Methods, apparatus, equipment and storage media for extracting power line point clouds

Abstract

This invention belongs to the field of point cloud target detection technology, and discloses a method, apparatus, equipment, and storage medium for extracting power line point clouds. The method includes: acquiring urban environmental point cloud data; filtering the urban environmental point cloud data according to cloth filtering and horizontal linear feature constraints to obtain initial power line point cloud data; performing multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data; and extracting point clouds according to the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data. This invention greatly suppresses various types of ground noise through multiple constraints such as linear features, vertical continuity, and local geometric consistency. Multi-scale voxel analysis and density-based clustering effectively avoid over-segmentation or accidental deletion of power line point clouds during the filtering process, ensuring that the extracted power lines are continuous and complete, and achieving more accurate power line point cloud extraction.

Description

Technical Field

[0001] This invention relates to the field of point cloud target detection technology, and in particular to a method, apparatus, equipment and storage medium for extracting power line point clouds. Background Technology

[0002] Power lines are a vital infrastructure ensuring the nation's economy and people's livelihoods, and their stable operation is crucial. Especially in complex urban environments, the rapid and effective execution of power line inspections has become a critical issue that the power industry urgently needs to address. Traditional power line inspection methods mainly rely on manual on-site surveys, which are not only labor-intensive and inefficient but also create blind spots in steep terrain and densely vegetated areas. LiDAR technology, with its high-efficiency and high-precision three-dimensional real-time data acquisition capabilities, is widely used in power line inspections, enabling digital management and intelligent inspection of transmission lines.

[0003] However, in urban scenarios, power lines are intertwined and covered by trees, making the efficient and accurate extraction of power lines a pressing issue in urban intelligent power line inspection. Current technologies still have room for improvement in computational efficiency and accuracy when processing large-scale point cloud data, especially in complex environments, where they cannot simultaneously guarantee both accuracy and completeness in power line extraction. Therefore, the accuracy of power line extraction still needs improvement. Summary of the Invention

[0004] In view of this, it is necessary to provide a method, apparatus, equipment and storage medium for extracting power line point clouds to solve the technical problem of inaccurate power line detection.

[0005] To address the above problems, this invention provides a method for extracting power line point clouds, comprising:

[0006] Acquire urban environmental point cloud data;

[0007] The initial power line point cloud data is obtained by performing point cloud filtering on the urban environment point cloud data based on cloth filtering and horizontal linear feature constraints.

[0008] Multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering are performed on the initial power line point cloud data to obtain reference power line point cloud data.

[0009] Point cloud extraction is performed based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data.

[0010] In one possible implementation, the step of obtaining initial power line point cloud data by filtering the urban environment point cloud data based on cloth filtering and horizontal linear feature constraints includes:

[0011] The urban environment point cloud data is simulated by cloth particle sedimentation and terrain iterative convergence based on the cloth simulation filtering algorithm to obtain ground point cloud data.

[0012] The ground point cloud data in the urban environment point cloud data is removed to obtain the point cloud data to be filtered, and the point cloud data to be filtered is divided into multiple point cloud clusters according to Euclidean clustering.

[0013] The point cloud is divided according to the spatial length of each point cloud cluster to obtain filtered point cloud data and interfering point cloud;

[0014] Linear feature recall is performed on the interfering point cloud by principal components with horizontal linear constraints to obtain recalled point cloud data. The filtered point cloud data and the recalled point cloud data are used as the initial power line point cloud data.

[0015] In one possible implementation, the step of performing multi-scale voxel connectivity analysis and vertical continuity-constrained point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data includes:

[0016] The initial power line point cloud data is converted into a three-dimensional voxel mesh, each of which includes spatial location coordinates, a unique index value, and the number of point clouds contained within it.

[0017] Near-ground voxels are extracted from the three-dimensional voxel grid according to a preset height threshold.

[0018] Based on the first reference scale, each near-ground voxel is searched upwards to obtain a connected voxel grid. The connected voxel grids in the three-dimensional voxel grid are then filtered out to obtain a reference voxel grid.

[0019] The vertical continuous point cloud in the reference voxel grid is extracted based on a second reference scale, where the second reference scale is smaller than the first reference scale.

[0020] The reference power line point cloud data is obtained by removing vertical continuous point clouds in the reference voxel grid whose consecutive counts are greater than a preset continuous threshold.

[0021] In one possible implementation, the step of extracting point cloud data based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data includes:

[0022] By performing DBSCAN density spatial clustering on the reference power line point cloud data using preset clustering parameters, multiple candidate point cloud clusters are obtained.

[0023] The covariance matrix is ​​constructed based on the cluster centroids of the candidate point cloud clusters. The linearity of each candidate point cloud cluster is obtained by performing eigenvalue decomposition on the covariance matrix through principal component analysis.

[0024] Filter out candidate point cloud clusters whose linearity is less than the first preset linearity threshold to obtain reference candidate point cloud clusters;

[0025] The reference candidate point cloud clusters are subjected to directional consistency constraint filtering and linear enhancement constraint filtering to obtain power line point cloud data.

[0026] In one possible implementation, the step of constructing a covariance matrix based on the cluster centroids of candidate point cloud clusters, and then performing eigenvalue decomposition on the covariance matrix using principal component analysis to obtain the linearity of each candidate point cloud cluster includes:

[0027] Construct a covariance matrix based on the cluster centroids of the candidate point cloud clusters, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues;

[0028] The distribution relationship of candidate point cloud clusters is obtained based on the magnitude relationship between the feature values;

[0029] When the distribution relationship is linear, the linearity of the candidate point cloud cluster is calculated based on the feature value.

[0030] In one possible implementation, the step of performing directional consistency constraint filtering and linear enhancement constraint filtering on the reference candidate point cloud cluster to obtain power line point cloud data includes:

[0031] Extract the feature vectors and feature values ​​of the reference candidate point cloud clusters, and obtain the main direction of the reference candidate point cloud clusters based on the feature vectors and feature values;

[0032] A local neighborhood is constructed based on the point clouds in the reference candidate point cloud cluster and a preset search radius, and the local normal direction of each point cloud in the reference candidate point cloud cluster is obtained based on the local neighborhood.

[0033] By eliminating reference candidate point cloud clusters whose main direction and local normal direction are inconsistent, the point cloud cluster to be constrained is obtained.

[0034] Linear enhanced constraint filtering is applied to the point cloud cluster to be constrained to obtain power line point cloud data.

