel to the scanning sensor in the aircraft is less than a preset forced trust distance threshold, and it is not a field-of-view boundary point; the second distance refers to the distance from the corresponding voxel; the preset forced trust distance threshold is used to characterize the observation distance at which the scanning sensor can definitely observe sufficiently and will not be affected by the observation angle. The criteria for sufficient observation at normal range are: the second distance from the voxel to the scanning sensor in the aircraft is less than the preset trust distance, and it is not a gap point or a field-of-view boundary point (i.e., a good viewing angle). The preset trust distance is used to characterize the distance at which the scanning sensor can observe sufficiently, but it will be affected by the viewing angle.

[0058] If the first candidate frontier meets the criteria for strong confidence at close range or the criteria for sufficient observation, the observation quality of the first candidate frontier can be directly marked as sufficient observation.

[0059] Optionally, for 3D reconstruction applications, observing the same surface from only a single angle is often insufficient to obtain high-quality reconstruction results—multi-angle observation can provide more uniform point cloud density, more accurate normal vector estimation, and more complete surface coverage. Therefore, embodiments of this application further propose a multi-angle coverage determination mechanism based on the distribution of observation angles. Specifically: if the first candidate front meets the close-range strong confidence determination condition or the conventional sufficient observation determination condition, then it is determined whether the number of observed angles of the first candidate front is greater than or equal to a preset coverage threshold; if the number of observed angle intervals of the first candidate front (typically 2~3, i.e., requiring observation from at least 2~3 different directions) is less than the preset coverage threshold, then the first candidate front is determined as the current valid front.

[0060] Here, in this embodiment of the application, an observation angle distribution record is maintained for each voxel in the observation quality map to track which directions the voxel has been observed from, thereby guiding the aircraft to perform supplementary scanning of the voxel area from unobserved angles. Specifically, the observation hemisphere space with the voxel normal vector as the axis is divided into K angle intervals (for example, dividing the horizontal 360° into 4 quadrants, K=4), and a K-bit bitmask is used to represent the coverage status (covered and uncovered) of each observation angle interval. Then, when a new scanning sensor scan frame is received, for each voxel that meets the single observation quality requirements (distance and viewing angle are qualified), the azimuth angle (observation angle) of the current observation direction relative to the voxel normal vector is calculated, the angle interval k to which it belongs is determined, and the bitmask is applied. The k-th position is marked as 1: The storage overhead of this mechanism is minimal (only 1-2 bytes per voxel), but it significantly improves the quality of 3D reconstruction: multi-angle observation makes the point cloud density of the same surface more uniform, the normal vector estimation more accurate, and occluded areas easier to discover and fill in, ultimately improving the integrity and accuracy of the 3D model.

[0061] In a simplified implementation, instead of using an observation angle interval bitmask, an observation counter can be maintained, and observations are marked as sufficient once a threshold is reached. This simplified approach is easier to implement, but it cannot guarantee the diversity of observation angles.

[0062] S402. For each second candidate frontier, if the voxel observation quality at a preset pixel distance in all preset directions of the corresponding normal vector is not sufficiently observed, then the second candidate frontier is determined as the third candidate frontier.

[0063] In this application embodiment, the preset direction may include a positive direction and a negative direction; the preset pixel distance may be at a voxel distance.

[0064] For example, along the second candidate front, check the voxel observation quality at one voxel distance in the positive direction of the corresponding normal vector, one voxel distance in the negative direction, and two voxel distances in the negative direction. If a voxel with sufficient observation exists at any of the above three locations, it means that the second candidate front is adjacent to a surface that has been sufficiently observed, and the determination of insufficient observation may be due to noise at the surface edge, so it is marked as sufficiently observed; if no voxel with sufficient observation exists at any of the above three locations, then the second candidate front is determined as the third candidate front.

[0065] Among them, the normal vector is the unit vector perpendicular to the local surface. It is calculated by principal component analysis (PCA) of the neighborhood point cloud: the eigenvector corresponding to the smallest eigenvalue of the neighborhood point cloud covariance matrix is ​​taken and the direction is adjusted to point to the side of the scanning sensor (i.e. the side of the observed area). It is used to describe the orientation of the observed surface to which the leading edge is attached, and to provide a geometric basis for subsequent noise suppression and scanning direction determination.

[0066] S403. For each third candidate frontier, if there is at least one voxel in the neighborhood voxel of the third candidate frontier that has a sufficiently observed state, then the third candidate frontier is determined as the current valid frontier; based on the numerical comparison result between the second distance corresponding to the current valid frontier and the preset trust distance, the reason for insufficient observation corresponding to the current valid frontier is determined.

[0067] In this embodiment of the application, the 26-neighborhood of the third candidate front (i.e. If at least one sufficiently observed voxel exists among all adjacent pixels within the window, then the third candidate front is the current valid front, the reason for its insufficiency is recorded, and it is ultimately marked as a distance-insufficient front or a poor-viewpoint front, and its normal vector is recorded for subsequent clustering. If it does not exist, then the voxel is an isolated point with insufficient observation, does not constitute a valid front, and is marked as sufficiently observed.

[0068] Specifically, if the second distance corresponding to the current effective frontier is greater than the preset trust distance, then the reason for insufficient observation of the current effective frontier is insufficient distance; otherwise (i.e., the frontier is a gap point or a field boundary point), the reason for insufficient observation of the current effective frontier is poor viewing angle. In this way, when there are two reasons for insufficient observation of the frontier, insufficient distance is used as the main classification criterion.

[0069] S102. Cluster the current efficient frontier and the first historical efficient frontier to obtain the current frontier cluster that satisfies the normal vector consistency condition and spatial distance constraint.

