A method, system, device and storage medium for large-scale point cloud registration based on power grid transmission lines

Abstract

The application discloses a kind of based on power grid transmission line Large-scale point cloud registration method, system, equipment and storage medium, the method includes: collecting the point cloud data of transmission line under different visual angle, forms source point cloud set and target point cloud set;To source point cloud set and the point cloud data undersampling in target point cloud set, dynamically adjust the sampling granularity of different density area, obtain representative point cloud data from source point cloud set and target point cloud set;Extract point cloud feature points, accelerate neighborhood search by KD-Tree structure, and adaptively adjust neighborhood search radius;Geometric topology model is constructed;Based on improved RANSAC algorithm, rough registration is carried out, non-collinear feature points are selected, and transformation model is optimized;Improved ICP algorithm is used for fine registration, and the registration accuracy is optimized by introducing a weighting factor using the normal vector feature.The application can efficiently and accurately process large-scale transmission line point cloud data, and is particularly suitable for power inspection, three-dimensional modeling and maintenance scenarios.

Description

TECHNICAL FIELD

[0001] The present application relates to computer vision and three-dimensional point cloud processing technology, and particularly relates to a large-scale point cloud registration method, system, device and storage medium based on power transmission line. BACKGROUND

[0002] With the development of remote sensing technologies such as unmanned aerial vehicles and LiDARs, three-dimensional point clouds have been widely used in the inspection, maintenance and modeling of power transmission lines. Point cloud data can accurately describe the three-dimensional structure of devices such as towers and conductors in power transmission lines. However, due to the large scale and complex structure of point cloud data in power transmission lines, traditional point cloud registration methods have many shortcomings in this application.

[0003] The ICP (Iterative Closest Point) algorithm is a classic point cloud registration method, but it is prone to local optimal solution when dealing with large-scale point clouds, and the computational efficiency is low. Especially in the complex geometric structure of power transmission lines (such as the linear structure of conductors and the vertical features of towers), the ICP algorithm cannot fully utilize these global features, resulting in a decrease in registration accuracy and robustness. In addition, existing feature matching methods, such as octree downsampling, ISS feature point extraction, and direction histogram, although perform well in some scenarios, still face the following technical bottlenecks in complex power transmission line scenarios:

[0004] (1) Low computational efficiency of large-scale point clouds: The range of point cloud data in power transmission lines is wide, and the density is uneven. Traditional methods have a huge amount of calculation when dealing with these large-scale data, making it difficult to meet the requirements of practical applications.

[0005] (2) Insufficient local features: Existing feature extraction algorithms usually rely on local geometric information and cannot effectively utilize the global structural features in power transmission lines, which cannot ensure accurate matching of key geometric structures.

[0006] (3) Poor registration robustness: In complex scenarios, noise points, missing points and other situations can easily affect existing registration methods, leading to registration failure or reduced accuracy. SUMMARY

[0007] The purpose of the present application is to provide a large-scale point cloud registration method, system, device and storage medium based on power transmission lines, which can solve the problems of low computational efficiency, insufficient registration accuracy and poor robustness of existing point cloud registration methods in dealing with complex power transmission line scenarios.

[0008] Technical solution: A large-scale point cloud registration method based on power transmission lines according to the present application comprises:

[0009] Collect point cloud data of the power transmission line from different perspectives to form a source point cloud set P and a target point cloud set Q;

[0010] Adaptive octree downsampling is respectively performed on the point cloud data of the source point cloud set P and the target point cloud set Q, and the interval of the selected points in different density regions is dynamically adjusted, representative point cloud data is obtained from the source point cloud set P to generate a first point cloud data set, and representative point cloud data is obtained from the target point cloud set Q to generate a second point cloud data set;

[0011] Point cloud feature points are extracted from the first point cloud data set and the second point cloud data set respectively by using an improved ISS algorithm, the neighborhood search is accelerated by using a KD-Tree structure in the process of extracting the point cloud feature points, and the neighborhood search radius is adaptively adjusted according to the local geometric characteristics of the point cloud, so that a feature point set suitable for complex scene structures is obtained;

[0012] Based on the feature point set, a direction histogram feature description model is constructed in combination with geometric topological information to generate a geometric topological model; the geometric topological model captures the spatial distribution direction of the feature points in the feature point set within the neighborhood, and reflects the local structural characteristics of the point cloud, thereby providing stable description features for subsequent registration;

[0013] An improved RANSAC algorithm is used for coarse registration of the feature point sets of the source point cloud set and the target point cloud set, non-collinear feature points meeting geometric relationships are selected, and a rigid transformation matrix is optimized; by minimizing the matching error, an initial alignment matrix of the source point cloud set to the target point cloud set is obtained;

[0014] Based on the coarse registration result, an improved ICP algorithm is used for fine registration; a weighting factor is introduced by using a normal vector feature and combined with KD-Tree accelerated search to gradually optimize the rigid transformation matrix to improve the registration accuracy, and finally an accurate registration result of the source point cloud set and the target point cloud set is output.

