Method, device, medium and product for planning a scanning path on an aircraft surface

Abstract

The application discloses a kind of scanning path planning methods, equipment, medium and product for aircraft upper surface, which comprises: based on curvature, normal vector and point cloud density joint detection algorithm, the overall point cloud data of the aircraft to be scanned is divided into data, to obtain each target point cloud partition corresponding to each target functional partition in the upper surface of the aircraft, after rectification treatment along the direction of normal vector, generate initial candidate viewpoint set and screen out safe viewpoint set.According to the visibility matrix constructed by the sensor model, the optimal viewpoint set is screened out from the safe viewpoint set using the greedy algorithm, and at least one optimal access sequence is solved using a hybrid solution strategy.The scanning path corresponding to each optimal access sequence is constructed and respectively issued to the unmanned aerial vehicle.The technical scheme can improve the detection efficiency and safety of the upper surface of the aircraft, and enhance the degree of automation and detection stability of the system.

Description

Technical Field

[0001] This invention relates to the field of autonomous inspection and path planning technology for unmanned aerial vehicles (UAVs), and in particular to a scanning path planning method, device, medium, and product for the upper surface of an aircraft. Background Technology

[0002] The upper surface of an aircraft is a high-risk area for external damage such as lightning strikes and hail impacts. Traditional manual inspection suffers from low efficiency, high missed detection rates, and high operational risks. In recent years, highly mobile and omnidirectionally accessible unmanned aerial vehicles (UAVs) have been gradually applied to automated inspection of aircraft surfaces. However, aircraft surfaces are large-scale, highly curved, complex free-form surfaces with significant self-occlusion and non-convexity characteristics. Planning a detection path that is fully covered, highly efficient, and collision-free still faces many challenges.

[0003] Existing coverage path planning methods include: random sampling method, which achieves coverage by randomly generating viewpoints within a safe space, but suffers from high viewpoint redundancy and circuitous paths; traditional slicing method, which generates reciprocating scanning paths along the fuselage axis with fixed step sizes, is difficult to adapt to complex curved surfaces and is prone to generating detection blind spots; and genetic algorithms, ant colony algorithms, and other algorithms based on sampling or swarm intelligence plan paths in an intelligent optimization manner, but suffer from high computational complexity and difficulty in ensuring real-time performance and path smoothness.

[0004] Existing technologies do not fully consider the spatial interference relationships between various aircraft components. Locally secure viewpoints are prone to collisions with other components in the global space. Furthermore, the pursuit of full coverage often results in redundant viewpoints and a surge in path length, making it difficult to achieve efficient detection under limited flight range. It is impossible to balance coverage integrity, path efficiency, and global safety. Summary of the Invention

[0005] This invention provides a scanning path planning method, device, medium, and product for the upper surface of an aircraft, to solve the problems of global space collision risk, viewpoint redundancy, and excessively long paths during the flight of a UAV to inspect the surface of an aircraft.

[0006] According to one aspect of the present invention, a scanning path planning method for the upper surface of an aircraft is provided, comprising: Based on the joint detection algorithm of curvature, normal vector and point cloud density, the overall point cloud data of the aircraft to be scanned is divided into data segments to obtain each target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft. After performing consistency correction processing on the normal vector direction of each point cloud point in each target point cloud partition, an initial candidate viewpoint set is generated along the corrected normal vector direction. Based on the global Euclidean distance field that matches the overall point cloud data, a safe viewpoint set is selected from the initial candidate viewpoint set. Based on the visibility matrix constructed by the sensor model, a greedy algorithm based on a preset coverage threshold condition is used to select the optimal set of viewpoints from the set of safe viewpoints. Based on the asymmetric traveling salesman problem modeled by the viewpoint access order, a hybrid solution strategy is used to solve for at least one optimal access sequence that matches the optimal viewpoint set; A scanning path corresponding to each of the optimal access sequences is constructed, and each scanning path is sent to the UAV so that each UAV can scan and photograph the upper surface of the aircraft to be scanned.

[0007] According to another aspect of the present invention, a scanning path planning device for an aircraft upper surface is provided, comprising: The point cloud partitioning module is used to divide the overall point cloud data of the aircraft to be scanned based on the joint detection algorithm of curvature, normal vector and point cloud density, and obtain the target point cloud partitions corresponding to each target functional partition on the upper surface of the aircraft. The safe viewpoint set acquisition module is used to perform consistency correction processing on the normal vector direction of each point cloud point in each target point cloud partition, generate an initial candidate viewpoint set along the corrected normal vector direction, and filter out the safe viewpoint set from the initial candidate viewpoint set based on the global Euclidean distance field that matches the overall point cloud data. The optimal viewpoint set acquisition module is used to select the optimal viewpoint set from the safe viewpoint set based on the visibility matrix constructed by the sensor model and a greedy algorithm based on a preset coverage threshold condition. The optimal access sequence acquisition module is used to solve at least one optimal access sequence that matches the optimal viewpoint set based on the asymmetric traveling salesman problem modeled by the viewpoint access order, using a hybrid solution strategy. The scanning path generation module is used to construct scanning paths corresponding to each of the optimal access sequences, and to send each scanning path to the UAV so that each UAV can scan and photograph the upper surface of the aircraft to be scanned.

[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the scanning path planning method for the upper surface of an aircraft as described in any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the scanning path planning method for an aircraft upper surface as described in any embodiment of the present invention.

[0010] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the method as described in any embodiment of the present invention.

[0011] This invention, through the aforementioned technical solution, utilizes a joint detection algorithm based on curvature, normal vector, and point cloud density to divide the overall point cloud data of the aircraft to be scanned into target point cloud partitions corresponding to various target functional areas on the aircraft's upper surface. An initial candidate viewpoint set is generated along the normal vector direction of each point cloud point in each target point cloud partition after correction processing. Based on a global Euclidean distance field matching the overall point cloud data of the aircraft to be scanned, a safe viewpoint set can be selected from the initial candidate viewpoint set according to a preset safe distance. A greedy algorithm based on a preset coverage threshold condition is used in the visibility matrix constructed by the sensor model to select the optimal viewpoint set from the safe viewpoint set. An asymmetric traveling salesman problem with optimal sequence is modeled for the optimal viewpoint set. A hybrid solution strategy is used to obtain multiple optimal access sequences matching the optimal viewpoint set. Corresponding scanning paths can be constructed based on these multiple optimal access sequences and sent to the UAV to control the UAV to scan and photograph the upper surface of the aircraft. The above technical solution can obtain the optimal scanning path by solving the overall point cloud data of the aircraft's upper surface, effectively eliminating redundant observations, avoiding detours in the detection path, improving detection efficiency and operational safety, and enhancing the system's automation and detection stability.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a scanning path planning method for the upper surface of an aircraft according to Embodiment 1 of the present invention; Figure 2This is a flowchart of a scanning path planning method for the upper surface of an aircraft according to Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of overall point cloud data of an aircraft applicable to an embodiment of the present invention; Figure 4 This is a schematic diagram of a target functional partition on the upper surface of an aircraft to which an embodiment of the present invention applies; Figure 5 This is a schematic diagram of the initial candidate viewpoint set for the upper surface of an aircraft to which this embodiment of the invention applies; Figure 6 This is a schematic diagram of the safety viewpoint set for the upper surface of an aircraft to which this invention applies; Figure 7 This is a schematic diagram of the optimal viewpoint set for the upper surface of an aircraft to which this embodiment of the invention applies; Figure 8 This is a schematic diagram of the UAV scanning path applicable to the embodiments of the present invention; Figure 9 This is a schematic diagram of a scanning path planning device for the upper surface of an aircraft according to Embodiment 3 of the present invention; Figure 10 This is a schematic diagram of the structure of an electronic device that implements the scanning path planning method for the upper surface of an aircraft according to an embodiment of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] Example 1 Figure 1 This is a flowchart of a scanning path planning method for an aircraft upper surface provided in Embodiment 1 of the present invention. This embodiment is applicable to the automated inspection of large-scale, high-curvature, complex freeform surfaces of aircraft using UAVs, conducting detection path planning operations that balance coverage integrity, path efficiency, and global safety. This method can be executed by a scanning path planning device for an aircraft upper surface, which can be implemented in hardware and / or software and is generally configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes: S110. Based on the joint detection algorithm of curvature, normal vector and point cloud density, the overall point cloud data of the aircraft to be scanned is divided into data to obtain each target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft.