[0070] In this application's implementation, the first historical effective frontier refers to the historical frontier contained in the first historical frontier cluster that overlaps with the current effective frontier in the target cluster list. When the region where the member frontier of an unreachable frontier cluster in the target cluster list is located is covered by new scanning sensor data, a disassembly-re-clustering operation is performed on that frontier cluster: its member voxels that are still in the frontier state are merged with the newly added frontier voxels. This mechanism ensures that temporarily unreachable scanning targets are not permanently missed during the online exploration process where the environment is gradually revealed. For clusters that are always unreachable due to physical constraints (such as channel width being less than the aircraft's safe distance), they will remain unreachable in multiple rounds of reassessment, not participating in global route solving, but also not affecting the normal planning of other reachable clusters.

[0071] Specifically, this application proposes a normal vector-driven clustering method, which adds a normal vector consistency constraint to the spatial distance constraint, ensuring that only spatially adjacent frontiers with the same scanning direction are grouped into the same frontier cluster. The neighborhood of a cluster is defined as a pair of frontiers that simultaneously satisfy the following two conditions: Condition 1 (Spatial distance constraint): Euclidean distance between the two fronts Smaller than the cluster radius ; ; in, The position corresponding to the first frontier. This corresponds to the position of the second frontier.

[0072] Condition 2 (Normal Vector Consistency): The dot product of the two frontal normal vectors is greater than the direction threshold. ; ; in, Let be the normal vector corresponding to the first front. Let be the normal vector corresponding to the second front; the normal vector consistency condition is used to constrain the angle between the normal vectors of two fronts in the same front cluster.

[0073] Here, the clustering process employs an improved density clustering algorithm (such as DBSCAN) to expand the clusters based on the neighborhood relationships that satisfy the aforementioned dual constraints. During the expansion process, the spatial extent of the clusters is monitored in real time, and expansion stops when the maximum side length of the bounding box exceeds the size limit to prevent the clusters from becoming too large.

[0074] The output information for each front cluster includes: the cluster center location, the cluster principal normal vector (the normalized mean of all member normal vectors), the set of member fronts and their normal vectors, the spatial bounding box, and the waypoint information generated subsequently.

[0075] Here, the advantage of normal vector-driven clustering is that the frontiers of different scanning requirements are correctly separated, and each cluster corresponds to a scanning task with a clear direction, providing clear observation direction guidance for subsequent waypoint generation.

[0076] Optionally, invalid frontier clusters are cleaned up. The following two cleanup methods are performed on the target cluster list: First, when the spatial extent of a newly generated front cluster is too small (the maximum side length of the bounding box is below the lower limit of the size) or the number of member fronts is insufficient, the front cluster is determined to be an invalid target and marked as dormant, not participating in subsequent waypoint generation and global route planning. Voxels that are still in the front state within invalid front clusters will be re-evaluated and processed in clustering in subsequent frames—if new observations enable them to meet the sufficient observation conditions, they will be converted to a sufficient observation state; otherwise, they may be merged with other fronts to form an effective cluster.

[0077] Secondly, when the spacecraft has reached a certain scanning waypoint, the corresponding leading cluster is marked as completed and removed from the target cluster list, and its voxel is marked as fully observed.

[0078] Here, the cleanup mechanism prevents the system from continuously consuming computing resources for targets that are too small, invalid, or already completed.

[0079] S103. Based on the reasons for insufficient observation corresponding to the current effective frontier in the current frontier cluster and the reasons for insufficient observation corresponding to the second historical effective frontier in the second historical frontier cluster in the target cluster list, generate the target scanning waypoint with the highest comprehensive score and the optimal yaw angle corresponding to each target scanning waypoint for each current frontier cluster and the second historical frontier cluster, so as to control the aircraft to scan the target environment based on the target scanning waypoint and the corresponding optimal yaw angle.

[0080] In this embodiment, the second historical front cluster refers to the historical front cluster included in the target clustering list. Scan waypoints are discrete observation positions that the aircraft needs to reach sequentially during a 3D data acquisition mission. Each scan waypoint contains 3D coordinates and the desired sensor orientation (yaw angle), defining the pose of a single scan observation. After the aircraft completes scanning the target environment, it uses the scan data to perform 3D reconstruction of the target environment. 3D reconstruction refers to the process of constructing a 3D digital model of the real environment using data collected by sensors (LiDAR, cameras, etc.). The quality of the 3D reconstruction depends on the coverage and observation quality of the original acquired data.

[0081] The overall score corresponding to the target scan waypoint is obtained by comprehensively evaluating the target scan waypoint from the dimensions of visibility, distance, and direction consistency.

[0082] Specifically, refer to Figure 5 The diagram shown is a flowchart illustrating the generation process of target scan waypoints and their optimal yaw angles according to an embodiment of this application. Specifically, the following steps are used to generate corresponding target scan waypoints and their optimal yaw angles for any leading edge cluster: S501. Based on the reasons for insufficient observation of the effective frontier in the current frontier cluster, generate candidate scan waypoints and the optimal yaw angle for each candidate scan waypoint for the current frontier cluster.

[0083] In the embodiments of this application, firstly, a bar-shaped sampling strategy is used to generate sampling points around each leading cluster.

[0084] Specifically, the cylindrical sampling space is determined by the sampling height. Sampling radius and sampling azimuth angle Three-dimensional parameterization: (1) Sampling height dimension: from the preset minimum sampling height value h_min to the preset maximum sampling height value h_max, uniformly divided into N_h layers; (2) Sampling radius dimension: from the preset minimum sampling radius r_min to the preset maximum sampling radius r_max, uniformly divided into N_r layers; (3) Sampling azimuth dimension: in [0,2 The area is evenly divided into One direction.

[0085] Let the current center position of the leading cluster be p_c, then the absolute position of each sampling point is p_c+ Among them, the offset Defined as: .

[0086] In addition, cylindrical sampling better reflects the motion characteristics of the aircraft (primarily horizontal movement) compared to spherical sampling, avoiding sampling redundancy near the polar angle. The starting azimuth angles of different altitude layers are slightly offset to increase sampling uniformity.