[0015] Further, adaptive octree downsampling is respectively performed on the point cloud data of the source point cloud set and the target point cloud set, the interval of the selected points in different density regions is dynamically adjusted, representative point cloud data is obtained from the source point cloud set to generate a first point cloud data set, and representative point cloud data is obtained from the target point cloud set to generate a second point cloud data set, including:

[0016] Initialize the octree: determine the three-dimensional space range occupied by the point cloud data in the source point cloud set P and the target point cloud set Q of the power transmission line, and define it as [x min ,x max ,y min ,y max ,z min ,z max ];

[0017] Divide octree node: for the root node of octree, it is recursively divided into eight child nodes using the following formula, taking x direction as an example, y and z directions are similar:

[0018]

[0019] Where i = 0, 1, 2, 3, 4, 5, 6, 7, representing the index of the eight child nodes in the x direction;

[0020] Adaptive subdivision: calculate the point cloud density p in each child node, the formula is as follows:

[0021]

[0022] Where N represents the number of points in the node; V represents the spatial volume of the node, for a cuboid-shaped node, its volume is:

[0023] V = (x max -x min )(y max -y min )(z max -z min )

[0024] Set the density threshold of point cloud data in the transmission tower area and the density threshold of point cloud data in the empty area, when p > p threshold , further subdivision is carried out on the node; wherein the density threshold of point cloud data in the transmission tower area is greater than the density threshold of point cloud data in the empty area;

[0025] Downsampling selection of representative points: after the construction of octree, the representative points are selected from each node;

[0026] Through the above steps, a part of representative points are selected from the original large-scale point cloud data, and adaptive octree downsampling is completed.

[0027] Further, after the construction of octree, the representative points are selected from each node, including:

[0028] Using the method of node barycenter point, for a point set {P i (x i ,y i ,z i )} in an octree node, the calculation formula of barycenter point P center is as follows:

[0029]

[0030] Where N represents the number of points in the node.

[0031] Further, the improved ISS algorithm is used to extract point cloud feature points from the first point cloud data set and the second point cloud data set respectively. In the process of extracting point cloud feature points, the KD-Tree structure is used to accelerate the neighborhood search, and the neighborhood search radius is adaptively adjusted according to the local geometric characteristics of the point cloud, so as to obtain a feature point set suitable for complex scene structure, including:

[0032] The KD-Tree structure is used to determine the local neighborhood. For each point P = (x, y, z) in the point cloud data, the local neighborhood point cloud set N(P) within a radius r is determined;

[0033] For each point P in the local neighborhood point cloud set N(P) q , a local coordinate system is established, and a search radius is set, wherein the neighborhood search radius of the point cloud near the power transmission tower is greater than that of the point cloud in the smooth area of the power transmission line; all points within the search radius are searched, and the weight of each point is calculated, and the formula is:

[0034]

[0035] In the formula, ω i represents the weight of the point p q , which is inversely proportional to the number of points in its local neighborhood; r represents the search radius; p i represents the points in the neighborhood, and the sparser the points in the sampling, the higher the weight;

[0036] The three-dimensional weighted covariance matrix of each point is calculated, and the formula is:

[0037]

[0038] The eigenvalues of each point are solved according to the three-dimensional weighted covariance matrix of each point , and are arranged in descending order, and the relationship between the three eigenvalues reflects the main features of the data points and their relative importance; two threshold values ε1 and ε2 are set for the point cloud near the power transmission tower with complex structure and the point cloud in the smooth area of the power transmission line respectively; if the query point p q meets the following conditions, it is regarded as an ISS feature point:

[0039]

[0040] Further, based on the feature point set, the direction histogram feature is described in combination with the geometric topological information, and a geometric topological model reflecting the local structure characteristics of the point cloud is constructed, including:

[0041] For each feature point P and its local neighborhood point cloud set N(P), the direction vector of the neighborhood point relative to the feature point P is calculated;