[0018] Curvature refers to a geometric quantity that measures the degree of bending of a curve or surface; a larger value indicates a greater degree of bending. A normal vector is a non-zero vector in geometric space that is perpendicular to a tangent line at a point on a curve or a tangent plane at a point on a surface; it is a core geometric quantity describing the spatial orientation of curves and surfaces at corresponding positions. Point cloud data refers to a dataset consisting of the coordinates and related attribute information of a large number of discrete points in three-dimensional space, acquired through technologies such as laser scanning and visual imaging; it is a digital representation of the geometric shape of a real-world object's surface. Point cloud density refers to the number of points in a unit spatial area of ​​point cloud data; it is a core indicator for measuring the precision of point cloud data in representing the geometric shape of a target surface. Target functional zones refer to the areas on the upper surface of an aircraft divided according to the characteristics of different functional components constituting the aircraft. Target point cloud zones can be point cloud data used to describe the geometric shape of the upper surface of the aircraft in each target functional zone.

[0019] Understandably, when scanning and inspecting the upper surface of an aircraft, different target functional zones can be defined based on the characteristics of the various functional components that make up the entire aircraft. For example, based on the surface distribution characteristics of the fuselage, wings, and tail, the upper surface of the aircraft can be divided into seven target functional zones: the upper surface of the fuselage, the upper surface of the left wing, the upper surface of the right wing, the upper surface of the left horizontal tail, the upper surface of the right horizontal tail, the left side of the vertical tail, and the right side of the vertical tail. By scanning each target functional zone of the upper surface, overall point cloud data describing the geometry of the aircraft's upper surface can be obtained. Based on the curvature, normal vector, and point cloud density of the upper surface, a joint detection algorithm can be constructed. Based on the overall point cloud data of the upper surface and the constructed joint detection algorithm, the target point cloud zones corresponding to each target functional zone of the upper surface can then be obtained.

[0020] S120. After performing consistency correction processing on the normal vector direction of each point cloud point in each target point cloud partition, an initial candidate viewpoint set is generated along the corrected normal vector direction. Based on the global Euclidean distance field that matches the overall point cloud data, a safe viewpoint set is selected from the initial candidate viewpoint set.

[0021] The initial candidate viewpoint set can be defined as the set of viewpoints formed by offsetting each point cloud point from a preset safety threshold distance based on the normal vector direction of the point cloud data of the aircraft's upper surface. The global Euclidean distance field can be defined as a numerical field in three-dimensional global space that assigns the nearest straight-line distance to the target surface to each location point, accurately representing the distance and relative position between any point in space and the target surface. The safe viewpoint set can be defined as the set of viewpoints obtained by removing viewpoints corresponding to point cloud points in each target functional zone from the initial candidate viewpoint set whose straight-line distance to viewpoints corresponding to point cloud points in other target functional zones is less than the safety threshold distance.

[0022] It is understandable that the normal vectors of individual points in the collected overall aircraft point cloud data may exhibit inconsistent orientations due to acquisition or calculation errors. To ensure consistency in the normal vector directions of all points in the overall point cloud data of the aircraft's upper surface and to eliminate these errors, a consistency correction process can be performed on the normal vector directions of each point in each target point cloud partition, ensuring that the normal vector directions of all points point outward from the upper surface of the aircraft. By offsetting the normal vector directions of each point in the corrected overall point cloud data of the aircraft's upper surface by a preset safety threshold distance, corresponding viewpoints are formed, thus obtaining the initial candidate viewpoint set for each target functional partition. In the global Euclidean distance field constructed based on the overall point cloud data of the aircraft's upper surface, the viewpoints corresponding to the point points in each target functional partition of the initial candidate viewpoint set are traversed. Points whose straight-line distance to the viewpoints corresponding to point points in other target functional partitions is less than the safety threshold distance are filtered out and removed. All viewpoints retained in the initial candidate viewpoint set are considered the safe viewpoint set.

[0023] S130. Based on the visibility matrix constructed by the sensor model, a greedy algorithm based on a preset coverage threshold condition is used to select the optimal viewpoint set from the safe viewpoint set.

[0024] The visibility matrix can be a two-dimensional matrix constructed based on a sensor model, representing whether effective visual perception can be achieved between a viewpoint and corresponding point cloud points in the overall point cloud data of the aircraft's upper surface. The matrix elements directly reflect the visibility status of the corresponding viewpoint to the point cloud points. The optimal viewpoint set can be a set of viewpoints that meet the preset viewpoint and corresponding point cloud point coverage threshold conditions, selected stepwise from the safe viewpoint set using a greedy algorithm.

[0025] Understandably, in order to more accurately measure the relationship between each viewpoint in the safety viewpoint set and the corresponding point cloud points in the overall point cloud data of the aircraft's upper surface, a visibility matrix based on the sensor model of the collected point cloud data can be constructed. According to the detection requirements, a coverage threshold condition for the viewpoints in the safety viewpoint set to the point cloud points in the overall point cloud data of the aircraft's upper surface can be preset. A greedy algorithm based on the preset coverage threshold condition can be adopted. Starting from any viewpoint in the safety viewpoint set, candidate viewpoints that can maximize the improvement of the current overall viewpoint coverage can be selected step by step and included in the effective viewpoint set until the proportion of the point cloud point range covered by the included viewpoints to the point cloud points in the overall point cloud data of the aircraft's upper surface reaches the preset coverage threshold. The final effective viewpoint set can be used as the optimal viewpoint set.

[0026] S140. Based on the asymmetric traveling salesman problem modeled by the viewpoint access order, a hybrid solution strategy is adopted to find at least one optimal access sequence that matches the optimal viewpoint set.

[0027] The asymmetric traveling salesman problem can be defined as the problem of finding the minimum total cost path that traverses all viewpoints from the starting viewpoint and returns to the starting viewpoint exactly once, given that the round-trip path costs between viewpoints in the optimal viewpoint set are different. The optimal visit sequence can be defined as the order of visits that minimizes the total movement path and time when traversing all viewpoints in the optimal viewpoint set.

[0028] Understandably, after determining the optimal viewpoint set, in order to minimize the time taken for the UAV to inspect the aircraft's upper surface at each viewpoint in the optimal viewpoint set, it is necessary to calculate the optimal access sequence for each viewpoint in the optimal viewpoint set. Therefore, an asymmetric traveling salesman problem can be modeled based on the viewpoint access order in the optimal viewpoint set, and the constructed asymmetric traveling salesman problem can be solved using a hybrid solution strategy. This will yield at least one optimal access sequence that matches the optimal viewpoint set, thus determining the order in which the UAV inspects the aircraft's upper surface at each viewpoint in the optimal viewpoint set.

[0029] S150. Construct scanning paths corresponding to each optimal access sequence, and send each scanning path to the UAV so that each UAV can scan and photograph the upper surface of the aircraft to be scanned.

[0030] Understandably, based on the optimal access sequence of each viewpoint in a set of multiple optimal viewpoints, multiple optimal scanning paths can be determined for the UAV to scan and photograph the upper surface of the aircraft. Each scanning path is then issued to the UAV that executes the scanning and photographing task along that path, so that the UAV scans and photographs each target functional area on the upper surface of the aircraft according to the received scanning path command, thereby improving the efficiency of the UAV in detecting each target functional area on the upper surface of the aircraft from each viewpoint in the set of optimal viewpoints.

[0031] This invention, through the aforementioned technical solution, utilizes a joint detection algorithm based on curvature, normal vector, and point cloud density to divide the overall point cloud data of the aircraft to be scanned into target point cloud partitions corresponding to various target functional areas on the aircraft's upper surface. An initial candidate viewpoint set is generated along the normal vector direction of each point cloud point in each target point cloud partition after correction processing. Based on a global Euclidean distance field matching the overall point cloud data of the aircraft to be scanned, a safe viewpoint set can be selected from the initial candidate viewpoint set according to a preset safe distance. A greedy algorithm based on a preset coverage threshold condition is used in the visibility matrix constructed by the sensor model to select the optimal viewpoint set from the safe viewpoint set. An asymmetric traveling salesman problem with optimal sequence is modeled for the optimal viewpoint set. A hybrid solution strategy is used to obtain multiple optimal access sequences matching the optimal viewpoint set. Corresponding scanning paths can be constructed based on these multiple optimal access sequences and sent to the UAV to control the UAV to scan and photograph the upper surface of the aircraft. The above technical solution can obtain the optimal scanning path by solving the overall point cloud data of the aircraft's upper surface, effectively eliminating redundant observations, avoiding detours in the detection path, improving detection efficiency and operational safety, and enhancing the system's automation and detection stability.