[0087] Then, the sampling points are filtered using the following three conditions to obtain candidate scan waypoints: (a) Obstacle distance filtering: Remove sampling points whose distance to the nearest obstacle is less than a preset safety threshold; (b) Working area filtering: Delete sampling points that exceed the preset scanning area boundary; (c) Clustering filtering: Density clustering is performed on the remaining sampling points after filtering (a) and (b) to remove isolated outliers.

[0088] Finally, based on the reasons for insufficient observation corresponding to the effective frontier in the current frontier cluster, the optimal yaw angle for the candidate scan waypoints corresponding to the current frontier cluster is generated.

[0089] In this embodiment of the application, for each candidate scan waypoint corresponding to the current frontier cluster, the number of visible frontiers of the aircraft at different yaw angles at the candidate scan position is evaluated; the yaw angle with the most visible frontiers is determined as the optimal yaw angle corresponding to the candidate scan waypoint.

[0090] Specifically, the front in the current front cluster that satisfies all of the following observation conditions is a visible front at the corresponding candidate scan waypoint and the corresponding yaw angle: (a) Observation distance condition: For a frontier that is not sufficiently observed due to insufficient distance, if the distance at which the aircraft observes the frontier at the candidate scan position and the corresponding yaw angle is less than the preset trust distance threshold, then the observation condition is satisfied; For a frontier that is not sufficiently observed due to poor viewing angle, if the distance at which the aircraft observes the frontier at the candidate scan position and the corresponding yaw angle is less than the preset forced trust distance threshold, then the observation distance condition is satisfied. (b) Observation angle condition: If the observation angle quality score of the frontal line observed at the candidate scan position and the corresponding yaw angle is greater than or equal to the preset observation quality threshold, then the observation angle condition is met. Here, the observation angle quality score is calculated based on the dot product 's' of the unit direction vector from the candidate scan waypoint to the leading edge and the normal vector of the leading edge. Its formula is as follows: ; in, Let be the unit direction vector pointing from the candidate scan waypoint to the leading edge. This is the front normal vector (pointing to one side of the observed region). When the candidate scan waypoint is located on the unobserved side of the front and directly facing the surface, and Nearly in the same direction, Approaching 1, Approaching 1; when the candidate scan waypoint is located on the flank of the leading edge (edge ​​observation), Approaching 0 Close to 0. This is a smoothing factor (typically π / 100), controlling the steepness of the transition from 0 to 1 in the score. When If the observation quality is below the observation quality threshold, the front does not meet the observation angle condition; otherwise, the front meets the observation angle condition. This design encourages the selection of waypoints on the unobserved side of the front, from which the sensor can scan the unknown region behind the front when looking at the front surface.

[0091] (c) Unobstructed observation path condition: If the ray from the front to the candidate scanning waypoint is not blocked by known obstacles (i.e. there are no voxels in the ray path that have been scanned point cloud, and there are no physical surfaces that block the line of sight), then the corresponding voxel satisfies the unobstructed observation path condition. (d) At the candidate scan position and corresponding yaw angle, the leading edge of the aircraft is within the field of view of the scanning sensor.

[0092] S502. Perform a comprehensive evaluation of each candidate scan waypoint across multiple dimensions to obtain a comprehensive score for each candidate scan waypoint.

[0093] In this application's implementation, the evaluation dimensions are visibility, distance, or orientation consistency. Visibility measures the potential exploration value of candidate scan waypoints for unknown environments; distance measures the spatial distance cost between the candidate scan waypoint and the aircraft's current position; orientation consistency measures the consistency between the direction from the aircraft's current position to the candidate scan waypoint and the aircraft's current direction of motion. Specifically, each candidate scan waypoint... Overall score The expression (suitable for scenarios with high real-time requirements) is: ; in, The number of visible frontiers at the optimal yaw angle (visibility dimension, weight 1.0). Candidate scan waypoints To the current position of the aircraft Distance (distance dimension, weight) ). It is a minimal constant that prevents division by zero; The direction from the aircraft's current position to the candidate scan waypoints relative to the direction of the aircraft's current velocity dot product (directional consistency dimension, weight) Since this method is an online planning method, the aircraft is usually in motion at the planning time. This component uses the current velocity and direction information to encourage the selection of waypoints along the direction of motion, reducing the time loss caused by frequent turns.

[0094] Here, visibility ensures scan coverage, the distance component reduces long-distance flight, and the orientation consistency component utilizes the aircraft's current motion state to reduce frequent turning. When the aircraft is stationary (such as when it has just taken off or is hovering for scanning), its speed is zero, orientation consistency is automatically set to zero, and the score degenerates into a two-objective optimization of visibility and distance.

[0095] Optionally, the visibility dimension can be normalized to coverage to avoid the problem of large clusters naturally dominating while small clusters are ignored. Simultaneously, the distance dimension can be replaced with a flight time estimate based on cruise speed to improve the overall score. With a more explicit physical meaning, its expression (suitable for scenarios with high requirements for 3D reconstruction quality) can also be: ; in, This represents the total number of leading-edge voxels corresponding to the current leading-edge cluster for this candidate scan waypoint. This refers to the aircraft's cruise speed. The physical meaning of this score is the increase in coverage that can be achieved per unit flight time, i.e., information gain efficiency. It is a minimal constant that prevents division by zero.

[0096] Optionally, when the aforementioned multi-angle coverage determination mechanism is used in the embodiments of this application, the visibility dimension is further refined to the number of newly uncovered angle intervals that the viewpoint can contribute to the leading voxels (rather than a simple count of visible voxels). This allows the scoring function to prioritize waypoints that can provide new observation angles, avoiding repeated scanning from already covered angles. The expression can be: ; in, To observe the front from this candidate scan waypoint At that time, the number of newly covered observation angle intervals can be increased (i.e., the angle interval corresponding to the observation direction has not yet been set in the bit mask of f). This is the set of visible frontiers for the candidate scan waypoint and its corresponding optimal yaw angle.