[0042] Let the neighborhood point be Q iN(P), direction vector The calculation method is That is, the direction vector is obtained by subtracting the coordinates of the feature points from the coordinates of the neighborhood points, so as to obtain a set of vector sets reflecting the spatial distribution direction of the neighborhood points relative to the feature points;

[0043] The vector set is divided according to the direction space, the number of direction vectors of the neighborhood points falling in each interval is counted, and a histogram is constructed;

[0044] Set the distance threshold d threshold And the angle threshold θ threshold If the distance between two neighborhood points is less than d threshold And the angle formed by them and the feature point is less than θ threshold , it is considered that there is a connection between the two points, the connected point pairs in the neighborhood are identified, and the geometric topology model is established, and the local topological information of the point cloud is quantized into a feature descriptor.

[0045] Further, the improved RANSAC algorithm is used to coarsely register the feature point sets of the source point cloud set and the target point cloud set, select the non-collinear feature points that meet the geometric relationship, and optimize the rigid transformation matrix; By minimizing the matching error, the initial alignment matrix of the source point cloud set to the target point cloud set is obtained, including:

[0046] Randomly select four corresponding data points in the source point cloud set P and the target point cloud set Q, respectively denoted as p1, p2, p3, p4 and q1, q2, q3, q4;

[0047] Solve the rigid transformation matrix H by least squares method, so that the source point cloud set is subjected to rigid rotation and translation transformation to obtain a new point cloud set P` = R c P+t c , Where R c is a rotation matrix, and t c is a translation vector;

[0048] Calculate the distance projection error of the new source point cloud set PP` of the source point cloud set P under the rigid transformation matrix H and the corresponding points of the target point cloud set Q;

[0049] If the distance projection error of the point pair is less than the set threshold error d t (d t =d 2 (p` i ,Hp i ), wherein d is the distance between the corresponding points, then the point is added to the sample inlier set I, and the matching score S i , that is, the ratio of the inliers to the total number of point clouds;

[0050] Set the number of cycles N and the matching score threshold S min, if the matching score of the point cloud is less than S min , then break the loop, otherwise repeat the above steps until the iteration number reaches N;

[0051] Count the number of inliers under different rigid transformation matrices, and take the rigid transformation matrix with the most inliers as the best rigid transformation matrix to optimize the rigid transformation matrix and obtain the source point cloud after rough registration.

[0052] Further, based on the rough registration result, an improved ICP algorithm is used for fine registration; a weighting factor is introduced using the normal vector feature and combined with KD-Tree accelerated search, and the rigid transformation matrix is gradually optimized to improve the registration accuracy, and finally the accurate registration result of the source point cloud set and the target point cloud set is output, including:

[0053] Input the source point cloud set P and the target point cloud set Q to be registered;

[0054] KD-Tree index is established for the source point cloud set P and the target point cloud set Q respectively; taking the source point cloud set as an example, for each point p, search the neighborhood point set N(P), and construct the covariance matrix C.

[0055]

[0056] where q is a point in the neighborhood, and the eigenvector corresponding to the smallest eigenvalue of the covariance matrix is the normal vector of the point p;

[0057] Using the normal vector feature of the source point cloud set and the target point cloud set, a weight value is given to each matched point pair in the registration process, and according to the normal vector information, a weighting factor ω is defined to optimize the matching, assuming and are the normal vectors of the corresponding point pair, and the weight ω can be expressed as:

[0058]

[0059] where σ represents a smoothing parameter;

[0060] Descriptors are constructed based on each point and its normal information to enhance the uniqueness of each feature point;

[0061] Traverse all points p in the source point cloud set P i , search for the nearest neighbor point q i of each p i in the target point cloud set Q using KD-Tree, and form a set s of matched point pairs (p i , q i );

[0062] Calculate the average distance d m and the standard deviation σ between the matched points in the matched point set s, if the distance d siThe distance satisfies the 3σ criterion, d si <d m If the value is +3σ, the matching pair is retained; otherwise, it is discarded.

[0063] According to d si Sort the matching point set s from smallest to largest, and select the top 70% of matching point pairs (p i ,q i The coordinates of ) are used to solve for the rotation matrix R and translation vector t using the weighted least squares method;

[0064] The rigid transformation matrix [RT] is obtained by solving the SVD method, which satisfies q i ≈R·p i +t;

[0065] Calculate the registration error evaluation function E for this registration. k (RT):

[0066]

[0067] Where, ω i Indicates the weighting factor; n represents the number of valid matching pairs; k represents the k-th transformation, if E is equal to or greater than E in the previous two transformations. k Difference D k =E k -E k-1 If the set threshold is met, the process is considered converged; otherwise, the iteration continues.