[0035] In one possible implementation, the step of performing linear enhanced constraint filtering on the point cloud cluster to obtain power line point cloud data includes:

[0036] Calculate the cluster length and cluster linearity of the point cloud cluster to be constrained;

[0037] Remove the point cloud clusters to be constrained whose cluster length is less than a preset length threshold and whose cluster linearity is less than or equal to a second preset linearity threshold, and obtain the point cloud clusters to be density constrained, wherein the second preset linearity threshold is greater than the first preset linearity threshold.

[0038] Calculate the spatial density of the point cloud cluster to be density constrained;

[0039] Remove point cloud clusters from the density-constrained point cloud clusters whose spatial density is greater than or equal to a preset density threshold and whose cluster linearity is less than or equal to a third preset linearity threshold, to obtain power line point cloud data where the third preset linearity threshold is greater than the first preset linearity threshold.

[0040] Furthermore, to achieve the above objectives, the present invention also proposes a power line point cloud extraction device, the power line point cloud extraction device comprising:

[0041] The acquisition module is used to acquire urban environmental point cloud data;

[0042] The filtering module is used to perform point cloud filtering on the urban environment point cloud data according to the cloth filtering and horizontal linear feature constraints to obtain the initial power line point cloud data.

[0043] The filtering module is also used to perform multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data.

[0044] The power line point cloud data extraction module is used to extract point cloud data based on the density characteristics and geometric features of the reference power line point cloud data.

[0045] Furthermore, to achieve the above objectives, the present invention also proposes an electronic device comprising: a memory, a processor, a display, and a power line point cloud extraction program stored in the memory and executable on the processor, the power line point cloud extraction program being configured to implement the steps of the power line point cloud extraction method as described above.

[0046] In addition, to achieve the above objectives, the present invention also proposes a storage medium storing a power line point cloud extraction program, which, when executed by a processor, implements the steps of the power line point cloud extraction method described above.

[0047] The beneficial effects of adopting the above implementation method are:

[0048] By employing multiple constraints such as linear features, vertical continuity, and geometric consistency, various types of ground feature noise are significantly suppressed. Multi-scale voxel analysis and density-based clustering effectively prevent power line point clouds from being over-segmented or mistakenly deleted during the filtering process, ensuring that the extracted power lines are continuous and complete, thus achieving more accurate power line point cloud extraction. Attached Figure Description

[0049] 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.

[0050] Figure 1 This is a flowchart illustrating the first embodiment of the power line point cloud extraction method of the present invention;

[0051] Figure 2 This is a schematic diagram illustrating the segmentation effect of ground points and non-ground points in the first embodiment of the power line point cloud extraction method of the present invention;

[0052] Figure 3 This is a schematic diagram of the point cloud voxelization result of the first embodiment of the power line point cloud extraction method of the present invention;

[0053] Figure 4 This is a schematic diagram of the vertical connectivity analysis results of the first embodiment of the power line point cloud extraction method of the present invention;

[0054] Figure 5 This is a schematic diagram of the vertical continuity analysis results of the first embodiment of the power line point cloud extraction method of the present invention;

[0055] Figure 6 This is a flowchart illustrating the second embodiment of the power line point cloud extraction method of the present invention;

[0056] Figure 7 This is a schematic diagram illustrating the effect of normal direction consistency in the second embodiment of the power line point cloud extraction method of the present invention.

[0057] Figure 8 This is a complete flowchart of power line point cloud extraction in the second embodiment of the power line point cloud extraction method of the present invention;

[0058] Figure 9 This is a structural block diagram of the first embodiment of the power line point cloud extraction device of the present invention;

[0059] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0061] In the description of the embodiments of this application, unless otherwise stated, "a plurality of" means two or more.

[0062] In this embodiment of the invention, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, apparatus, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product or device.

[0063] The naming or numbering of steps in the embodiments of the present invention does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The execution order of the named or numbered process steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effect can be achieved.

[0064] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0065] The executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or power line point cloud extraction device capable of performing the above functions. The following description uses a power line point cloud extraction device as an example to illustrate this embodiment and the subsequent embodiments.

[0066] This invention provides a method for extracting power line point clouds, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the power line point cloud extraction method of the present invention.

[0067] In this embodiment, the power line point cloud extraction method includes steps S10 to S40:

[0068] Step S10: Obtain urban environmental point cloud data.

[0069] Understandably, point cloud data can be understood as a collection of millions or even billions of three-dimensional points in space. Each point contains at least three-dimensional coordinates (X, Y, Z), and sometimes also information such as intensity and color.

[0070] It should be noted that urban environmental point cloud data may include power line point cloud data to be extracted as well as environmental interference point cloud data; among which, environmental interference point cloud data may include vegetation point cloud data, building point cloud data, and point cloud data of other urban facilities (streetlights, traffic lights, etc.).

[0071] It should be understood that urban environmental point cloud data is a complex spatial dataset with highly mixed and mutually occluded features of multiple land features.

[0072] It is worth noting that urban environmental point cloud data can be acquired through LiDAR or other point cloud data acquisition devices with high efficiency and high precision in real-time 3D data acquisition.

[0073] Point cloud data can be acquired by actively emitting laser pulses into the urban environment through airborne, vehicle-mounted, or ground-based LiDAR systems, and accurately measuring the three-dimensional coordinates (X, Y, Z) of each point by receiving the echoes, thereby generating a massive, discrete set of spatial points.

[0074] Step S20: Perform point cloud filtering on the urban environment point cloud data according to the cloth filtering and horizontal linear feature constraints to obtain the initial power line point cloud data.

[0075] It is understood that the constraints in this embodiment and the following embodiments can be similarly understood as a filtering strategy. For example, linear feature constraints can be understood as pre-set linear feature conditions. By traversing each point cloud data, point cloud data that does not meet the linear feature conditions are filtered out.

[0076] It should be noted that linear feature constraints can be linear feature constraints in the horizontal space. The purpose of linear feature constraints in the horizontal space is to reduce the data size, simplify the subsequent processing flow, and at the same time retain key information related to power lines, so as to provide a high-quality data foundation for subsequent power line extraction.

[0077] In one feasible implementation, step S20 may include steps A21 to A24:

[0078] Step A21: Based on the cloth simulation filtering algorithm, the urban environment point cloud data is subjected to cloth particle sedimentation simulation and terrain iterative convergence to obtain ground point cloud data.

[0079] Understandably, ground point cloud data can be a collection of all point clouds that represent "ground" or "terrain".

[0080] It should be noted that the cloth particle settling simulation is a reverse simulation process based on a physical model. The "cloth particles" do not actually exist, but are a virtual elastic cloth plane composed of regular grids that covers the original point cloud. Each grid node is a "particle".