[0097] Optionally, during scoring, it can be checked whether the fronts of other frontal clusters can be observed simultaneously within the sensor field of view of the candidate scan waypoint. If a waypoint can cover multiple frontal clusters simultaneously, additional points are awarded, and the overall score is adjusted accordingly. The expression (suitable for scenarios with high requirements for 3D reconstruction quality) can be: ; in, To observe the number of other leading-edge clusters at this candidate scan waypoint, This mechanism assigns a weight to the multi-cluster coverage reward. It encourages the selection of efficient waypoints that can cover multiple targets in a single scan, reducing the total number of waypoints and flight time.

[0098] S503. The candidate scan waypoint with the highest comprehensive score is determined as the target scan waypoint corresponding to the current leading cluster.

[0099] Furthermore, referring to Figure 6 The diagram shows a flowchart illustrating the process of controlling an aircraft to scan the target environment according to an embodiment of this application. The process involves controlling the aircraft to scan the target environment based on the target waypoints and their corresponding optimal yaw angles, including: S601. Construct the flight cost matrix based on the latest road network topology.

[0100] In this embodiment, the topology network includes multiple environmental nodes and edges between reachable pairs of environmental nodes; the weight value corresponding to each edge is the actual path length between the corresponding pairs of environmental nodes; the environmental nodes are obtained by dividing the scanned environmental regions in the target environment; the flight cost matrix contains the flight cost between the target scan waypoints. Specifically, the target scan waypoints are inserted into the latest topology network using a graph search algorithm (such as A... Calculate the flight cost of the shortest path between target waypoints. The flight cost is defined as the sum of the Euclidean distances between adjacent environmental nodes along the path, and its expression is: ; in, The flight cost between target scan waypoint i and target scan waypoint j; The set of environment nodes on the shortest path between target scan waypoint i and target scan waypoint j; The three-dimensional position coordinates of the (k+1)th environmental node on the shortest path; The three-dimensional position coordinates of the k-th environmental node on the shortest path; It represents the Euclidean distance between the (k+1)th environment node and the kth environment node.

[0101] Optionally, altitude change penalties and cruise speed normalization can be further introduced to convert flight cost into a flight time estimate, then the expression for flight cost can also be: ; in, The penalty coefficient is highly variable; The height of the (k+1)th environment node. The height of the k-th environment node; This refers to the aircraft's cruising speed. The cost of this flight is measured in time.

[0102] Furthermore, for unreachable target waypoint pairs in the topological network (shortest path search failure), a large penalty is imposed on the flight cost of such target waypoint pairs to ensure that reachable paths are prioritized during global route optimization.

[0103] Furthermore, let the current position of the aircraft be node 0, and the N target waypoints be nodes 1 to N, construct a flight cost matrix M of (N+1)×(N+1).

[0104] In this embodiment, the matrix element M(i,j) represents the flight cost from node i to node j. Since the target scanning waypoint is temporarily inserted into the topology network, the graph search path from node i to node j via the topology network and the search path from j to i may pass through different intermediate nodes (affected by the search starting point, heuristic function, and graph structure). Therefore, the cost matrix may be asymmetric, i.e., M(i,j) ≠ M(j,i).

[0105] Optionally, to improve the quality of the flight cost matrix, embodiments of this application perform a triangular inequality correction (Floyd-Warshall) on the matrix:

[0106] in, Let $\matrix_i$ be the flight cost of a matrix element from node $i$ to node $k$. Let be the flight cost of the matrix element from node k to node j. This correction ensures that the flight cost matrix satisfies the triangle inequality, preventing the TSP solver from generating unreasonable routes.

[0107] Optionally, embodiments of this application introduce a return cost adjustment mechanism in the flight cost matrix: the cost of returning from the target scan waypoint to the starting point is set as follows: ; in, Based on the return trip cost, For adjustment coefficients, The distance from target scan i to the aircraft's current position. This mechanism encourages the aircraft to choose a more distant target scan waypoint as its last stop when returning after completing a scan, thus making the route generally expand outward and reducing repeated crossings of already scanned areas.

[0108] In summary, cost calculation based on topological road networks can more accurately reflect the impact of obstacles on the flight path compared to the Euclidean distance approximation, especially in scenarios where large obstacles need to be detoured, where the difference between the two is significant.

[0109] Furthermore, the topological network is incrementally updated as the spacecraft explores. Newly explored areas automatically generate new topological nodes and edges, and the connections of existing nodes are re-verified when the environment changes. The implementation process is as follows: The environmental space within the boundary of the newly scanned area is uniformly divided into three-dimensional region grids of fixed dimensions (e.g., each region is a 3.3m × 3.3m × 2.0m cuboid). The number of regions is determined by the ratio of the total size of the scanned areas to the size of a single region. For each region grid, traversable skeleton nodes are extracted through safe space sampling and connectivity analysis. Specifically: (a) Generate collision-free safe spheres (Bubble) within the region grid by recursively subdividing the space to cover the free space.

[0110] In this embodiment, the regional grid is used as the initial sub-region; in the scanned point cloud map, the distance to the nearest point to the center of the sub-region is queried; a safety sphere is constructed with this distance as the radius; if the safety sphere has covered the entire sub-region, the process ends; if the safety sphere has not covered the entire sub-region, the sub-region is further subdivided to obtain multiple new sub-regions and processed recursively until all sub-regions are covered by the safety sphere or the size of the sub-region is less than a preset threshold.

[0111] In a preferred embodiment, the subregion is further subdivided by the following steps: when the radius of the safety sphere is greater than or equal to the minimum safety radius, the inscribed cube of the safety sphere is calculated; the subregion is divided into at most 26 smaller cuboids by the faces of the inscribed cube to obtain new subregions; when the radius of the safety sphere is less than the minimum safety radius, the subregion is bisected along each axis and then recursively divided.