[0068] Based on the same inventive concept, the present invention provides a large-scale point cloud registration system based on power grid transmission lines, comprising:

[0069] The data acquisition module is used to collect point cloud data of the transmission line from different perspectives, forming source point cloud set P and target point cloud set Q;

[0070] The downsampling module is used to perform adaptive octree downsampling on the point cloud data of the source point cloud P and the target point cloud Q respectively, and dynamically adjust the interval of the selected points in different density regions. It obtains representative point cloud data from the source point cloud P to generate the first point cloud dataset, and obtains representative point cloud data from the target point cloud Q to generate the second point cloud dataset.

[0071] The feature point extraction module is used to extract point cloud feature points from the first point cloud dataset and the second point cloud dataset respectively using the improved ISS algorithm. During the extraction of point cloud feature points, the KD-Tree structure is used to accelerate the neighborhood search, and the neighborhood search radius is adaptively adjusted according to the local geometric characteristics of the point cloud, so as to obtain a feature point set that adapts to the complex scene structure.

[0072] A model construction module is configured to construct a direction histogram feature description model based on the set of feature points and geometric topology information, and generate a geometric topology model; the geometric topology model captures the spatial distribution direction of the feature points in the set of feature points in a neighborhood, and reflects the local structure characteristics of the point cloud, thereby providing stable description features for subsequent registration;

[0073] A coarse registration module is configured to perform coarse registration on the set of feature points of the source point cloud set and the target point cloud set by using an improved RANSAC algorithm, select non-collinear feature points that meet geometric relationships, and optimize a rigid transformation matrix; and by minimizing matching errors, an initial alignment matrix of the source point cloud set to the target point cloud set is obtained.

[0074] A fine registration module is configured to perform fine registration based on the coarse registration result by using an improved ICP algorithm; a weighting factor is introduced by using a normal vector feature and combined with KD-Tree accelerated search, and a rigid transformation matrix is gradually optimized to improve registration accuracy, and finally an accurate registration result of the source point cloud set and the target point cloud set is output.

[0075] Based on the same inventive concept, an electronic device based on a power transmission line large-scale point cloud registration method, comprising a processor and a memory, the memory stores computer instructions, and the processor is configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the electronic device implements the steps of the above-mentioned power transmission line large-scale point cloud registration method.

[0076] Based on the same inventive concept, a computer readable storage medium stores a computer program, and when the program is executed by a processor, the steps of the above-mentioned power transmission line large-scale point cloud registration method are implemented.

[0077] Advantages: Compared with the prior art, the significant technical effects of the present application are:

[0078] Adaptive octree downsampling strategy: the adaptive octree method is applied to intelligently and dynamically downsample point cloud data in different density regions. The problem that traditional fixed downsampling is difficult to adapt to the complex structure and uneven density of the power transmission line is solved. Efficient data compression is achieved while key structural information is retained, and the data processing efficiency is improved.

[0079] Improved feature point extraction based on ISS algorithm: by using KD-Tree to accelerate neighborhood search, and by adaptively adjusting the neighborhood search radius according to local geometric characteristics, ISS feature points with high representativeness are extracted. The problem that traditional feature extraction algorithms are inefficient on large and uneven density point cloud data and are difficult to capture key geometric structures is solved. Fast and stable feature point extraction is achieved, and the representativeness and uniformity of the registration points are enhanced.

[0080] The direction histogram is combined with a geometric topology model to construct, the geometric topology information of the feature point is combined, the spatial distribution of the neighborhood of the feature point is captured through the direction histogram, and the geometric topology model is formed. The problem that the existing description method has limited geometric information expression for the point cloud of a complex scene and is difficult to guarantee accuracy is solved. The integrity and reliability of the point cloud feature description are improved, and the registration accuracy in a complex scene is improved.

[0081] The improved RANSAC coarse registration and non-collinear feature point selection are improved, the improved RANSAC method is used for coarse registration, and non-collinear feature points are selected to improve the matching robustness. The problem that the traditional coarse registration algorithm is difficult to avoid collinearity in the random sampling process and leads to unstable registration is solved. The accuracy and efficiency of the coarse registration are improved, and reliable initial alignment is provided for subsequent fine registration.