[0081] The specific simulation process involves these virtual cloth particles "sinking" from top to bottom under the influence of gravity. When a particle comes into contact with a high-point cloud (considered as non-ground objects such as power lines, trees, and houses), its movement is stopped; while when a particle comes into contact with a low-point cloud (considered as potential ground points), it continues to fall until it reaches a steady state.

[0082] Furthermore, by simulating the settling of the fabric, the algorithm can initially identify the undulating contours of the terrain and separate objects such as power lines suspended high up from points close to the ground.

[0083] It should be noted that terrain iterative convergence is a progressively precise and self-optimizing computational process that gradually improves the results through repeated calculations. In each iteration, the algorithm fine-tunes the position of the cloth particles and the shape of the terrain based on the current cloth shape and the point cloud in contact with it. When the difference between the results of two consecutive iterations is less than a preset threshold, the algorithm is considered to have "converged." At this point, the cloth shape has stably adhered to the real terrain surface and no longer undergoes significant changes.

[0084] It should be noted that terrain iteration convergence ensures that the generated terrain model can accurately adapt to "complex urban road scenarios", avoiding misjudging natural terrain such as gentle slopes and road shoulders as non-ground objects, thereby achieving high-precision separation of ground point clouds.

[0085] Step A22: Remove the ground point cloud data from the urban environment point cloud data to obtain the point cloud data to be filtered, and divide the point cloud data to be filtered into multiple point cloud clusters according to Euclidean clustering.

[0086] It should be understood that the urban environmental point cloud data is processed by a cloth simulation filtering algorithm to obtain two parts of point cloud data, one part being the ground point cloud data and the other part being the point cloud data to be filtered.

[0087] In practical implementation, effective filtering of ground point clouds plays a crucial role in the efficient extraction of target features during power line point cloud data processing. Based on the spatial distribution characteristics of power lines in real-world scenarios, there is a significant elevation difference between near-ground laser point clouds and power line point clouds. Since ground point clouds constitute a large proportion of vehicle-mounted mobile laser scanning (MLS) datasets, their effective filtering can significantly reduce the computational complexity of subsequent processing.

[0088] However, it should be noted that this invention employs a cloth-based simulated filtering algorithm (CSF) to construct an adaptive digital elevation model (DEM), achieving ground point cloud separation through cloth particle dynamics simulation and terrain iterative convergence mechanism. For example... Figure 2 As shown, Figure 2 The image in the middle shows a schematic diagram illustrating the segmentation effect between ground points and non-ground points. Figure 2 All objects can be divided into two colors: dark color represents non-ground point cloud and light color represents ground point cloud. This can efficiently remove ground point cloud in complex urban road scenes, while preserving the complete power line point cloud, and constructing a high-confidence non-ground point cloud subset for subsequent power line feature extraction.

[0089] It should be noted that the point cloud data in urban scenes contains some discrete or irregularly distributed point cloud data, such as trees, buildings, power line towers, etc. In order to initially distinguish between power line point clouds and non-power line point clouds, Euclidean clustering is used to construct a point cloud segmentation model to quickly divide the non-ground point cloud into several point cloud clusters. Based on the distribution of point cloud clusters in horizontal space, the area where the power line point cloud is located can be initially located.

[0090] Step A23: Divide the point cloud according to the spatial length of each point cloud cluster to obtain filtered point cloud data and interfering point cloud.

[0091] It should be noted that, due to the obvious long linearity of electric field lines, the length of the point cloud cluster in horizontal space can be used as the main criterion for distinguishing electric field line point cloud clusters. If the length of the point cloud cluster is long enough, the cluster may contain electric field line point clouds and their supporting structures, such as large buildings; otherwise, it represents discrete and irregularly distributed targets.

[0092] In practical implementation, the spatial length of each point cloud cluster can be extracted using the following formula:

[0093]

[0094] Where D is the length of the point cloud cluster, X max X min Y max Y min These represent the maximum values ​​of the projection of each cluster onto the XOY plane after clustering.

[0095] It should be noted that a conventionally accepted length value for power lines is pre-set as the spatial length threshold. The spatial length of each extracted point cloud cluster is compared with the spatial length threshold. Point cloud clusters with spatial lengths greater than the spatial length threshold are used as filtered point cloud data, while point cloud clusters with spatial lengths less than or equal to the spatial length threshold are used as interference point clouds.

[0096] Step A24: Perform linear feature recall on the interfering point cloud using principal components with horizontal linear constraints to obtain recalled point cloud data. Use the filtered point cloud data and the recalled point cloud data as initial power line point cloud data.

[0097] It should be noted that while it can initially locate the area where the power line point cloud is located, due to the limitations of vehicle-mounted LiDAR imaging, the occlusion effect during data acquisition, and the complexity of the urban environment, some power line point clouds may be isolated and scattered, making them easy to be mistakenly deleted during clustering and affecting the completeness of extraction.

[0098] It should be emphasized that, based on the linear feature constraints of principal component analysis (PCA), point cloud clusters with significant linear features are recalled, ensuring the integrity of subsequent power line extraction.

[0099] It should be noted that principal component analysis (PCA) of interfering point clouds yields three mutually perpendicular eigenvectors (principal components) and their corresponding eigenvalues. The first principal component direction represents the direction of the longest distribution of point clouds, while the other eigenvalue explains the shape of the point cloud. If the eigenvalue of a point cloud cluster is significantly greater in the first principal component direction than in the other two directions, the cluster is geometrically distinctly linear or rod-shaped.

[0100] Furthermore, the horizontal linear constraint can be understood as determining the point cloud geometry as linear through PCA eigenvalues ​​and determining its spatial attitude as roughly horizontal through the principal component directions.

[0101] Therefore, performing linear feature recall on the interference point cloud using principal components with horizontal linear constraints to obtain the recalled point cloud data can be understood as taking the point cloud that satisfies the horizontal linear constraints from the interference point cloud as the recalled point cloud data.

[0102] In this implementation, by using cloth-based simulated filtering to remove massive amounts of ground points, the data scale and computational complexity are greatly reduced. Subsequently, Euclidean clustering and spatial length constraints are used for rapid and effective coarse screening to initially separate targets and interference objects with significant morphology. Furthermore, linear feature discrimination based on principal component analysis is introduced to intelligently review and remedy misjudgments generated by the coarse screening, specifically recalling fragmented point clouds that were mistakenly deleted but conform to the geometric characteristics of power lines. This ensures the high confidence (accuracy) of the final "initial power line point cloud data" and effectively compensates for the missed detection problem (completeness) caused by data defects through the recall mechanism, providing more accurate and complete preliminary point cloud data for subsequent power line point cloud extraction.