[0032] Example 2 Figure 2 This is a flowchart of a scanning path planning method for the upper surface of an aircraft, provided in Embodiment 2 of the present invention. This embodiment is a specific embodiment based on the above embodiments, including: a specific method for dividing the overall point cloud data of the upper surface of the aircraft, and a specific method for constructing the optimal scanning path for the UAV. Figure 2 As shown, the method includes: S210. Based on the joint detection algorithm of curvature, normal vector and point cloud density, the overall point cloud data of the aircraft to be scanned is divided into data to obtain the target point cloud partitions corresponding to each target functional partition on the upper surface of the aircraft.

[0033] Optionally, based on a joint detection algorithm of curvature, normal vector, and point cloud density, the overall point cloud data of the aircraft to be scanned is divided into data partitions to obtain target point cloud partitions corresponding to each target functional partition on the upper surface of the aircraft. This can include: Based on the global point cloud density that matches the overall point cloud data of the aircraft to be scanned, the optimal neighborhood size is dynamically calculated, and based on the curvature and normal vector direction of each point cloud point in the overall point cloud data, the curvature threshold and normal vector angle threshold are dynamically calculated. All the overall point cloud data is added to the undivided point set, and the point cloud point with the smallest curvature is selected from the current undivided point set as the current growth seed. Then, the current point cloud partition corresponding to the current growth seed is initialized and constructed. The target search region matching the current growth seed is determined based on the optimal neighborhood size. All point cloud points located within the target search region, with curvature less than the curvature threshold and normal vector angle between them and the current growth seed less than the normal vector angle threshold are selected from the unsegmented point set and added to the current point cloud partition. Return to the operation of selecting the point cloud point with the smallest curvature from the current set of undivided points as the current growth seed, until the processing of all point cloud points in the set of undivided points is completed; After the iteration ends, all generated point cloud partitions are divided into fragmented partitions and standard partitions. The fragmented partitions are then merged into the nearest standard partition to obtain multiple candidate point cloud partitions. Each of the candidate point cloud partitions is input into a pre-built classifier for semantic segmentation to determine the functional partition to which each candidate point cloud partition belongs, and to obtain each target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft; wherein, the target functional partitions include: fuselage, upper surfaces of left and right wings, upper surfaces of left and right horizontal tails, and left and right sides of vertical tail. After determining the target point cloud partition corresponding to the fuselage, the point cloud points located on the lower surface of the fuselage are filtered out.

[0034] The global point cloud density refers to the average number of points per unit area within the overall spatial range of the aircraft surface to be detected, used to characterize the global distribution uniformity and completeness of the point cloud data. The optimal neighborhood refers to the best neighborhood range and corresponding point set that accurately characterizes the curvature and normal vector of the point cloud in the calculation of local point cloud features. The unsegmented point set refers to the set of point cloud points in the overall aircraft point cloud data that have not been filtered by curvature and normal vector angle thresholds. The current growth seed refers to the point cloud point with the smallest curvature selected from the unsegmented point set as the initial point cloud point for filtering points that meet the curvature and normal vector angle threshold conditions. The current point cloud partition refers to the set of point cloud points that meet the curvature and normal vector angle threshold conditions, filtered from the unsegmented point set based on the current growth seed. The target search region refers to the range of point cloud data with the optimal neighborhood size in the unsegmented point set, filtered based on the current growth seed and meeting the curvature and normal vector angle threshold conditions. Fragmented partitions can refer to isolated or fragmented small-scale collections of point cloud points formed after filtering all point cloud points from an undivided point set based on curvature and normal vector angle thresholds. Standard partitions can refer to larger-scale collections of geometrically continuous and characteristically consistent point cloud points formed after filtering all point cloud points from an undivided point set based on curvature and normal vector angle thresholds. Candidate point cloud partitions can refer to multiple point cloud partitions formed after filtering all point cloud points from an undivided point set based on curvature and normal vector angle thresholds, awaiting confirmation of their respective functional partitions.

[0035] Understandably, for example Figure 3 When segmenting the overall point cloud data of the aircraft shown, different numbers of neighboring points need to be selected based on the different global point cloud density distributions to determine the optimal neighborhood size. After determining the global point cloud density normalization index, a mapping relationship between the global point cloud density distribution and the number of neighboring points can be constructed to adapt to the requirements of determining the optimal neighborhood size for different point cloud densities. In other words, when obtaining the global point cloud density, the optimal neighborhood size matching the aircraft to be scanned can be calculated based on the mapping relationship between different global point cloud density ranges and different optimal neighborhood sizes. Alternatively, for simplified calculation, a pre-set optimal neighborhood size can be directly used for point cloud partitioning; this embodiment does not impose any limitations on this approach.

[0036] Calculating the curvature of each point in the overall point cloud data allows us to construct the corresponding neighborhood covariance matrix. For any point in the overall point cloud data... Its k neighborhood point set covariance matrix It can be defined as: ,in, Let the centroid be the neighborhood centroid. For the covariance matrix... After performing eigenvalue decomposition, we can obtain the eigenvectors corresponding to the three eigenvalues. Therefore, point P in the point cloud can be determined. i The corresponding curvature threshold C i for: .

[0037] For point cloud points The degree of geometric abrupt change in its neighborhood can be determined by examining all point cloud points within the neighborhood. With point cloud point P i The sum of the angles between the normal vectors is used for quantization, and the specific quantization formula is as follows: ,in, , Point P i , P j The normal vector, absolute value It can be used to handle situations where the direction of the normal vector may be reversed due to calculation, thereby allowing the calculation of point cloud point P. i Corresponding normal vector angle threshold .

[0038] To simplify calculations, after calculating the curvature threshold and normal angle threshold corresponding to each point cloud point, various normalization or statistical averaging methods can be used to calculate the curvature threshold and normal angle threshold corresponding to the aircraft to be scanned. Alternatively, preset curvature thresholds and normal angle thresholds can be used directly for subsequent point cloud partitioning, etc.

[0039] After determining the optimal neighborhood size, curvature threshold, and normal angle threshold corresponding to the aircraft to be scanned, an unsegmented point set can be constructed based on the overall point cloud data. The point with the smallest curvature in this unsegmented point set is selected as the current growth seed, and the corresponding current point cloud partition is initialized to store the selected point points that meet the requirements. The optimal neighborhood size corresponding to the current growth seed can be used to determine the target search region in the overall point cloud data. All point points in the unsegmented point set located within the target search region with curvature and normal angle less than the aforementioned determined curvature threshold and normal angle less than the aforementioned determined normal angle threshold are selected. All selected point points that meet the requirements are added to the current point cloud partition corresponding to the current growth seed. Following the above logic, the point with the smallest curvature in the unsegmented point set is selected sequentially as the current growth seed, and the operation of selecting point points that meet the curvature threshold and normal angle threshold conditions is performed until the selection process for all point points in the unsegmented point set is completed.

[0040] After the unsegmented point set undergoes point cloud point filtering, there may be unassigned discrete point cloud points (i.e., point cloud partitions containing only one point cloud point) and point cloud partitions with small ranges (or, in other words, point cloud partitions with fewer than a set threshold number of point cloud points). These point cloud points and point cloud partitions can be designated as fragmented partitions, while point cloud partitions with larger ranges can be designated as standard partitions. Furthermore, based on metrics such as iterative nearest neighbor analysis, point cloud points in fragmented partitions are merged into the nearest standard partitions with the most similar geometric features as candidate point cloud partitions. This ensures that the unsegmented point set, after being divided into multiple point cloud partitions, possesses good spatial coherence and physical integrity.