[0082] The improved ICP fine registration of the normal vector is combined with the ICP algorithm of the normal vector for fine registration, and the feature description is strengthened by introducing a weighting factor. The problem that the traditional ICP algorithm is sensitive to noise and abnormal points and is easy to fall into local optimum is solved. The robustness and accuracy of the point cloud registration are enhanced, and the fine registration of large-scale data is suitable. BRIEF DESCRIPTION OF DRAWINGS

[0083] Figure 1 is a flowchart of a large-scale point cloud registration method based on a power grid transmission line disclosed by an embodiment of the application;

[0084] Figure 2 is a structural schematic diagram of a large-scale point cloud registration system based on a power grid transmission line disclosed by an embodiment of the application;

[0085] Figure 3 is a structural schematic diagram of a large-scale point cloud registration system based on a power grid transmission line disclosed by an embodiment of the application. DETAILED DESCRIPTION

[0086] The application will be described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art will understand that the purposes and advantages realized by the application are not limited to the above specific descriptions, and the above and other purposes realized by the application will be more clearly understood according to the following detailed description.

[0087] Those skilled in the art should understand that the exemplary components, systems and methods described in combination with the embodiments disclosed in the application can be realized in hardware, software or a combination of both. Whether to be realized in hardware or software depends on the specific application and design and condition of the technical solution. Professional technicians can use different methods to realize the functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0088] Reference to“an embodiment” or“the embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase“in one embodiment” or“in at least one embodiment” in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily all referring to a particular embodiment which is“preferred” over other embodiments. As used herein, the term“exemplary” or“illustrative” means serving as an example, instance or illustration. Any implementation described herein as“exemplary” or“illustrative” is not necessarily to be construed as preferred or advantageous over other implementations.

[0089] Embodiment 1

[0090] Please refer to Figure 1 , Figure 1 is a flowchart of a large-scale point cloud registration method based on power grid transmission lines disclosed by an embodiment of the application. Wherein, Figure 1 The large-scale point cloud registration method described is applied in a power system, such as three-dimensional modeling and inspection analysis of a transmission line, and the embodiment of the application is not limited. As shown in Figure 1 The large-scale point cloud registration method based on power grid transmission lines can include the following operations:

[0091] S1, data acquisition: acquiring point cloud data of a transmission line under different perspectives to form a source point cloud set P and a target point cloud set Q.

[0092] Obtain two groups of transmission line point cloud sets under different perspectives collected by unmanned aerial vehicle inspection, which are respectively a source point cloud set P and a target point cloud set Q.

[0093] S2, adaptive octree downsampling: respectively performing adaptive octree downsampling on point cloud data of the source point cloud set P and the target point cloud set Q, and dynamically adjusting the interval (i.e. sampling granularity) of selected points in different density regions, obtaining representative point cloud data from the source point cloud set P to generate a first point cloud data set, and obtaining representative point cloud data from the target point cloud set Q to generate a second point cloud data set. The specific steps are as follows:

[0094] S2.1, initialize octree: after obtaining the point cloud data to be processed, first determine the three-dimensional space range occupied by the point cloud data in the source point cloud set P and the target point cloud set Q of the transmission line, defined as [x min , x max , y min , y max , z min , z max ].

[0095] In this embodiment, the range of geographical coordinates during actual data collection is set. For example, if the power transmission line is located in an area extending 1000 meters in the east-west direction, 200 meters in the north-south direction, and 50 meters in height from the ground to the top of the power transmission tower, and the geographical coordinate origin is located at the lower left corner of the area, then x min = 0, x max = 1000, y min = 0, y max = 200, z min = 0, z max = 50.

[0096] S2.2, divide the octree node: for the root node of the octree, it is recursively divided into eight child nodes using the following formula (take the x direction as an example, the y and z directions are similar):

[0097]

[0098] where i = 0, 1, 2, 3, 4, 5, 6, 7, representing the index of the eight child nodes in the x direction.