[0103] The above are merely feasible implementations of step S20 provided in this embodiment. This embodiment does not specifically limit the specific implementation of step S20.

[0104] Step S30: Perform multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data.

[0105] Understandably, the point cloud clusters containing power lines in the initial power line point cloud data also retain typical features such as vegetation, buildings, and power infrastructure. Compared to power lines, vegetation and buildings have vertical continuity. Therefore, a non-power line point filtering algorithm based on multi-scale voxel continuity constraints in vertical space can be used for point cloud filtering, namely multi-scale voxel connectivity analysis and vertical continuity constraints.

[0106] It should be noted that multi-scale voxel connectivity analysis can be understood as analyzing whether there is vertical spatial connectivity among voxels in point cloud data from the perspective of different voxel sizes, and performing point cloud filtering based on this connectivity. At the same time, vertical continuity constraint point cloud filtering can be understood as determining whether each voxel is an electric field point cloud based on vertical continuity, because electric field point clouds are obviously horizontal and do not have vertical continuity.

[0107] It should be noted that multi-scale voxel continuity constraints in vertical space are used. Through multi-scale voxel connectivity analysis and vertical continuity constraints, major noise points such as utility poles, walls, and buildings are effectively removed, reducing the interference of noise on the extraction results and improving the integrity of power line extraction.

[0108] In one feasible implementation, step S30 may include steps A31 to A35:

[0109] Step A31: Convert the initial power line point cloud data into a three-dimensional voxel grid. Each three-dimensional voxel grid includes spatial location coordinates, a unique index value, and the number of point clouds contained within it.

[0110] It should be understood that voxels can transform disordered discrete point cloud data into a three-dimensional mesh with topological relationships, which not only reduces the computational complexity of data processing, but also highlights the linear characteristics of electric field lines by leveraging the mesh, laying the foundation for efficient and accurate extraction.

[0111] It should be noted that converting the initial power line point cloud data into a three-dimensional voxel mesh makes it easier to represent the point cloud data after linear feature constraints in the horizontal space using a regular dense voxel mesh. Each voxel has voxel position coordinates (i.e., spatial position coordinates) which can be represented as (c, r, h), a unique index value, and the number of points contained within the current voxel, represented as B.

[0112] In practical implementation, the voxel position coordinates can be represented using the following formula:

[0113]

[0114] Where (x, y, z) are the original input point cloud coordinates, (x, y, z) min y min , z min ) represent the minimum values ​​of the point cloud data on the x-axis, y-axis, and z-axis, respectively, and d represents the voxel size.

[0115] It should be noted that after voxelization, there is a correspondence between the point cloud data and voxels. All points can be found in voxels through voxel indexing. Therefore, voxels can be used as the smallest unit in point cloud processing. A diagram illustrating the voxelization result can be found in [reference needed]. Figure 3 .

[0116] Step A32: Extract near-ground voxels from the three-dimensional voxel grid according to a preset height threshold.

[0117] It should be noted that the preset height threshold can be set according to the relevant provisions in the "Design Code for 110kV~750kV Overhead Transmission Lines" (GB50545-2010), which stipulates that the minimum ground clearance requirement for 110kV transmission lines in non-residential areas is 6 meters, and the minimum ground clearance requirement for residential areas is 7 meters.

[0118] In practical implementation, a voxel space is established with d1 (to balance the accuracy and efficiency of the algorithm, the initial voxel size d1≤1m) as the voxel size. Any voxel S(c v r v h v If h exists v <2d 1， This voxel is considered a near-surface voxel and is denoted as:

[0119]

[0120] Step A33: Based on the first reference scale, perform an upward search on each near-ground voxel to obtain a connected voxel grid. Filter out the connected voxel grids in the three-dimensional voxel grid to obtain a reference voxel grid.

[0121] Understandably, the first and second reference scales can be different voxel sizes.

[0122] It should be noted that the connected voxel grid obtained by searching upwards for each near-ground voxel based on the first reference scale can be obtained by using near-ground voxels as seed voxels and connectivity as a constraint, and adopting a bottom-up search strategy.

[0123] In practical implementation, for any near-ground voxel D i (c, r, h) i ), search vertically upwards according to the voxel index, if D exists i+1(c, r, h) i+1 ), and the number of point clouds within a voxel B i+1 If ≠0, then D is considered to be... i With D i+1 The vertical direction is connected, so continue searching upwards sequentially for D. i+2 (c, r, h) i+2 ), D i+3 (c, r, h) i+3 ).

[0124] It is worth noting that since targets such as vegetation and buildings are directly connected to the ground in vertical space and have connectivity, while targets such as electric field lines are not connected to other targets in vertical space, the connected regions with near-ground voxels as seed voxels can be identified as non-electric field line points.

[0125] It should be noted that the near-surface voxel is defined as h. v A set of voxels <2d1 (where d1 can be understood as the first reference scale), but in h v In regions >2d1, numerous continuous interference objects still exist in the vertical space. Within a space of initial voxel size d1, due to the excessively large voxels, electric field line targets and non-electric field line targets are grouped into the same voxel. Expanding the near-ground voxel definition range may reduce the integrity of electric field line extraction. A schematic diagram of the vertical connectivity analysis results can be found in [reference needed]. Figure 4 .

[0126] Step A34: Extract the vertical continuous point cloud from the reference voxel grid based on a second reference scale, where the second reference scale is smaller than the first reference scale.

[0127] It should be noted that extracting the vertical continuous point cloud from the reference voxel grid based on the second reference scale can be understood as resetting a standard smaller than the first reference scale, performing vertical connectivity analysis on the voxel grid again, and then performing vertical connectivity analysis on the voxel grid.

[0128] It should be emphasized that, considering the lack of vertical continuity in the electric field lines, further vertical continuity constraints are designed in the voxel space of d2 (where d2 can be understood as the second reference scale): firstly, starting from the lowest level h of the voxel space... v Starting from 0, traverse all voxels layer by layer, searching vertically upwards according to the voxel index. Let any voxel V(c) be denoted as V(c). v r v The continuity of 0 in the vertical direction is L. v The initial value is 0. If there exists a voxel W(c v r v ,1), and the number of point clouds within a voxel B W If L ≠ 0, then voxel V and voxel W are considered to be perpendicularly adjacent and continuous.v =1; secondly, continue indexing vertically upwards, if consecutively existing (c v r v ,2),(c v r v ,3),...,L v The count increases until the continuity disappears.

[0129] Step A35: Remove vertical continuous point clouds in the reference voxel grid whose continuous count is greater than a preset continuous threshold to obtain reference power line point cloud data.