[0041] An aircraft can be divided into the following categories: Figure 4 The multiple target functional partitions shown specifically include: the fuselage, the upper surfaces of the left and right wings, the upper surfaces of the left and right horizontal stabilizers, and the left and right sides of the vertical stabilizer. Therefore, a classifier can be constructed by combining prior knowledge such as the symmetry of the aircraft structure, the relative positions of components, and naming rules. Each candidate point cloud partition is input into the pre-constructed classifier for semantic segmentation to determine the functional partition corresponding to each candidate point cloud partition. Based on the functional partitions corresponding to each candidate point cloud partition, the target point cloud partition corresponding to the target functional partition is selected.

[0042] It is important to note that, except for the fuselage partition which includes both the upper and lower fuselage surfaces, all other functional partitions correspond to the upper surface of the aircraft. Therefore, the fuselage partition needs to be processed separately. Specifically, the geometric centroid of the target fuselage point cloud partition corresponding to the fuselage partition can be calculated, and a horizontal reference plane passing through this centroid can be constructed. By solving the directed distance from each point in the target fuselage point cloud partition to the horizontal reference plane, points on the lower fuselage surface can be filtered and removed, retaining only the point cloud of the upper fuselage surface. This results in a semantically clear and well-defined set of target point clouds representing the upper surface of the aircraft.

[0043] S220. After performing consistency correction processing on the normal vector direction of each point cloud point in each target point cloud partition, an initial candidate viewpoint set is generated along the corrected normal vector direction. Based on the global Euclidean distance field that matches the overall point cloud data, a safe viewpoint set is selected from the initial candidate viewpoint set.

[0044] Optionally, after performing consistent bias correction on the normal vector directions of each point cloud point in each target point cloud partition, an initial candidate viewpoint set is generated along the corrected normal vector directions, including: For each point in each target point cloud partition, the initial normal vector corresponding to each point is calculated using principal component analysis with a preset number of its nearest neighbors. According to the preset spatial guidance direction of the functional area to which each target point cloud partition belongs, the initial normal vectors are uniformly corrected so that the normal vector of each point cloud point after correction points to the external space of the surface of the aircraft to be scanned. Voxel grid downsampling is performed on the point cloud points in each target point cloud partition to obtain a representative set of sampling points; Each sampling point in the representative sampling point set is translated along its corresponding corrected normal vector direction by a preset security check distance to obtain a six-dimensional candidate viewpoint corresponding to each sampling point, forming the initial candidate viewpoint set.

[0045] Principal component analysis (PCA) refers to transforming multidimensional correlated point cloud data into a set of linearly independent principal components through orthogonal linear transformations, achieving dimensionality reduction and feature extraction while preserving the main features of the data. A pre-defined spatial guidance direction is also included. Voxel grid downsampling involves dividing a 3D point cloud into a regular 3D voxel grid and selecting representative points within each voxel to reduce the amount of point cloud data while preserving the overall geometric features of the target. A six-dimensional candidate viewpoint refers to the viewpoint represented by the 3D coordinates of a point cloud point in 3D space and the corresponding normal vector components along the 3D coordinate axes. For example, the 3D coordinates of the current point cloud point are x, y, z, and the corresponding normal vector components along the three coordinate axes are n. x n y n z The corresponding six-dimensional candidate viewpoint is v=(x, y, z, n). x n y n z ).

[0046] Understandably, in generating such Figure 5 When the UAV scans the initial candidate viewpoint set on the upper surface of the aircraft, it can sequentially acquire each point cloud point in the target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft as the current point cloud point. A covariance matrix is ​​constructed based on a preset number of nearest neighbors corresponding to the current point cloud point. Principal component analysis is used to perform eigenvalue decomposition on this covariance matrix to obtain the corresponding eigenvectors. The eigenvector corresponding to the smallest eigenvalue is the initial normal vector corresponding to the current point cloud point. Then, based on the preset spatial guidance direction of the upper surface functional partition to which each target point cloud partition belongs, the dot product of the initial normal vector and the preset spatial guidance direction is calculated. The resulting dot product is then uniformly sign-corrected, ensuring that the normal vectors of all point cloud points point towards the external space of the upper surface of the aircraft.

[0047] The specific sign correction method can be as follows: Assume the initial normal vector of point i in the point cloud. Guided by the preset space The dot product is It can be done through the formula: The corrected normal vector of point i in the point cloud is calculated. Among them, sgn() is the standard symbolic function.

[0048] Voxel grid downsampling is performed on the point cloud points in each target point cloud partition to obtain a representative set of sampling points with uniform distribution and reasonable density. Each sampling point in the representative sampling point set is selected as the current sampling point. The current sampling point is translated along the direction of the normal vector after correction processing by a preset safety check distance. That is, the calculation result of the current sampling point's three-dimensional coordinates plus the safety check distance is combined with the current sampling point's normal vector after correction processing to obtain a six-dimensional candidate viewpoint corresponding to the current sampling point. All sampling points are traversed to form an initial candidate viewpoint set.

[0049] Optionally, based on a global Euclidean distance field matching the overall point cloud data, a safe viewpoint set is selected from the initial candidate viewpoint set, including: Construct a three-dimensional spatial bounding box that can completely cover the overall point cloud data of the aircraft to be scanned and the initial candidate viewpoint set; The three-dimensional bounding box is discretized into a three-dimensional voxel mesh, and based on the overall point cloud data, the shortest Euclidean distance from each voxel sampling point in the three-dimensional voxel mesh to the aircraft surface is calculated using a distance transformation algorithm. A global Euclidean distance field is constructed based on the shortest Euclidean distance corresponding to each of the voxel sampling points. Based on the position of each candidate viewpoint in the initial candidate viewpoint set in the global Euclidean distance field, the corresponding shortest Euclidean distance is queried, and candidate viewpoints whose distance is greater than a preset global safe distance threshold are selected to form the safe viewpoint set.

[0050] The three-dimensional bounding box refers to the smallest regular three-dimensional geometric body that can completely enclose the overall point cloud data of the aircraft and the initial candidate viewpoint set. It can be used to limit the target range, simplify spatial positioning, and perform fast clipping operations. The three-dimensional voxel mesh refers to a regular three-dimensional grid structure formed by uniformly dividing the three-dimensional bounding box according to a preset size. Each grid cell serves as an independent voxel unit and can be used for indexing, sampling, neighborhood retrieval, and geometric feature quantization of point cloud points and viewpoints. The distance transformation algorithm calculates the Euclidean distance from each point cloud point in the three-dimensional voxel mesh to the boundary of the target region to form a distance field, which can be used for region segmentation, center extraction, path planning, and morphological operations. A voxel sampling point refers to a discrete point within a three-dimensional voxel mesh cell, selected or calculated according to sampling rules, used to characterize the spatial position and geometric features of that voxel. The global safe distance threshold refers to a pre-set, uniformly applicable minimum safe distance limit in global space to ensure that the UAV does not collide with the aircraft surface. It is used to constrain the viewpoint path and scanning attitude, ensuring the safety and reliability of the detection process.

[0051] Understandably, to more accurately describe the positional relationship between the initial candidate viewpoint set and the point cloud data of the overall aircraft, a 3D spatial bounding box can be constructed that fully covers all point cloud data of the overall aircraft and all viewpoints in the initial candidate viewpoint set. This 3D spatial bounding box is discretized into multiple 3D voxel meshes according to a preset size. Each voxel mesh is processed sequentially, and multiple voxel sampling points are selected. The shortest Euclidean distance from each voxel sampling point to the upper surface of the aircraft is calculated using a distance transformation algorithm. After processing all the 3D voxel meshes, a global Euclidean distance field can be constructed based on the shortest Euclidean distances from all voxel sampling points to the upper surface of the aircraft. The shortest Euclidean distance from each candidate viewpoint in the initial candidate viewpoint set to the upper surface of the aircraft in the global Euclidean distance field is then sequentially queried. Candidate viewpoints whose shortest Euclidean distance is greater than a preset global safe distance threshold are selected. All qualified candidate viewpoints selected from the initial candidate viewpoint set are combined as follows: Figure 6 The set of security viewpoints shown.

[0052] S230: Construct a sensor imaging model based on the key shooting parameters of the camera equipment mounted on the drone.

[0053] The key shooting parameters include focal length, sensor size, image resolution, effective depth of field range, and maximum allowable surface incident angle.