[0112] (b) Perform disjoint-set clustering on the overlapping safety spheres (two safety spheres are considered connected when the center distance between the two spheres is less than the sum of their radii minus the preset minimum safety radius, i.e., the overlapping area of ​​the two spheres is large enough) to obtain safety sphere clusters; take the center of the safety sphere closest to the center of the region in each safety sphere cluster as the environment node of that safety sphere cluster; (c) Perform reachability checks on environmental nodes in adjacent regions (verified through path search). Establish edges between reachable environmental node pairs, with the weight of the edge being the actual path length.

[0113] Optionally, for frontier clusters that fail the accessibility check in the current planning cycle (i.e., there is no passable path from the aircraft's current position to the target scan waypoint corresponding to the frontier cluster in the topology network), the system marks the frontier cluster as temporarily unreachable, retains it in the target cluster list instead of deleting it, but does not participate in the cost matrix construction in this round of ATSP solution.

[0114] S602. Optimize the flight order between all target scan waypoints based on the flight cost matrix to obtain the latest global scan route, and control the aircraft to scan the target environment based on the latest global scan route and the optimal yaw angle corresponding to each target scan waypoint.

[0115] In this embodiment, the flight cost matrix is ​​converted into the standard format of the ATSP problem, and a heuristic TSP solver (such as the LKH algorithm) is invoked to find the optimal or near-optimal access order. The solver outputs an ordered access sequence of target scan waypoints, i.e., the global route. The aircraft visits each target scan waypoint sequentially according to the global route. After reaching each target scan waypoint and completing the scan, the scan target classification and clustering are updated based on the newly acquired scan data, waypoints are regenerated, and the global route is optimized. This online replanning mechanism enables the system to adapt to the gradual revelation of the environment and continuously optimize the scanning strategy.

[0116] Optionally, controlling the aircraft to scan the target environment based on the latest global scan route includes: determining whether the aircraft's current scan waypoint meets a preset stop condition; if so, controlling the aircraft to scan the target environment according to the latest global scan route.

[0117] The preset stop conditions are that the currently executing waypoint has been completed, the currently executing waypoint is unreachable, or the flight cost has significantly deteriorated. In a preferred embodiment, the currently executing waypoint remains unchanged if the current flight cost does not exceed a preset multiple (e.g., 1.5 times) of the previous round's flight cost. In another embodiment, the system can also decide whether to switch targets by monitoring the effectiveness status of the scanned target cluster (e.g., whether the cluster size is below the lower limit, whether the waypoint has been reached, etc.).

[0118] In one alternative implementation, refer to Figure 7 The diagram shown is a schematic representation of the observation quality provided in an embodiment of this application; refer to... Figure 8 The diagram shown is a schematic of the multi-angle coverage determination mechanism provided in this application embodiment (taking four angle intervals as an example, with a coverage threshold N_cov=3: requiring observations from at least three different angle intervals before being marked as sufficient observation); refer to Figure 9 The diagram shown illustrates the gap points and field of view boundaries provided in this embodiment. Red dots represent gap points (depth abrupt changes), the area between the near and far surfaces represents "potentially unobserved areas behind occlusion," and the purple ray direction indicates the direction without echoes. (Refer to...) Figure 10 The image shown is a schematic diagram of a depth map provided in an embodiment of this application; refer to... Figure 11The diagram shown is a comparison between normal vector-driven clustering and pure distance clustering provided in the embodiments of this application. When based on distance-driven clustering, the frontiers in different directions are mixed and clustered, with the main normal vector being a compromise direction and the waypoint orientation being unclear. When based on normal vector-driven clustering, the frontiers in different directions are correctly separated, each cluster direction is clear, and the waypoint orientation is accurate.

[0119] Reference Figure 12 The diagram shown is a schematic diagram of the generation of bar-shaped sampling waypoints provided in an embodiment of this application.

[0120] Reference Figure 13 The diagram illustrates the evaluation dimensions of candidate scanning waypoints provided in this application embodiment. For (a) the visibility component dimension, waypoints with the largest number of visible frontiers are selected, with green representing visible frontiers and red representing occluded frontiers; for (b) the distance component, the closer to the aircraft, the larger the component value, encouraging the selection of nearby waypoints; for (c) the directional consistency component, choosing waypoints close to the aircraft's current speed direction is encouraged to reduce frequent turns. In this application embodiment, the candidate waypoint with the highest comprehensive score is selected as the scanning waypoint for clustering, taking into account the weighted sum of the above components.

[0121] Reference Figure 14 The diagram shown is a schematic representation of the topology network provided in an embodiment of this application; refer to... Figure 15 The diagram shown is a schematic diagram of TSP global route optimization provided in an embodiment of this application.

[0122] Example 1: Autonomous Scanning and 3D Reconstruction of an Indoor Factory Environment. This example assumes the scanned scene is an indoor factory workshop with an area of ​​approximately 50m × 30m × 8m. The unmanned aerial vehicle (UAV) is equipped with a solid-state LiDAR, and the onboard computing platform includes an ARM embedded processor (responsible for perception and decision-making) and a flight control computer (responsible for motion execution). System parameter settings: voxel size 0.4m, trust distance 6.0m, forced trust distance 2.0m, gap detection threshold 0.8. Cluster radius 1.0m, normal vector consistency threshold 0.15. Cylindrical sampling: 3 layers with height [1.0, 3.5]m, 3 layers with radius [1.5, 4.0]m, 8 azimuth directions, and a total of 72 candidate waypoints. Scoring weights: distance weight 0.3, directional consistency weight 0.2.

[0123] Processing flow: (1) The user triggers the exploration mission. The aircraft takes off and the sensors begin scanning.

[0124] (2) The point cloud map (i.e. the observation quality map) is updated incrementally for each frame of point cloud, and at the same time, it is projected onto the depth map to perform observation quality classification.

[0125] (3) Perform normal vector-driven clustering on the front edge to generate front edge clusters.

[0126] (4) Perform bar sampling and multi-evaluation dimension comprehensive scoring on each frontier cluster to generate target scanning waypoints.

[0127] (5) Insert the target scanning waypoints into the topology network (this embodiment uses a skeleton graph based on safety bubbles) and calculate the flight cost between waypoints.