[0130] It should be emphasized that if continuity L v If the value is greater than a certain threshold (i.e., a preset continuity threshold), it proves that voxel V has a large continuity in vertical space and does not meet the characteristics of electric field lines; otherwise, it may be a candidate voxel for electric field lines. A schematic diagram of the vertical continuity analysis results can be found in the provided image. Figure 5 .

[0131] It should be emphasized that in areas where the main objects are near the ground and vertical structures (buildings) connected to the ground, large-scale voxels can be used for efficient removal. However, in areas close to power lines, since non-power line targets and power line targets are at similar heights in the voxel space, using larger voxels would classify them into the same voxel. Using smaller voxel sizes can further filter out interference while ensuring the integrity of the power lines.

[0132] It is worth noting that in this embodiment, two levels of voxels are set (where d1 > d2). However, the number of voxel scales can be adjusted appropriately according to the differences in different usage scenarios. If the power line is too high above the ground, multiple levels of voxels can be set.

[0133] In this embodiment, the computational complexity is reduced by converting discrete point clouds into three-dimensional voxel meshes, while highlighting linear features. Through bottom-up connectivity analysis, large areas of vegetation and complete buildings can be accurately marked as non-power line targets and filtered out. Multi-scale filtering is performed by taking into account efficiency, macroscopic integrity and fineness, and local purity. Furthermore, "vertical discontinuity" is used for fine filtering to remove these vertically continuous structures that do not conform to the morphological characteristics of power lines, thereby improving the accuracy of the final power line point cloud data.

[0134] The above are merely feasible implementations of step S30 provided in this embodiment. This embodiment does not specifically limit the specific implementation of step S30.

[0135] Step S40: Extract the point cloud based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data.

[0136] It should be understood that the reference power line point cloud data still contains hollow tree structures, some residual walls, and a small number of discrete noise points. Accurately extracting power line point clouds from these complex backgrounds can utilize the density characteristics and geometric features of point cloud data to effectively distinguish and remove residual non-power line points, thereby improving the accuracy and completeness of power line point cloud extraction.

[0137] It should be noted that point cloud extraction based on density features can be understood as performing DBSCAN clustering on reference power line point cloud data. The density similarity of power line point clouds can be used to cluster them into multiple clusters to achieve preliminary extraction of power lines, while removing sparse noise points. Furthermore, candidate clusters of power line point clouds are distinguished based on their spatial distribution characteristics (i.e., density features), and principal component analysis (PCA) is performed to construct multi-stage linear feature enhancement constraints to achieve accurate extraction of power line clusters.

[0138] It should be noted that point cloud extraction based on geometric features can be achieved by using principal component analysis to extract the feature values ​​of each cluster of point clouds, and then filtering and extracting the point cloud data based on these feature values ​​to obtain power line point cloud data.

[0139] It should be noted that the complex urban environment causes power lines to interact and interfere with each other, intertwining with trees, buildings, and other objects, posing a significant challenge to the accurate extraction of power lines. Existing MLS methods still suffer from insufficient extraction accuracy and robustness when handling such complex scenarios. There is a need to improve the accuracy and completeness of power line extraction in complex urban environments. This embodiment effectively addresses the problem of power lines intertwining with trees, buildings, and other objects by combining the characteristics of urban vehicle-mounted LiDAR data, providing reliable technical support for urban power line inspection.

[0140] It is worth noting that, due to the complexity of the urban environment and the large amount of content contained in the point cloud data, the extraction of power line point clouds in this embodiment is carried out in several ways. First, to address the problem of the large and complex volume of the original point cloud data, a data preprocessing method with linear feature constraints in the horizontal space is designed to extract data containing power lines. Second, considering the floating characteristics of power lines, a multi-scale voxel continuity analysis method in the vertical space is designed to further extract candidate points for power lines. Third, considering the linear geometric characteristics of power lines, a DBSCAN clustering extraction algorithm with enhanced linear features is designed to achieve precise extraction of power line points in complex urban scenes.

[0141] This embodiment provides a method for extracting power line point clouds. Through multiple constraints such as linear features, vertical continuity, and local geometric consistency, it greatly suppresses various types of ground noise. Multi-scale voxel analysis and density-based clustering effectively avoid over-segmentation or accidental deletion of power line point clouds during the filtering process, ensuring that the extracted power lines are continuous and complete, and achieving more accurate power line point cloud extraction.

[0142] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 6 Step S40, the power line point cloud extraction method further includes steps S41~S44:

[0143] Step S41: Perform DBSCAN density spatial clustering on the reference power line point cloud data using preset clustering parameters to obtain multiple candidate point cloud clusters.

[0144] It should be noted that the preset clustering parameters include the neighborhood radius and the minimum number of points in the cluster. The neighborhood radius is a key parameter for clustering and can be set based on the spatial distribution characteristics of electric field lines. Generally, the value is 0.2-0.5 (determined by the point cloud density), and the minimum number of points (minPTs) is generally 3-5 (to ensure the local integrity of the power line) in order to reduce time complexity.

[0145] It should be noted that nearest neighbor search can be performed by dividing the point cloud data into k candidate clusters using a KD-Tree neighbor search. C={C k} (k=0, 1, 2, ..., n) and the noise point set E.

[0146] Step S42: Construct a covariance matrix based on the cluster centroids of the candidate point cloud clusters, and perform eigenvalue decomposition on the covariance matrix through principal component analysis to obtain the linearity of each candidate point cloud cluster.

[0147] It should be noted that, in order to further distinguish candidate clusters of power line point clouds, multi-stage linear feature enhancement constraints can be constructed by combining principal component analysis (PCA) to achieve accurate extraction of power line clusters. As the core method for point cloud geometric feature extraction, principal component analysis can represent the linearity of clusters through the eigenvalue decomposition of the covariance matrix, thereby achieving accurate discrimination of power lines.

[0148] It is worth noting that, in view of the limitations and the problem of false detection in the traditional principal component analysis (PCA) linear detection method, this embodiment adopts the DBSCAN clustering extraction algorithm with linear feature enhancement. Based on DBSCAN clustering, it combines PCA point cloud data to perform linear feature enhancement, thereby achieving more accurate and complete extraction of power line points and effectively improving the extraction accuracy.

[0149] In one feasible implementation, step S42 may include steps A421-A423:

[0150] Step A421: Construct a covariance matrix based on the cluster centroids of the candidate point cloud clusters, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues.

[0151] It should be noted that for any number containing N k Candidate clusters at points The covariance matrix can be constructed using the following formula:

[0152]

[0153] Where, μ k p is the cluster centroid. i k Candidate cluster C k The i-th point in the array.