[0054] Among them, the sensor imaging model can refer to the geometric mapping model established based on imaging parameters such as sensor size, focal length, and resolution, which is used to characterize the projection relationship from three-dimensional spatial points to two-dimensional imaging planes and imaging constraints.

[0055] It is understandable that to construct a sensor imaging model that matches the camera equipment on the drone, a pyramidal field-of-view model can be used to represent the imaging area of ​​the camera equipment. The key parameters of this model include focal length, sensor size, image resolution, effective depth of field range, and maximum allowable surface incident angle. By integrating the above key parameters into the constructed pyramidal field-of-view model, the construction and initialization of the sensor imaging model are completed, providing a constraint basis for subsequent viewpoint selection.

[0056] S240. Based on the sensor imaging model, construct a binarized visibility matrix, wherein the number of rows in the visibility matrix matches the total number of viewpoints at the safe viewpoint, and the number of columns matches the total number of point clouds contained in each target point cloud partition. The set row vector in the visibility matrix represents all target points belonging to the target point cloud partition that can be observed without obstruction from the set safe viewpoint.

[0057] It is understandable that a binary visibility matrix V∈{0,1} is constructed based on the sensor imaging model. (M×N) Where M is the number of rows in the visibility matrix, representing the total number of viewpoints at safe viewpoints, and N is the number of columns in the visibility matrix, representing the total number of point clouds contained in each target point cloud partition on the aircraft's upper surface. The defined row vector V in the visibility matrix... ij =1 is used to indicate the first [value] in the safety viewpoint set. i Each viewpoint can observe the first target point cloud region on the aircraft's upper surface without obstruction. j For each target point, the visibility matrix can be used to define the row vectors that represent the safe viewpoint can observe without obstruction of all target points belonging to each target point cloud partition.

[0058] In addition, the geometric and physical constraints comprehensively considered in the construction of visibility moments may include: the target point is within the effective depth of field of the sensor, the angle between the line-of-sight vector of the target point relative to the viewpoint and the optical axis of the sensor is within the field of view, the angle between the line-of-sight vector and the normal vector at the target point is less than a set threshold, and spatial acceleration queries can be performed by constructing a binary tree spatial index structure for recursive partitioning of multidimensional spatial data to ensure that the line-of-sight vector intersects with the unobstructed surface of the aircraft.

[0059] S250. Based on the visibility matrix, a greedy algorithm is used to iteratively select safe viewpoints from the safe viewpoint set and add them to the optimal viewpoint set. In each iteration, a safe viewpoint that can newly cover the most uncovered target point cloud points is selected until the proportion of the total number of target point cloud points covered by the optimal viewpoint set to the total number of all target point cloud points reaches the preset coverage threshold.

[0060] Among them, target point cloud points can refer to point cloud points in the overall point cloud data of the aircraft used to determine the viewpoint coverage.

[0061] It is understandable that selecting security viewpoints from a set of security viewpoints and adding them to the list is appropriate. Figure 7 The optimal viewpoint set shown can be transformed into a weighted set coverage problem, which is solved using a greedy iterative strategy. In each iteration, the marginal coverage gain corresponding to each remaining candidate viewpoint in the safe viewpoint set is calculated, and the viewpoint with the largest gain is added to the optimal viewpoint set. The iteration terminates when the proportion of the total number of target point cloud points covered by the optimal viewpoint set to the total number of all target point cloud points reaches the preset coverage threshold. Finally, the optimal viewpoint set that meets the coverage threshold requirement and has the smallest size is output.

[0062] S260. Based on the asymmetric traveling salesman problem modeled by the viewpoint access order, a hybrid solution strategy is used to solve for at least one optimal access sequence that matches the optimal viewpoint set.

[0063] Optionally, based on the asymmetric traveling salesman problem modeled by viewpoint access order, a hybrid solution strategy is employed to solve for at least one optimal access sequence matching the optimal viewpoint set, including: Based on the functional partition to which each optimal viewpoint in the optimal viewpoint set belongs, the optimal viewpoint set is divided into multiple optimal viewpoint subsets; For each subset of optimal viewpoints, an asymmetric cost matrix is ​​constructed, where each element of the asymmetric cost matrix represents the total access cost required to move from one optimal viewpoint to another, and the total access cost includes Euclidean distance cost and UAV yaw angle change cost. For each optimal viewpoint subset, the nearest neighbor algorithm is used to construct the initial access sequence corresponding to each optimal viewpoint subset. A preset local search algorithm is used to perform local iterative optimization on each initial access sequence to minimize the total access cost and obtain the optimal access sequence corresponding to each optimal viewpoint subset.

[0064] The asymmetric cost matrix can be a two-dimensional cost matrix in which the path cost from viewpoint i to viewpoint j is not equal to the path cost from viewpoint j to viewpoint i, and its matrix elements do not satisfy symmetry. It can be used to describe the asymmetric cost constraints in the process of viewpoint transformation or matching. The local search algorithm can be a local search optimization algorithm that optimizes the path length and path rationality by iteratively removing two intersecting edges in the viewpoint access path and reconnecting them.

[0065] Understandably, the optimal access sequence matching the optimal viewpoint set can be determined by dividing the optimal viewpoint set into multiple optimal viewpoint subsets corresponding to each target functional partition on the aircraft's upper surface, based on the target functional partitions to which each viewpoint belongs. Optionally, each optimal viewpoint subset can be assigned according to the results of the previous semantic partitioning. For example, viewpoints on the four partitions of the forward fuselage, the rear fuselage plus tail, the left wing, and the right wing can be respectively assigned as four optimal viewpoint subsets. Alternatively, viewpoints on the entire fuselage can be assigned as one optimal viewpoint subset, viewpoints on the two wings as another, and viewpoints on the tail section as yet another, resulting in three optimal viewpoint subsets in total.

[0066] By dividing the optimal viewpoint set into multiple optimal viewpoint subsets and planning optimal coverage paths for each subset, optimal paths for multiple UAVs can be obtained. Parallel operation can significantly improve detection efficiency.

[0067] Furthermore, an asymmetric cost matrix can be constructed for each optimal viewpoint subset to calculate the total access cost between any two viewpoints. Each element in the asymmetric cost matrix represents the total access cost required to move from one optimal viewpoint to another within the current optimal viewpoint subset. The total access cost includes not only the Euclidean distance in three-dimensional space but also the yaw angle change cost caused by the UAV adjusting its scanning observation attitude.

[0068] The asymmetric cost matrix corresponding to each optimal viewpoint subset is selected sequentially. Using the nearest neighbor algorithm, any viewpoint from the current optimal viewpoint subset is chosen as the current viewpoint. The unvisited viewpoint with the smallest cost from the current viewpoint is selected and used as the current viewpoint. This process is repeated iteratively to select the unvisited viewpoint with the smallest cost from the current viewpoint, thus constructing an initial access sequence matching the current optimal viewpoint subset. To avoid the initial viewpoint access sequence getting trapped in local optima, a pre-defined local search algorithm is used as a basis to introduce the 2-Opt operator for local iterative optimization. This can cut two non-adjacent edges in the viewpoint paths included in the initial viewpoint access sequence and reconnect them. If the total cost of the new path decreases, its topology is preserved. This process is repeated until the path cannot be optimized or the maximum number of iterations is reached, thus obtaining the optimal access sequence corresponding to each optimal viewpoint subset.

[0069] S270. Construct scanning paths corresponding to each of the optimal access sequences, and send each scanning path to the UAV so that each UAV can scan and photograph the upper surface of the aircraft to be scanned.

[0070] Optionally, construct scan paths corresponding to each optimal access sequence, including: Among all optimal access sequences, obtain the current optimal access sequence; Using the optimal viewpoints contained in the current optimal access sequence as path anchor points, multiple control points are obtained based on the cubic uniform B-spline inverse calculation formula. Based on each control point, trajectory fitting is performed on the cubic uniform B-spline curve to calculate the initial smooth trajectory point set corresponding to the current optimal access sequence. Based on the minimum turning radius constraint of the UAV, the initial smooth trajectory point set is subjected to dynamic constraint feasibility verification, and the initial smooth trajectory point set is subjected to local relaxation processing based on the verification result to obtain the optimized smooth trajectory point set. Based on the pre-constructed global Euclidean distance field, collision detection is performed on the optimized smooth trajectory point set to identify safety risk points in the optimized smooth trajectory point set whose shortest distance to the aircraft surface is less than the global safe distance. For each of the aforementioned safety risk points, a repulsion vector is applied in the direction away from its nearest obstacle point to correct the position of each safety risk point, resulting in an updated set of optimized smooth trajectory points. Return to the operation of performing collision detection on the optimized smooth trajectory point set based on the pre-constructed global Euclidean distance field until there are no safety risk points in the optimized smooth trajectory point set; The optimized smooth trajectory point set obtained at the end of the iteration is determined as the scan path corresponding to the current optimal access sequence.