[0128] (6) Construct the flight cost matrix, correct it with Floyd-Warshall, and then call the LKH solver to solve the ATSP and output the global route.

[0129] (7) The aircraft visits waypoints sequentially along the route (in this embodiment, Bubble A is used). Path search and MINCO trajectory optimization for flight execution.

[0130] (8) After reaching the target waypoint, repeat steps (2) to (7) until there are no more reachable frontiers, and the exploration is complete.

[0131] In this embodiment, exploring a 50m×30m×8m factory workshop takes approximately 15-25 minutes. The point cloud map uses incremental kd-tree storage, with memory usage linearly related to the actual number of observation points, approximately 30-80MB, significantly lower than the memory overhead of voxel raster representation for the same coverage area. The collected point cloud can be used for 3D Gaussian splashing or mesh reconstruction to generate a digital twin model of the factory.

[0132] Example 2: Large-scale scanning of an outdoor building complex. This example assumes the scanning scene is an outdoor building complex, covering an area of ​​approximately 200m × 200m. Due to the large scale of the environment, traditional voxel raster methods have insufficient memory. This example operates directly on the point cloud, with memory usage only linearly related to the actual number of observation points, approximately 100~300MB, and can run normally on embedded platforms. The processing flow is the same as in Example 1.

[0133] Example 3: Power Substation Inspection. This example is applied to the autonomous inspection of power substations. The aircraft first uses the method of this application to acquire a complete 3D point cloud of the substation, and then performs 3D reconstruction to generate a digital twin model. In this scenario, multi-level observation quality classification is particularly valuable—substations contain a large number of small structures (insulators, conductors, hardware, etc.), and the poor viewing angle front can identify observation blind spots caused by obstruction, guiding the aircraft to fully scan these small structures from multiple angles, improving the integrity of the reconstruction.

[0134] Example 4: Security Monitoring Scenario. This example is applied to security monitoring in a large park. The aircraft performs a full-coverage scan of the park, and the acquired 3D point cloud is used to construct a security situation map. Global route optimization ensures that the aircraft completes the full-coverage scan in the shortest possible time. An online replanning mechanism enables the aircraft to continuously update the environmental map during patrol and promptly detect environmental changes.

[0135] This concludes the description of the embodiments. The embodiments given in this section are specific illustrations of the method and are used to help understand the implementation of this application. However, the scope of protection of this application is not limited thereto. For example, point cloud map management can adopt incremental kd-trees, octrees, or other dynamic spatial index structures; path search can adopt A... RRT Alternatively, other path planning algorithms can be used; trajectory execution can employ MINCO, B-spline, or other trajectory optimization methods; TSP solving can utilize LKH, genetic algorithms, simulated annealing, or other heuristic / exact solvers. Various specific algorithm implementations can be replaced or extended within the basic technical framework of this invention.

[0136] In summary, compared with the problems existing in the prior art, the embodiments of this application have the following advantages and positive effects: (1) Multi-level observation quality classification to accurately identify scanning requirements.

[0137] This application's embodiments subdivide surface voxels containing scanned point clouds into three observation states, distinguishing between data loss caused by insufficient distance and poor viewing angle. This fine classification allows subsequent waypoint generation to selectively choose observation strategies—generating closer waypoints for areas with insufficient distance and generating waypoints with different angles for areas with poor viewing angles, thus improving the quality of scanned data and the completeness of 3D reconstruction. The normal vector verification mechanism effectively suppresses noise interference with front-end detection.

[0138] (1.1) Multi-angle coverage determination improves the quality of 3D reconstruction.

[0139] The multi-angle coverage determination mechanism in this application requires that the same surface be observed from multiple different directions before it is marked as sufficiently observed. This ensures that the point cloud data required for 3D reconstruction has multi-angle coverage characteristics, enabling the reconstruction algorithm to obtain more uniform point cloud density, more accurate normal vector estimation, and more complete surface coverage, significantly improving the accuracy and integrity of the 3D model. The storage overhead of this mechanism is extremely small (only 1-2 bytes added per voxel) and does not affect the real-time performance of the system.

[0140] (2) Normal vector-driven clustering preserves the directionality of scanning requirements.

[0141] This application's embodiments introduce a normal vector consistency constraint in clustering, ensuring the correct separation of fronts from different scanning directions. Each cluster corresponds to a scanning task with a clearly defined direction, and the cluster's principal normal vector provides clear observation direction guidance for waypoint generation, avoiding the waypoint quality degradation caused by the mixing of fronts from different directions in traditional methods.

[0142] (3) Multi-dimensional evaluation of candidate scan waypoints to balance coverage and flight efficiency.

[0143] The scoring function in this application comprehensively considers three factors: visibility, flight distance, and consistency of movement direction. The visibility component ensures scan coverage, the distance component reduces long-distance flight, and the consistency of movement direction component reduces frequent turning. The weights of the three components can be flexibly adjusted according to specific application scenarios to adapt to different efficiency-coverage trade-offs.

[0144] (4) Cost calculation based on topological road network accurately reflects the actual flight cost.

[0145] This application's embodiments calculate the actual flight cost between waypoints using a topological road network, replacing the Euclidean distance approximation. In scenarios where large obstacles need to be detoured, the cost based on the topological road network is highly consistent with the actual flight cost, while the Euclidean distance may significantly underestimate the actual cost. Accurate cost estimation makes the results of global route optimization more reasonable.

[0146] (5) ATSP global route optimization reduces redundant flight.

[0147] This application's embodiments model global route planning as an ATSP problem and utilize a heuristic solver to optimize the waypoint visit order. Compared to greedy strategies, ATSP solutions can optimize routes from a global perspective, significantly reducing repeated shuttles between scanned and unscanned areas. The return cost adjustment mechanism and triangle inequality correction in the flight cost matrix further improve route quality. The target waypoint persistence mechanism avoids flight jitter caused by frequent target switching.