[0154] Furthermore, by performing eigenvalue decomposition C on the covariance matrix... k v j k =λ j k v j k， Obtain the eigenvalues ​​in descending order and the corresponding feature vector v1 k v2 k v3 k .

[0155] It should be noted that the eigenvalue represents the amount of information in each eigenvector, while the eigenvector represents the direction of each principal component.

[0156] Step A422: Obtain the distribution relationship of candidate point cloud clusters based on the magnitude relationship between the feature values.

[0157] It should be noted that the magnitude relationship between the three eigenvalues ​​can represent the geometric features of the candidate cluster: if λ1 k ≈λ2 k ≈λ3 k This indicates that these clusters have a discrete distribution; if λ1 k ,λ2 k 》λ3 k This indicates that these clusters are distributed in a planar manner; if λ1 k 》λ2 k ,λ3 k This indicates that these clusters have a linear distribution.

[0158] Step A423: When the distribution relationship is linear, calculate the linearity of the candidate point cloud cluster based on the feature value.

[0159] It should be noted that the following formula can be used as a reference for determining the characteristics of a linear distribution:

[0160]

[0161] Here, L represents the linearity of the cluster; the larger the value, the higher the linearity of the cluster.

[0162] In this embodiment, by combining density clustering (DBSCAN) with geometric feature analysis, it is possible to accurately distinguish between power lines and various types of interference objects in complex scenes. Through the dual constraints of feature value relationship discrimination and linear quantification threshold, residual planar, discrete, and weakly linear interference objects can be strictly filtered out. The final output power line point cloud has extremely high purity, ensuring the integrity and accuracy of the power line point cloud extraction task.

[0163] The above are merely feasible implementations of step S42 provided in this embodiment. This embodiment does not specifically limit the specific implementation of step S42.

[0164] Step S43: Filter out candidate point cloud clusters with linearity less than the first preset linearity threshold to obtain reference candidate point cloud clusters.

[0165] It should be noted that the linearity constraint can be preset to L1 for candidate cluster C. k If its linearity index L < L1, it is determined to be a nonlinear structure (such as wall fragments or tree clusters); otherwise, it is a reference candidate point cloud cluster that meets the linear characteristics.

[0166] Step S44: Perform directional consistency constraint filtering and linear enhancement constraint filtering on the reference candidate point cloud cluster to obtain power line point cloud data.

[0167] It should be noted that non-electric line clusters also exist in the candidate electric line clusters in the reference candidate point cloud clusters. For example, there are scattered wall structures that cannot be filtered out by the multi-scale voxel continuity constraint in vertical space, mixed clusters of electric lines and trees, and small tree clusters with certain linear characteristics (such as single slender branches). These structures are used to further distinguish between the design direction consistency constraint and the linear enhancement constraint in this paper.

[0168] In one feasible implementation, step S44 may include steps A441 to A444:

[0169] Step A441: Extract the feature vector and feature value of the reference candidate point cloud cluster, and obtain the main direction of the reference candidate point cloud cluster based on the feature vector and feature value.

[0170] It should be noted that due to the characteristics of DBSCAN clustering, it may cluster power line point clouds and tree point clouds with similar density distributions into the same cluster, which may lead to a decrease in the accuracy of the extraction results.

[0171] Each candidate point cloud cluster has three eigenvalues, each with varying values, the largest being λ1. k The corresponding feature vector v1k This is the main direction of the cluster.

[0172] Step A442: Construct a local neighborhood based on the point clouds in the reference candidate point cloud cluster and a preset search radius, and obtain the local normal direction of each point cloud in the reference candidate point cloud cluster based on the local neighborhood.

[0173] It should be noted that for any point p in the cluster i k A neighborhood is constructed with radius R to obtain a local neighborhood. The normals of the point cloud in the local neighborhood are estimated sequentially using the covariance matrix to obtain the local normal directions.

[0174] In detail, the normal to a point on the electric field line should be approximately perpendicular to the principal direction and consistent with the normal directions of other points in the neighborhood. The orthogonality between its normal and the principal direction of the cluster, as well as its consistency with the normal directions of other points in the neighborhood, should be calculated separately. The specific calculation method can be found in the following formulas:

[0175]

[0176]

[0177] in, ( ) are points and neighboring points The normal unit vector is N, where N is the number of points in the neighborhood.

[0178] Step A443: Remove reference candidate point cloud clusters whose main direction and local normal direction are inconsistent to obtain the point cloud cluster to be constrained.

[0179] It should be noted that if the orthogonality between the principal direction and the local normal direction is lower than a set threshold... If the normal at that point is approximately perpendicular to the principal direction, then the normal at that point is considered to be perpendicular to the principal direction; if the consistency value between the principal direction and the local normal direction exceeds a set threshold... If the point is consistent with its neighboring points in terms of direction, it is considered a candidate point for the power line only when both of the above constraints are satisfied.

[0180] In practical implementation, a schematic diagram illustrating the effect of calculating the orthogonality between its normal and the cluster principal direction, as well as its consistency with the normal directions of other points in the neighborhood, can be found here. Figure 7 , Figure 7 The red area represents point clouds of electric field lines, and the green area represents point clouds of tree clusters. The image effectively distinguishes between electric field lines and tree clusters within the same cluster.

[0181] Step A444: Perform linear enhanced constraint filtering on the point cloud cluster to be constrained to obtain power line point cloud data.

[0182] It should be noted that, specifically, the cluster length and cluster linearity of the point cloud cluster to be constrained can be calculated first;

[0183] Point cloud clusters to be constrained are removed if their cluster length is less than a preset length threshold and their cluster linearity is less than or equal to a second preset linearity threshold, resulting in point cloud clusters to be density-constrained, where the second preset linearity threshold is greater than the first preset linearity threshold; the spatial density of the point cloud clusters to be density-constrained is calculated; point cloud clusters to be constrained are removed if their spatial density is greater than or equal to a preset density threshold and their cluster linearity is less than or equal to a third preset linearity threshold, resulting in power line point cloud data where the third preset linearity threshold is greater than the first preset linearity threshold.

[0184] It should be noted that the linearity of clusters is improved by directional consistency constraints, but some non-electric line clusters with high linearity still cannot be directly distinguished by linearity judgment. Therefore, this invention further designs linear enhancement constraints for different non-electric line clusters.

[0185] Among them, linear enhancement constraint filtering can be used to further design length constraints for small tree clusters (such as single slender branches) with certain linear characteristics, and calculate the projection length of any cluster along the main direction. Set length threshold ,Will Clusters are defined as small clusters (often tree fragments) and other non-power line interference. However, to ensure the integrity of the algorithm and avoid accidental deletion, a linearity constraint is set. ( For small clusters, perform linear enhancement judgment: only retain Small clusters with low linearity can be removed, and small non-power line point clouds such as tree points can be effectively removed by length constraints and linear enhancement constraints.