[0071] Cubic uniform B-spline curves refer to piecewise continuous and smooth curves constructed from cubic polynomial basis functions with uniformly distributed node vectors. They are characterized by local controllability, good continuity, and gentle curvature changes, and are often used for viewpoint path smoothing and trajectory optimization. The cubic uniform B-spline inverse calculation formula refers to the mathematical formula for calculating the corresponding control points based on a given optimal viewpoint sequence. It is used to inversely derive the control parameters of the cubic uniform B-spline curve from the trajectory path, achieving accurate fitting and generation of the viewpoint motion trajectory.

[0072] It is understandable that UAVs experience discontinuous speeds and rapid changes in angular velocity at viewpoint switching points, requiring them to frequently perform continuous deceleration, hovering, turning, and acceleration maneuvers. To adapt to the continuous operation characteristics of UAVs, when constructing the scanning paths corresponding to each optimal access sequence, the optimal access sequence corresponding to each target functional area on the aircraft's upper surface can be sequentially obtained as the current optimal access sequence. Using the optimal viewpoints in the current optimal access sequence as path anchor points, multiple corresponding control points can be calculated using the cubic uniform B-spline inverse calculation formula. These control points are then fitted with the cubic uniform B-spline curve to obtain the initial smooth trajectory point set corresponding to the current optimal access sequence. To further obtain an optimized smooth trajectory point set, the feasibility of dynamic constraints on each viewpoint in the initial smooth trajectory point set can be verified. Based on the constraint of the minimum turning radius of UAV flight, the curvature of each viewpoint in the initial smooth trajectory point set is calculated. For viewpoints that do not meet the minimum turning radius constraint, local relaxation can be performed by inserting additional viewpoints or adjusting curve parameters to satisfy the minimum turning radius constraint, thus obtaining the optimized smooth trajectory point set.

[0073] The generated optimized smooth trajectory point set may have issues such as crossing the aircraft's upper surface or being too close to point cloud points on the aircraft's upper surface. To ensure the flight safety of the UAV when scanning the aircraft's upper surface from various viewpoints, collision detection needs to be performed on the obtained optimized smooth trajectory point set. Based on a pre-constructed global Euclidean distance field, safety risk points in the optimized smooth trajectory point set whose shortest distance to the aircraft's upper surface is less than the global safe distance are identified. KDTree (a binary tree structure for fast nearest neighbor retrieval of high-dimensional spatial data) can be used to densely sample and perform collision detection on the path segments between viewpoints in the optimized smooth trajectory point set. When a path is detected to be less than the safe threshold in distance to point cloud points on the aircraft's upper surface, indicating a potential collision hazard, a repulsive force vector is applied along the normal direction of the point cloud points on the aircraft's upper surface or in the vertically upward direction, pushing the corresponding path viewpoint to a safe area. Finally, based on the optimized smooth trajectory point set that has been iteratively corrected and has no safety risk points, the optimal access sequence corresponding to the current optimal access sequence is output, satisfying dynamic constraints and without collisions. Figure 9 The smooth scan path shown.

[0074] This invention, through the aforementioned technical solution, can dynamically calculate curvature thresholds and normal vector angle thresholds based on point cloud density. It identifies and divides the overall point cloud data of the aircraft into target point cloud partitions corresponding to each target functional area on the aircraft's upper surface. This ensures that the normal vectors of all point clouds in each target point cloud partition, after correction processing, point outwards from the aircraft's surface. Voxel grid downsampling is performed on each point cloud in each target point cloud partition, and the sampling points are shifted by a preset safety check distance to obtain an initial candidate viewpoint set. Based on a global Euclidean distance field matching the overall point cloud data of the aircraft to be scanned, candidate viewpoints with distances greater than a preset global safety distance threshold are selected from the initial candidate viewpoint set to form a safe viewpoint set. In the binary visibility matrix constructed based on the sensor model, a greedy algorithm iteratively selects safe viewpoints from the safe viewpoint set that can newly cover the most uncovered target point cloud points, until the proportion of selected safe viewpoints to the total number of point cloud points reaches a coverage threshold. This yields an optimal viewpoint set composed of the selected safe viewpoints. Modeling the optimal viewpoint set involves constructing an asymmetric cost matrix from the optimal viewpoint subsets of each target functional zone on the aircraft's upper surface. Using nearest neighbor and local search algorithms, multiple optimal access sequences minimizing total visits can be obtained. Based on these optimal access sequences and the cubic uniform B-spline inverse calculation formula, corresponding initial smooth trajectory point sets are constructed. After dynamic constraint feasibility verification and safety risk point removal, the updated optimized smooth trajectory point set generates the scanning path. This technical solution comprehensively considers the geometric characteristics of the UAV scanning viewpoint and the actual flight scenario, resulting in a safer and smoother UAV scanning flight path through algorithmic processing.

[0075] Example 3 Figure 9 This is a schematic diagram of a scanning path planning device for an aircraft upper surface provided in Embodiment 3 of the present invention. Figure 9 As shown, the device includes: a point cloud partition acquisition module 910, a safe viewpoint set acquisition module 920, an optimal viewpoint set acquisition module 930, an optimal access sequence acquisition module 940, and a scan path generation module 950.

[0076] The point cloud partitioning acquisition module 910 is used to divide the overall point cloud data of the aircraft to be scanned based on the joint detection algorithm of curvature, normal vector and point cloud density, and obtain each target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft.

[0077] The safe viewpoint set acquisition module 920 is used to perform consistency correction processing on the normal vector directions of each point cloud point in each target point cloud partition, generate an initial candidate viewpoint set along the corrected normal vector directions, and select a safe viewpoint set from the initial candidate viewpoint set based on the global Euclidean distance field that matches the overall point cloud data.

[0078] The optimal viewpoint set acquisition module 930 is used to select the optimal viewpoint set from the safe viewpoint set based on the visibility matrix constructed by the sensor model and a greedy algorithm based on a preset coverage threshold condition.

[0079] The optimal access sequence acquisition module 940 is used to solve at least one optimal access sequence that matches the optimal viewpoint set based on the asymmetric traveling salesman problem modeled by the viewpoint access order, using a hybrid solution strategy.

[0080] The scanning path generation module 950 is used to construct scanning paths corresponding to each of the optimal access sequences, and to send each of the scanning paths to the UAVs so that each UAV can scan and photograph the upper surface of the aircraft to be scanned.

[0081] This invention, through the aforementioned technical solution, utilizes a joint detection algorithm based on curvature, normal vector, and point cloud density to divide the overall point cloud data of the aircraft to be scanned into target point cloud partitions corresponding to various target functional areas on the aircraft's upper surface. An initial candidate viewpoint set is generated along the normal vector direction of each point cloud point in each target point cloud partition after correction processing. Based on a global Euclidean distance field matching the overall point cloud data of the aircraft to be scanned, a safe viewpoint set can be selected from the initial candidate viewpoint set according to a preset safe distance. A greedy algorithm based on a preset coverage threshold condition is used in the visibility matrix constructed by the sensor model to select the optimal viewpoint set from the safe viewpoint set. An asymmetric traveling salesman problem with optimal sequence is modeled for the optimal viewpoint set. A hybrid solution strategy is used to obtain multiple optimal access sequences matching the optimal viewpoint set. Corresponding scanning paths can be constructed based on these multiple optimal access sequences and sent to the UAV to control the UAV to scan and photograph the upper surface of the aircraft. The above technical solution can obtain the optimal scanning path by solving the overall point cloud data of the aircraft's upper surface, effectively eliminating redundant observations, avoiding detours in the detection path, improving detection efficiency and operational safety, and enhancing the system's automation and detection stability.