[0148] (6) Online replanning mechanism to adapt to the gradual revelation of the environment.

[0149] In this embodiment, after the aircraft reaches each waypoint, it re-executes the scanning target classification, clustering, waypoint generation, and route optimization based on the newly acquired scanning data. This online replanning mechanism enables the system to continuously adapt to the gradual revelation of the environment, constantly optimizing the scanning strategy during the exploration process until all discovered surfaces are fully observable.

[0150] (7) A complete autonomous scanning process that supports full-link automation of 3D reconstruction.

[0151] This application establishes a complete autonomous scanning planning process from target identification to global route output. The entire process requires no manual intervention; it can be executed automatically simply by providing the scanning area boundary and aircraft parameters. The planned scanning route can directly drive the aircraft to perform data acquisition tasks, and the acquired point cloud data can be used for subsequent 3D reconstruction (such as 3D Gaussian splashing, NeRF, Mesh reconstruction, etc.), achieving end-to-end automation from data acquisition planning to 3D model construction.

[0152] Based on the same inventive concept, this application also provides an environmental scanning device for three-dimensional reconstruction, which corresponds to the environmental scanning method for three-dimensional reconstruction. Since the principle of the device in this application is similar to the environmental scanning method for three-dimensional reconstruction described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0153] Reference Figure 16 The diagram shown is a schematic of an environmental scanning device for 3D reconstruction provided in an embodiment of this application. The device includes: The acquisition module 1601 is used to acquire the current effective frontier determined by the depth map within the current planning period and the reasons for insufficient observation corresponding to the current effective frontier, as well as the first historical effective frontier in the first historical frontier cluster in the target cluster list that overlaps with the current effective frontier; the depth map contains the first distance between each scanned pixel in the target environment and the nearest obstacle on the corresponding scan ray; the frontier refers to the voxel that needs to be scanned by the aircraft; Clustering module 1602 is used to cluster the current effective frontier and the first historical effective frontier to obtain current frontier clusters that satisfy the normal vector consistency condition and spatial distance constraint; the normal vector consistency condition is used to constrain the angle between the normal vectors of two frontiers in the same frontier cluster; The generation module 1603 is used to generate the target scanning waypoints with the highest comprehensive scores and the optimal yaw angles corresponding to each target scanning waypoint for each current frontier cluster and the second historical frontier cluster, based on the insufficient observation reasons corresponding to the current effective frontier in the current frontier cluster and the insufficient observation reasons corresponding to the second historical effective frontier in the target cluster list. This is used to control the aircraft to scan the target environment based on the target scanning waypoints and the corresponding optimal yaw angles. The comprehensive score corresponding to the target scanning waypoint is obtained by comprehensively scoring the target scanning waypoint from the visibility dimension, distance dimension, and orientation consistency dimension.

[0154] like Figure 17As shown in the embodiment of this application, an electronic device 1700 includes a processor 1701, a memory 1702, and a bus. The memory 1702 stores machine-readable instructions executable by the processor 1701. When the electronic device is running, the processor 1701 communicates with the memory 1702 via the bus. The processor 1701 executes the machine-readable instructions to perform the steps of the environmental scanning method for three-dimensional reconstruction described above.

[0155] Specifically, the memory 1702 and processor 1701 mentioned above can be general-purpose memory and processor, without any specific limitations. When the processor 1701 runs the computer program stored in the memory 1702, it can execute the above-mentioned environmental scanning method for three-dimensional reconstruction.

[0156] Corresponding to the above-described environmental scanning method for 3D reconstruction, this application embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the above-described environmental scanning method for 3D reconstruction.

[0157] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0158] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0159] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0160] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the environmental scanning method for 3D reconstruction described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, external hard drives, ROM, RAM, magnetic disks, or optical disks.

[0161] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An environmental scanning method for 3D reconstruction, characterized in that, The method includes: The system obtains the current effective frontier determined by the depth map within the current planning period, the reasons for insufficient observation corresponding to the current effective frontier, and the first historical effective frontier within the first historical frontier cluster in the target cluster list that overlaps with the current effective frontier; the depth map contains the first distance between each scanned pixel in the target environment and the nearest obstacle on the corresponding scan ray; the frontier refers to the voxel that needs to be scanned by the aircraft; The step of obtaining the current effective frontier determined by the depth map within the current planning period and the corresponding reasons for insufficient observation of the current effective frontier includes: determining the field-of-view boundary points and gap points of the aircraft in the target environment based on the first distance between each scanned pixel point in the depth map and the nearest obstacle on the corresponding scan ray within the current planning period; the gap points refer to pixels used to indicate occlusion boundaries or scan blind spots of the aircraft; and determining the current effective frontier in the voxel set to be updated and the corresponding reasons for insufficient observation of each current effective frontier based on the gap points and the field-of-view boundary points. Clustering is performed on the current effective frontier and the first historical effective frontier to obtain current frontier clusters that satisfy the normal vector consistency condition and spatial distance constraint; the normal vector consistency condition is used to constrain the angle between the normal vectors of two frontiers in the same frontier cluster; Based on the reasons for insufficient observation corresponding to the current effective frontier in the current frontier cluster and the reasons for insufficient observation corresponding to the second historical effective frontier in the second historical frontier cluster in the target cluster list, the target scanning waypoints with the highest comprehensive scores and the optimal yaw angles corresponding to each target scanning waypoint are generated for each current frontier cluster and the second historical frontier cluster. The aircraft is then controlled to scan the target environment based on the target scanning waypoints and the corresponding optimal yaw angles. The comprehensive score corresponding to the target scanning waypoint is obtained by comprehensively scoring the target scanning waypoint from the dimensions of visibility, distance, and orientation consistency.