[0186] Specifically, for the remaining wall structure, this paper further designs density constraints. Specifically, this is achieved by calculating the point density of each cluster. To achieve initial differentiation, it is defined as the number of points within a cluster. With space volume The ratio ( Considering the slender shape of the electric field line, its spatial volume is approximated as that of a cylinder. (Where l is the principal axis length of the cluster, calculated by PCA principal direction projection; r is the average cross-sectional radius, estimated by the standard deviation of the point cloud distribution perpendicular to the principal axis direction). Firstly, the clusters of crossing power lines exhibit lower density due to their increased volume (spatial distribution in the crossing region), allowing for initial differentiation from the wall. While wall point clouds may exhibit similar density values ​​to some power lines (such as single dense power lines) (walls may show higher density values ​​due to the large number of points, while power lines may show higher density values ​​due to their small size), the density of wall point clouds may be similar to that of some power lines (such as single dense power lines). However, the two differ significantly in their geometric characteristics: the point cloud of the wall is essentially a planar structure, while the electric field line has strong linear characteristics.

[0187] To further explain, a linear enhancement judgment can be added to the density constraint: for clusters with similar density values, a linearity constraint can be applied. Further screening ( By employing a dual constraint of initial segmentation based on density features and refined verification with high linearity, the system effectively distinguishes between power lines and residual wall point clouds, particularly addressing the density confusion issue between single dense power lines and walls.

[0188] In practical implementation, a complete flowchart for power line point cloud extraction can be found by referring to [the provided text]. Figure 10 The extraction process is divided into three stages. The first row of three images represents the continuity constraints in the horizontal space, namely the original point cloud diagram, the ground point segmentation diagram, and the continuity constraint diagram. The second row of three images represents the multi-scale voxel continuity constraints in the vertical space, namely the point cloud voxelization diagram, the connectivity constraint diagram, and the continuity constraint diagram. The third row of three images represents the DBSCAN clustering extraction based on linear enhancement, namely the DBSCAN clustering diagram, the linear enhancement diagram based on geometric feature constraints, and the electric field line extraction result diagram.

[0189] In this embodiment, the overall extension direction of the candidate cluster is quantified by extracting the main direction of the point cloud cluster. The local normal direction of each point in the point cloud cluster is calculated to characterize the microscopic geometric structure inside the point cloud cluster. By calculating the normal direction in the local neighborhood of each point, the orientation of the local surface at the location of the point can be reflected for point cloud filtering, accurately identifying and filtering out "pseudo-linear" interference, and enhancing the geometric purity of the extraction results.

[0190] The above are merely feasible implementations of step S44 provided in this embodiment. This embodiment does not specifically limit the specific implementation of step S44.

[0191] This embodiment provides a method for extracting power line point clouds. Based on the continuous and uniform density of the power line point cloud distribution, density clustering is performed to initially divide the chaotic reference point cloud into discrete candidate clusters and remove sparse noise. Principal component analysis is then performed to quantify the linearity of each candidate cluster. Point cloud screening is conducted based on a preset threshold, and fine filtering is performed based on the structural consistency of each point cloud. This accurately identifies and eliminates "pseudo-linear" interference. Finally, linear enhancement and density constraint filtering ensure that the output power line point cloud not only has significant linear features in geometric shape but also has a continuous and complete spatial distribution. Ultimately, this achieves accurate, complete, and reliable extraction of power line point clouds in complex scenes.

[0192] To better implement the power line point cloud extraction method in this embodiment of the invention, based on the power line point cloud extraction method, correspondingly, as follows: Figure 9 As shown, this embodiment of the invention also provides a power line point cloud extraction device, the power line point cloud extraction device 900 comprising:

[0193] Module 901 is used to acquire urban environmental point cloud data;

[0194] Filtering module 902 is used to perform point cloud filtering on the urban environment point cloud data according to the cloth filtering and horizontal linear feature constraints to obtain initial power line point cloud data.

[0195] The filtering module 902 is also used to perform multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data.

[0196] The power line point cloud data extraction module 903 is used to extract point cloud data based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data.

[0197] The power line point cloud extraction device 900 provided in the above embodiments can realize the technical solutions described in the above power line point cloud extraction method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above power line point cloud extraction method embodiments, and will not be repeated here.

[0198] like Figure 10 As shown, the present invention also provides an electronic device 1000. The electronic device 1000 includes a processor 1001, a memory 1002, and a display 1003. Figure 10 Only some components of the electronic device 1000 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0199] In some embodiments, memory 1002 may be an internal storage unit of electronic device 1000, such as a hard disk or memory of electronic device 1000. In other embodiments, memory 1002 may also be an external storage device of electronic device 1000, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 1000.

[0200] Furthermore, the memory 1002 may include both internal storage units of the electronic device 1000 and external storage devices. The memory 1002 is used to store application software and various types of data installed on the electronic device 1000.

[0201] In some embodiments, processor 1001 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 1002 or process data, such as the power line point cloud extraction method of the present invention.

[0202] In some embodiments, display 1003 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 1003 is used to display information from electronic device 1000 and to display a visual user interface. Components 1001-1003 of electronic device 1000 communicate with each other via a system bus.

[0203] In some embodiments of the present invention, when the processor 1001 executes the power line point cloud extraction program in the memory 1002, the following steps can be implemented:

[0204] Acquire urban environmental point cloud data; perform point cloud filtering on the urban environmental point cloud data based on cloth filtering and horizontal linear feature constraints to obtain initial power line point cloud data; perform multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data; extract point cloud data based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data.

[0205] It should be understood that when the processor 1001 executes the power line point cloud extraction program in the memory 1002, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0206] Furthermore, the embodiments of the present invention do not specifically limit the type of the electronic device 1000 mentioned. The electronic device 1000 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 1000 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0207] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the power line point cloud extraction method provided by the methods described above, the method comprising:

[0208] Acquire urban environmental point cloud data; perform point cloud filtering on the urban environmental point cloud data based on cloth filtering and horizontal linear feature constraints to obtain initial power line point cloud data; perform multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data; extract point cloud data based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data.