[0082] Optionally, the point cloud partitioning module 910 can be specifically used to: dynamically calculate the optimal neighborhood size based on the global point cloud density matching the overall point cloud data of the aircraft to be scanned; and dynamically calculate the curvature threshold and the normal vector angle threshold based on the curvature and normal vector direction of each point cloud point in the overall point cloud data; add all the overall point cloud data to the unsegmented point set, and after selecting the point cloud point with the smallest curvature from the current unsegmented point set as the current growth seed, initialize and construct the current point cloud partition corresponding to the current growth seed. Determine the target search region matching the current growth seed based on the optimal neighborhood size, and filter all point cloud points in the unsegmented point set that are located within the target search region, have a curvature less than the curvature threshold, and have a normal vector angle with the current growth seed less than the normal vector angle threshold, and add them to the current point cloud partition. Return to the operation of selecting the point cloud point with the smallest curvature from the current unsegmented point set as the current growth seed until the processing of all point cloud points in the unsegmented point set is completed. After the iteration ends, all generated point cloud partitions are divided into fragmented partitions and standard partitions. The fragmented partitions are then merged into the nearest standard partition to obtain multiple candidate point cloud partitions. Each candidate point cloud partition is input into a pre-built classifier for semantic segmentation to determine the functional partition to which each candidate point cloud partition belongs, and to obtain the target point cloud partitions corresponding to each target functional partition on the upper surface of the aircraft. The target functional partitions include: the fuselage, the upper surfaces of the left and right wings, the upper surfaces of the left and right horizontal tails, and the left and right sides of the vertical tail. After determining the target point cloud partitions corresponding to the fuselage, the point cloud points located on the lower surface of the fuselage are filtered out.

[0083] Optionally, the safety viewpoint set acquisition module 920 can be specifically used for: for each point in each target point cloud partition, using a preset number of its nearest neighbors, calculating the initial normal vector corresponding to each point in the point cloud partition using principal component analysis. Based on the preset spatial guidance direction of the functional partition to which each target point cloud partition belongs, uniformly correcting the sign of each initial normal vector so that the normal vector of each corrected point in the point cloud partition points to the external space of the surface of the aircraft to be scanned. Performing voxel grid downsampling processing on the point in each target point cloud partition to obtain a representative sampling point set. Shifting each sampling point in the representative sampling point set along its corresponding corrected normal vector direction by a preset safety inspection distance to obtain a six-dimensional candidate viewpoint corresponding to each sampling point, forming the initial candidate viewpoint set.

[0084] Optionally, the safe viewpoint set acquisition module 920 can also be specifically used to: construct a three-dimensional spatial bounding box that can completely cover the overall point cloud data of the aircraft to be scanned and the initial candidate viewpoint set; discretize the three-dimensional spatial bounding box into a three-dimensional voxel grid; and, based on the overall point cloud data, use a distance transformation algorithm to calculate the shortest Euclidean distance from each voxel sampling point in the three-dimensional voxel grid to the aircraft surface; construct a global Euclidean distance field according to the shortest Euclidean distances corresponding to each voxel sampling point; query the corresponding shortest Euclidean distance according to the position of each candidate viewpoint in the initial candidate viewpoint set in the global Euclidean distance field; and filter out candidate viewpoints whose distances are greater than a preset global safe distance threshold to form the safe viewpoint set.

[0085] Optionally, the optimal viewpoint set acquisition module 930 can be specifically used to: construct a sensor imaging model based on the key shooting parameters of the camera device mounted on the UAV, wherein the key shooting parameters include focal length, sensor size, image resolution, effective depth of field, and maximum allowable surface incidence angle. Based on the sensor imaging model, a binarized visibility matrix is ​​constructed, wherein the number of rows in the visibility matrix matches the total number of viewpoints in the safe viewpoints, and the number of columns matches the total number of point clouds contained in each target point cloud partition. A set row vector in the visibility matrix represents all target points belonging to the target point cloud partition that the set safe viewpoints can observe without obstruction. Based on the visibility matrix, a greedy algorithm iteratively selects safe viewpoints from the safe viewpoint set to add to the optimal viewpoint set. Each iteration selects a safe viewpoint that can newly cover the most uncovered target point cloud points, until the proportion of the total number of target point cloud points covered by the optimal viewpoint set to the total number of all target point cloud points reaches a preset coverage threshold.

[0086] Optionally, the optimal access sequence acquisition module 940 can be specifically used to: divide the optimal viewpoint set into multiple optimal viewpoint subsets according to the functional partitions to which each optimal viewpoint belongs; construct an asymmetric cost matrix for each optimal viewpoint subset, wherein each matrix element in the asymmetric cost matrix represents the total access cost required to move from one optimal viewpoint to another, the total access cost including Euclidean distance cost and UAV yaw angle change cost; for the asymmetric cost matrix corresponding to each optimal viewpoint subset, use the nearest neighbor algorithm to construct an initial access sequence corresponding to each optimal viewpoint subset; and use a preset local search algorithm to perform local iterative optimization on each initial access sequence to minimize the total access cost, thereby obtaining the optimal access sequence corresponding to each optimal viewpoint subset.

[0087] Optionally, the scanning path generation module 950 can be specifically used to: obtain the current optimal access sequence from all optimal access sequences; use the optimal viewpoints included in the current optimal access sequence as path anchor points; obtain multiple control points based on the cubic uniform B-spline inverse calculation formula; and perform trajectory fitting based on the cubic uniform B-spline curve according to each control point to calculate the initial smooth trajectory point set corresponding to the current optimal access sequence. Based on the minimum turning radius constraint of the UAV, perform a dynamic constraint feasibility check on the initial smooth trajectory point set, and perform local relaxation processing on the initial smooth trajectory point set according to the check result to obtain an optimized smooth trajectory point set. Based on the pre-constructed global Euclidean distance field, perform collision detection on the optimized smooth trajectory point set, identify safety risk points in the optimized smooth trajectory point set whose shortest distance to the aircraft surface is less than the global safe distance, and apply a repulsion vector to each safety risk point along the direction away from its nearest obstacle point to correct the position of each safety risk point, thus obtaining an updated optimized smooth trajectory point set. Return to the operation of performing collision detection on the optimized smooth trajectory point set according to the pre-constructed global Euclidean distance field until there are no safety risk points in the optimized smooth trajectory point set. The optimized smooth trajectory point set finally obtained at the end of the iteration is determined as the scan path corresponding to the current optimal access sequence.

[0088] The scanning path planning device for the upper surface of an aircraft provided in this embodiment of the invention can execute the scanning path planning method for the upper surface of an aircraft provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0089] Example 4 Figure 10 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0090] like Figure 10As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0091] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0092] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a scan path planning method for the upper surface of an aircraft.

[0093] That is, based on the joint detection algorithm of curvature, normal vector and point cloud density, the overall point cloud data of the aircraft to be scanned is divided into data to obtain each target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft. After performing consistency correction processing on the normal vector direction of each point cloud point in each target point cloud partition, an initial candidate viewpoint set is generated along the corrected normal vector direction. Based on the global Euclidean distance field that matches the overall point cloud data, a safe viewpoint set is selected from the initial candidate viewpoint set. Based on the visibility matrix constructed by the sensor model, a greedy algorithm based on a preset coverage threshold condition is used to select the optimal set of viewpoints from the set of safe viewpoints. Based on the asymmetric traveling salesman problem modeled by the viewpoint access order, a hybrid solution strategy is used to solve for at least one optimal access sequence that matches the optimal viewpoint set; A scanning path corresponding to each of the optimal access sequences is constructed, and each scanning path is sent to the UAV so that each UAV can scan and photograph the upper surface of the aircraft to be scanned.

[0094] In some embodiments, the scanning path planning method for an aircraft upper surface can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the scanning path planning method for an aircraft upper surface described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the scanning path planning method for an aircraft upper surface by any other suitable means (e.g., by means of firmware).