2. The environmental scanning method for three-dimensional reconstruction according to claim 1, characterized in that, The following steps are used to determine the field-of-view boundary points and gap points of the aircraft in the target environment: Based on the ratio of the first distance between each scanned pixel and its corresponding neighboring pixel in the depth map and the first distance between each scanned pixel and its corresponding neighboring pixel on the corresponding scanning ray and the nearest obstacle, gap points are determined from all scanned pixels. Based on the comparison between the first distance corresponding to each scanned pixel and the preset effective distance threshold, each scanned pixel is divided into an effective region and an invalid region; pixels in the effective region have real obstacle echoes on the corresponding scanning ray; The boundary point between the effective region and the invalid region is defined as the field of view boundary point.

3. The environmental scanning method for three-dimensional reconstruction according to claim 1, characterized in that, The step of determining the current effective frontier in the voxel set to be updated and the reasons for insufficient observation corresponding to each current effective frontier based on the gap point and the field of view boundary point includes: For each first candidate frontier in the set of voxels to be updated that is not marked as sufficiently observed, if the first candidate frontier does not meet the close-range strong trust determination condition and the conventional sufficient observation determination condition, then the first candidate frontier is determined as the second candidate frontier; the close-range strong trust determination condition is that the second distance from the voxel to the scanning sensor in the aircraft is less than a preset forced trust distance threshold and is not a field of view boundary point; the conventional sufficient observation determination condition is that the second distance from the voxel to the scanning sensor in the aircraft is less than a preset trust distance and is not a gap point or a field of view boundary point; For each second candidate front, if the voxel observation quality of the second candidate front at a preset pixel distance in all preset directions of the corresponding normal vector is not sufficiently observed, then the second candidate front is determined as the third candidate front. For each third candidate frontier, if there is at least one voxel in the neighborhood voxel of the third candidate frontier that is in a sufficiently observed state, then the third candidate frontier is determined as the current valid frontier; based on the numerical comparison between the second distance corresponding to the current valid frontier and the preset trust distance, the reason for insufficient observation corresponding to the current valid frontier is determined.

4. The environmental scanning method for three-dimensional reconstruction according to claim 3, characterized in that, The method further includes: If the first candidate front meets the close-range strong trust determination condition or the conventional sufficient observation determination condition, then determine whether the number of observed angle intervals of the first candidate front is greater than or equal to a preset coverage threshold. If the number of observed angles of the first candidate frontier is less than the preset coverage threshold, then the first candidate frontier is determined as the current effective frontier.

5. The environmental scanning method for three-dimensional reconstruction according to claim 1, characterized in that, Generate the corresponding target scan waypoint and the corresponding optimal yaw angle for any leading edge cluster according to the following steps: Based on the reasons for insufficient observation corresponding to the effective front in the current front cluster, candidate scan waypoints and the optimal yaw angles corresponding to each candidate scan waypoint are generated for the current front cluster. Each candidate waypoint is comprehensively scored across multiple evaluation dimensions to obtain a comprehensive score. These evaluation dimensions include visibility, distance, and orientation consistency. Visibility measures the potential exploration value of a candidate waypoint in an unknown environment. Distance measures the spatial distance cost between the candidate waypoint and the aircraft's current position. Orientation consistency measures the consistency between the direction from the aircraft's current position to the candidate waypoint and the aircraft's current direction of motion. The candidate scan waypoint with the highest overall score is determined as the target scan waypoint corresponding to the current leading-edge cluster.

6. The environmental scanning method for three-dimensional reconstruction according to claim 1, characterized in that, The step of controlling the aircraft to scan the target environment based on the target waypoints and the corresponding optimal yaw angle includes: Based on the topological road network, a flight cost matrix is ​​constructed; the topological road network contains multiple environmental nodes and edges between reachable pairs of environmental nodes; the weight value of each edge corresponds to the actual path length between the corresponding pairs of environmental nodes; the environmental nodes are obtained by dividing the scanned environmental regions in the target environment; the flight cost matrix contains the flight cost between the scanned waypoints of the target. The flight order between all target scan waypoints is optimized based on the flight cost matrix to obtain the latest global scan route. The aircraft is then controlled to scan the target environment based on the latest global scan route and the optimal yaw angle corresponding to each target scan waypoint.

7. An environmental scanning device for three-dimensional reconstruction, characterized in that, The device includes: The acquisition module is used to acquire the current effective frontier determined by the depth map within the current planning period and the reasons for insufficient observation corresponding to the current effective frontier, as well as the first historical effective frontier in the first historical frontier cluster in the target cluster list that overlaps with the current effective frontier; the depth map contains the first distance between each scanned pixel in the target environment and the nearest obstacle on the corresponding scan ray; the frontier refers to the voxel that needs to be scanned by the aircraft; Specifically, the acquisition module is used to determine, within the current planning cycle, the field-of-view boundary points and gap points of the aircraft in the target environment based on the first distance between each scanned pixel in the depth map and the nearest obstacle on the corresponding scan ray; the gap points refer to pixels used to indicate occlusion boundaries or the scanning blind spots of the aircraft; and based on the gap points and the field-of-view boundary points, determine the current effective frontier in the voxel set to be updated and the reasons for insufficient observation corresponding to each current effective frontier; The clustering module is used to cluster the current effective frontier and the first historical effective frontier to obtain current frontier clusters that satisfy the normal vector consistency condition and spatial distance constraint; the normal vector consistency condition is used to constrain the angle between the normal vectors of two frontiers in the same frontier cluster; The generation module is used to generate target scanning waypoints with the highest comprehensive scores and corresponding optimal yaw angles for each current frontier cluster and the second historical frontier cluster, based on the insufficient observation reasons corresponding to the current effective frontier in the current frontier cluster and the insufficient observation reasons corresponding to the second historical effective frontier in the target cluster list. This allows the aircraft to scan the target environment based on the target scanning waypoints and their corresponding optimal yaw angles. The comprehensive score corresponding to the target scanning waypoint is obtained by comprehensively scoring the target scanning waypoint from the dimensions of visibility, distance, and orientation consistency.

8. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the environmental scanning method for three-dimensional reconstruction as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the environmental scanning method for three-dimensional reconstruction as described in any one of claims 1 to 6.

Images