[0209] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0210] The above provides a detailed description of the power line point cloud extraction method provided by the present invention. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for extracting power line point clouds, characterized in that, include: Acquire urban environmental point cloud data; The process of filtering urban environmental point cloud data using cloth filtering and horizontal linear feature constraints to obtain initial power line point cloud data includes: performing cloth particle sedimentation simulation and terrain iterative convergence on the urban environmental point cloud data using a cloth simulation filtering algorithm to obtain ground point cloud data; removing ground point cloud data from the urban environmental point cloud data to obtain point cloud data to be filtered; dividing the point cloud data to be filtered into multiple point cloud clusters using Euclidean clustering; dividing the point cloud according to the spatial length of each point cloud cluster to obtain filtered point cloud data and interfering point cloud data; performing linear feature recall on the interfering point cloud using principal component analysis with horizontal linear constraints to obtain recalled point cloud data; and using the filtered point cloud data and the recalled point cloud data as the initial power line point cloud data. The initial power line point cloud data is subjected to multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering to obtain reference power line point cloud data. This includes: converting the initial power line point cloud data into a three-dimensional voxel mesh, where each three-dimensional voxel mesh includes spatial location coordinates, a unique index value, and the number of points contained within it; extracting near-ground voxels from the three-dimensional voxel mesh according to a preset height threshold; performing an upward search on each near-ground voxel based on a first reference scale to obtain a connected voxel mesh; filtering out the connected voxel meshes in the three-dimensional voxel mesh to obtain a reference voxel mesh; extracting vertically continuous point clouds from the reference voxel mesh based on a second reference scale, where the second reference scale is smaller than the first reference scale; and removing vertically continuous point clouds from the reference voxel mesh with a continuous count greater than a preset continuous threshold to obtain the reference power line point cloud data. Point cloud extraction is performed based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data.

2. The power line point cloud extraction method as described in claim 1, characterized in that, The step of extracting point cloud data based on the density characteristics and geometric features of the reference power line point cloud data to obtain power line point cloud data includes: By performing DBSCAN density spatial clustering on the reference power line point cloud data using preset clustering parameters, multiple candidate point cloud clusters are obtained. The covariance matrix is ​​constructed based on the cluster centroids of the candidate point cloud clusters. The linearity of each candidate point cloud cluster is obtained by performing eigenvalue decomposition on the covariance matrix through principal component analysis. Filter out candidate point cloud clusters whose linearity is less than the first preset linearity threshold to obtain reference candidate point cloud clusters; The reference candidate point cloud clusters are subjected to directional consistency constraint filtering and linear enhancement constraint filtering to obtain power line point cloud data.

3. The power line point cloud extraction method as described in claim 2, characterized in that, The process of constructing a covariance matrix based on the cluster centroids of candidate point cloud clusters, and then performing eigenvalue decomposition on the covariance matrix using principal component analysis to obtain the linearity of each candidate point cloud cluster includes: Construct a covariance matrix based on the cluster centroids of the candidate point cloud clusters, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues; The distribution relationship of candidate point cloud clusters is obtained based on the magnitude relationship between the feature values; When the distribution relationship is linear, the linearity of the candidate point cloud cluster is calculated based on the feature value.

4. The power line point cloud extraction method as described in claim 2, characterized in that, The process of performing directional consistency constraint filtering and linear enhancement constraint filtering on the reference candidate point cloud clusters to obtain power line point cloud data includes: Extract the feature vectors and feature values ​​of the reference candidate point cloud clusters, and obtain the main direction of the reference candidate point cloud clusters based on the feature vectors and feature values; A local neighborhood is constructed based on the point clouds in the reference candidate point cloud cluster and a preset search radius, and the local normal direction of each point cloud in the reference candidate point cloud cluster is obtained based on the local neighborhood. By eliminating reference candidate point cloud clusters whose main direction and local normal direction are inconsistent, the point cloud cluster to be constrained is obtained. Linear enhanced constraint filtering is applied to the point cloud cluster to be constrained to obtain power line point cloud data.

5. The power line point cloud extraction method as described in claim 4, characterized in that, The process of performing linear enhanced constraint filtering on the point cloud cluster to obtain power line point cloud data includes: Calculate the cluster length and cluster linearity of the point cloud cluster to be constrained; Remove the point cloud clusters to be constrained whose cluster length is less than a preset length threshold and whose cluster linearity is less than or equal to a second preset linearity threshold, and obtain the point cloud clusters to be density constrained, wherein the second preset linearity threshold is greater than the first preset linearity threshold. Calculate the spatial density of the point cloud cluster to be density constrained; Remove point cloud clusters from the density-constrained point cloud clusters whose spatial density is greater than or equal to a preset density threshold and whose cluster linearity is less than or equal to a third preset linearity threshold, to obtain power line point cloud data where the third preset linearity threshold is greater than the first preset linearity threshold.

6. A power line point cloud extraction device, characterized in that, The power line point cloud extraction device includes: The acquisition module is used to acquire urban environmental point cloud data; The filtering module is used to filter the urban environment point cloud data according to cloth filtering and horizontal linear feature constraints to obtain initial power line point cloud data. This includes: performing cloth particle sedimentation simulation and terrain iterative convergence on the urban environment point cloud data using a cloth simulation filtering algorithm to obtain ground point cloud data; removing ground point cloud data from the urban environment point cloud data to obtain point cloud data to be filtered; dividing the point cloud data to be filtered into multiple point cloud clusters using Euclidean clustering; dividing the point cloud according to the spatial length of each point cloud cluster to obtain filtered point cloud data and interfering point clouds; performing linear feature recall on the interfering point clouds using principal component analysis with horizontal linear constraints to obtain recalled point cloud data; and using the filtered point cloud data and the recalled point cloud data as the initial power line point cloud data. The filtering module is further configured to perform multi-scale voxel connectivity analysis and vertical continuity constraint point cloud filtering on the initial power line point cloud data to obtain reference power line point cloud data, including: converting the initial power line point cloud data into a three-dimensional voxel mesh, each three-dimensional voxel mesh including spatial location coordinates, a unique index value, and the number of points contained within it; extracting near-ground voxels from the three-dimensional voxel mesh according to a preset height threshold; performing an upward search on each near-ground voxel based on a first reference scale to obtain a connected voxel mesh; filtering out the connected voxel meshes in the three-dimensional voxel mesh to obtain a reference voxel mesh; extracting vertically continuous point clouds from the reference voxel mesh based on a second reference scale, the second reference scale being smaller than the first reference scale; and removing vertically continuous point clouds from the reference voxel mesh with a continuous count greater than a preset continuous threshold to obtain reference power line point cloud data. The power line point cloud data extraction module is used to extract point cloud data based on the density characteristics and geometric features of the reference power line point cloud data.

7. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps of the power line point cloud extraction method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the power line point cloud extraction method as described in any one of claims 1 to 5.

Images