[0095] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0096] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0097] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0098] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0099] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0100] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0101] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0102] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A scanning path planning method for the upper surface of an aircraft, characterized in that, include: Based on the joint detection algorithm of curvature, normal vector and point cloud density, the overall point cloud data of the aircraft to be scanned is divided into data segments to obtain each target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft. After performing consistency correction processing on the normal vector direction of each point cloud point in each target point cloud partition, an initial candidate viewpoint set is generated along the corrected normal vector direction. Based on the global Euclidean distance field that matches the overall point cloud data, a safe viewpoint set is selected from the initial candidate viewpoint set. Based on the visibility matrix constructed by the sensor model, a greedy algorithm based on a preset coverage threshold condition is used to select the optimal set of viewpoints from the set of safe viewpoints. Based on the asymmetric traveling salesman problem modeled by the viewpoint access order, a hybrid solution strategy is used to solve for at least one optimal access sequence that matches the optimal viewpoint set; A scanning path corresponding to each of the optimal access sequences is constructed, and each scanning path is sent to the UAV so that each UAV can scan and photograph the upper surface of the aircraft to be scanned.

2. The method according to claim 1, characterized in that, The joint detection algorithm based on curvature, normal vector, and point cloud density divides the overall point cloud data of the aircraft to be scanned into target point cloud partitions corresponding to each target functional area on the upper surface of the aircraft, including: Based on the global point cloud density that matches the overall point cloud data of the aircraft to be scanned, the optimal neighborhood size is dynamically calculated, and based on the curvature and normal vector direction of each point cloud point in the overall point cloud data, the curvature threshold and normal vector angle threshold are dynamically calculated. All the overall point cloud data is added to the undivided point set, and the point cloud point with the smallest curvature is selected from the current undivided point set as the current growth seed. Then, the current point cloud partition corresponding to the current growth seed is initialized and constructed. The target search region matching the current growth seed is determined based on the optimal neighborhood size. All point cloud points located within the target search region, with curvature less than the curvature threshold and normal vector angle between them and the current growth seed less than the normal vector angle threshold are selected from the unsegmented point set and added to the current point cloud partition. Return to the operation of selecting the point cloud point with the smallest curvature from the current set of undivided points as the current growth seed, until the processing of all point cloud points in the set of undivided points is completed; After the iteration ends, all generated point cloud partitions are divided into fragmented partitions and standard partitions. The fragmented partitions are then merged into the nearest standard partition to obtain multiple candidate point cloud partitions. Each of the candidate point cloud partitions is input into a pre-built classifier for semantic segmentation to determine the functional partition to which each candidate point cloud partition belongs, and to obtain each target point cloud partition corresponding to each target functional partition on the upper surface of the aircraft; wherein, the target functional partitions include: fuselage, upper surfaces of left and right wings, upper surfaces of left and right horizontal tails, and left and right sides of vertical tail. After determining the target point cloud partition corresponding to the fuselage, the point cloud points located on the lower surface of the fuselage are filtered out.

3. The method according to claim 1, characterized in that, After performing consistency correction processing on the normal vector directions of each point cloud point in each target point cloud partition, an initial candidate viewpoint set is generated along the corrected normal vector directions, including: For each point in each target point cloud partition, the initial normal vector corresponding to each point is calculated using principal component analysis with a preset number of its nearest neighbors. According to the preset spatial guidance direction of the functional area to which each target point cloud partition belongs, the initial normal vectors are uniformly corrected so that the normal vector of each point cloud point after correction points to the external space of the surface of the aircraft to be scanned. Voxel grid downsampling is performed on the point cloud points in each target point cloud partition to obtain a representative set of sampling points; Each sampling point in the representative sampling point set is translated along its corresponding corrected normal vector direction by a preset security check distance to obtain a six-dimensional candidate viewpoint corresponding to each sampling point, forming the initial candidate viewpoint set.

4. The method according to any one of claims 1-3, characterized in that, The process of selecting a safe viewpoint set from the initial candidate viewpoint set based on the global Euclidean distance field matched with the overall point cloud data includes: Construct a three-dimensional spatial bounding box that can completely cover the overall point cloud data of the aircraft to be scanned and the initial candidate viewpoint set; The three-dimensional bounding box is discretized into a three-dimensional voxel mesh, and based on the overall point cloud data, the shortest Euclidean distance from each voxel sampling point in the three-dimensional voxel mesh to the aircraft surface is calculated using a distance transformation algorithm. A global Euclidean distance field is constructed based on the shortest Euclidean distance corresponding to each of the voxel sampling points. Based on the position of each candidate viewpoint in the initial candidate viewpoint set in the global Euclidean distance field, the corresponding shortest Euclidean distance is queried, and candidate viewpoints whose distance is greater than a preset global safe distance threshold are selected to form the safe viewpoint set.

5. The method according to any one of claims 1-3, characterized in that, The step of selecting the optimal set of viewpoints from the safe viewpoint set using a greedy algorithm based on a preset coverage threshold condition, based on the visibility matrix constructed by the sensor model, includes: Based on the key shooting parameters of the camera equipment mounted on the drone, a sensor imaging model is constructed, wherein the key shooting parameters include focal length, sensor size, image resolution, effective depth of field range, and maximum allowable surface incident angle. Based on the sensor imaging model, a binarized visibility matrix is ​​constructed, wherein the number of rows in the visibility matrix matches the total number of viewpoints at the safe viewpoint, and the number of columns matches the total number of point clouds contained in each target point cloud partition. The set row vector in the visibility matrix represents all target points belonging to the target point cloud partition that can be observed without obstruction from the set safe viewpoint. Based on the visibility matrix, a greedy algorithm is used to iteratively select safe viewpoints from the safe viewpoint set and add them to the optimal viewpoint set. In each iteration, a safe viewpoint that can newly cover the most uncovered target point cloud points is selected until the proportion of the total number of target point cloud points covered by the optimal viewpoint set to the total number of all target point cloud points reaches a preset coverage threshold.

6. The method according to claim 4, characterized in that, The asymmetric traveling salesman problem, modeled based on viewpoint access order, employs a hybrid solution strategy to find at least one optimal access sequence matching the optimal viewpoint set, including: Based on the functional partition to which each optimal viewpoint in the optimal viewpoint set belongs, the optimal viewpoint set is divided into multiple optimal viewpoint subsets; For each subset of optimal viewpoints, an asymmetric cost matrix is ​​constructed, where each element of the asymmetric cost matrix represents the total access cost required to move from one optimal viewpoint to another, and the total access cost includes Euclidean distance cost and UAV yaw angle change cost. For each optimal viewpoint subset, the nearest neighbor algorithm is used to construct the initial access sequence corresponding to each optimal viewpoint subset. A preset local search algorithm is used to perform local iterative optimization on each initial access sequence to minimize the total access cost and obtain the optimal access sequence corresponding to each optimal viewpoint subset.

7. The method according to claim 6, characterized in that, The construction of the scan path corresponding to each optimal access sequence includes: Among all optimal access sequences, obtain the current optimal access sequence; Using the optimal viewpoints contained in the current optimal access sequence as path anchor points, multiple control points are obtained based on the cubic uniform B-spline inverse calculation formula. Based on each control point, trajectory fitting is performed on the cubic uniform B-spline curve to calculate the initial smooth trajectory point set corresponding to the current optimal access sequence. Based on the minimum turning radius constraint of the UAV, the initial smooth trajectory point set is subjected to dynamic constraint feasibility verification, and the initial smooth trajectory point set is subjected to local relaxation processing based on the verification result to obtain the optimized smooth trajectory point set. Based on the pre-constructed global Euclidean distance field, collision detection is performed on the optimized smooth trajectory point set to identify safety risk points in the optimized smooth trajectory point set whose shortest distance to the aircraft surface is less than the global safe distance. For each of the aforementioned safety risk points, a repulsion vector is applied in the direction away from its nearest obstacle point to correct the position of each safety risk point, resulting in an updated set of optimized smooth trajectory points. Return to the operation of performing collision detection on the optimized smooth trajectory point set based on the pre-constructed global Euclidean distance field until there are no safety risk points in the optimized smooth trajectory point set; The optimized smooth trajectory point set obtained at the end of the iteration is determined as the scan path corresponding to the current optimal access sequence.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, which enables the at least one processor to perform the scanning path planning method for the upper surface of an aircraft as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the scanning path planning method for the upper surface of an aircraft as described in any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the scanning path planning method for the upper surface of an aircraft according to any one of claims 1-7.

Images