[0099] S2.3, judge adaptive subdivision: calculate the point cloud density p in each child node, the formula is as follows:

[0100]

[0101] where N represents the number of points in the node; V represents the spatial volume of the node, for a cuboid-shaped node, its volume is:

[0102] V = (x max -x min )(y max -y min )(z max -z min )

[0103] Considering the characteristics of power transmission line point cloud data, the point cloud data is dense near the power transmission tower and sparse in the empty area. Set the density threshold of point cloud data in the power transmission tower area and the density threshold of point cloud data in the empty area; wherein the density threshold of point cloud data in the power transmission tower area is greater than the density threshold of point cloud data in the empty area;

[0104] In this embodiment, for the power transmission tower area, the density threshold of point cloud data is set to p threshold = 100 points per cubic meter, and the density threshold of point cloud data in the empty area is set to p threshol = 20 points per cubic meter. When p > p threshold , further subdivision is performed on the node.

[0105] S2.4, Select representative points after downsampling: After the construction of the octree is completed, a representative point is selected from each node. In this embodiment, the method of node barycenter point is adopted. For a point set in an octree node

[0106] {P i (x i ,y i ,z i )},the calculation formula of the barycenter point P center is as follows:

[0107]

[0108] Wherein, N represents the number of points in the node.

[0109] Through the above steps, a part of representative points can be selected from the original large-scale point cloud data, and adaptive octree downsampling is completed.

[0110] In the present application, when the point cloud is down-sampled, the point cloud is first divided into multiple octree structures, and the three-dimensional space is recursively subdivided to generate multiple sub-cubes. Considering the complexity and density difference of power transmission line point cloud data, an adaptive octree downsampling algorithm is adopted. High-density areas (such as power transmission towers) maintain high resolution, while low-density areas (such as open areas) reduce sampling points, ensuring the balance between local refinement and global simplification. This method overcomes the problem that fixed sampling scale cannot adapt to different point cloud densities.

[0111] After downsampling, the barycenter of the point cloud in each octree node is selected as the representative point to generate a representative point cloud dataset, ensuring that important geometric features are retained in the subsequent registration process.

[0112] S3, ISS feature point extraction: The improved ISS algorithm is used to extract point cloud feature points from the first point cloud dataset and the second point cloud dataset respectively. In the process of extracting point cloud feature points, the KD-Tree structure is used to accelerate the neighborhood search, and the neighborhood search radius is adaptively adjusted according to the local geometric characteristics of the point cloud, so as to obtain a feature point set that adapts to complex scene structures.

[0113] After octree downsampling, the point cloud data is still large, so it is necessary to use the ISS algorithm to extract feature points. The ISS algorithm selects feature points by calculating the eigenvalues of the covariance matrix of each point domain and selecting feature points through the linear relationship between the eigenvalues. To improve the efficiency of extracting feature points from large-scale point cloud data, the ISS algorithm is improved, combined with the KD-Tree structure for acceleration processing, and the feature point extraction parameters are adaptively adjusted according to the local geometric characteristics of the point cloud. The specific steps of step S3 are as follows:

[0114] S3.1, determine the local neighborhood by KD-Tree structure, for each point P=(x, y, z) in the point cloud data, determine its local neighborhood point cloud set N(P) within a radius r.

[0115] In this embodiment, starting from the root node, the distance d between the point represented by the root node and P is calculated, using the Euclidean distance formula where (x i ,y i ,z i ) is the point represented by the root node.

[0116] If d≤r, the point represented by the root node is added to the local neighborhood point cloud set N(P); if d≥r, and the root node is divided according to the x dimension, if x i , search in the left subtree, if x>x i , search in the right subtree. Until there is only one point in the subtree or the preset stopping condition is met, for example, the subtree depth reaches 10 layers.

[0117] S3.2, after obtaining the local neighborhood point cloud set N(P), for each point P q in the local neighborhood point cloud set N(P), a local coordinate system is established, and a search radius is set. Among them, the neighborhood search radius of the point cloud near the power transmission tower is greater than the neighborhood search radius of the point cloud in the smooth area of the power transmission line; in this embodiment, for the point cloud near the power transmission tower, due to the complex structure, the neighborhood search radius of the point cloud near the power transmission tower is set to r=0.8 meters, and in the relatively smooth area of the power transmission line, the neighborhood search radius of the point cloud in the smooth area of the power transmission line is set to r=0.3 meters. All points within the search radius are searched, and the weight of each point is calculated, the formula is:

[0118]

[0119] In the formula, ω i represents the weight of point p q proportional to the number of points in its local neighborhood; r represents the search radius; p i represents the points in the neighborhood, the sparser the points in the sampling, the higher the weight.

[0120] S3.3, calculate the three-dimensional weighted covariance matrix of each point, the formula is:

[0121]

[0122] S3.4, solve the eigenvalue of each point according to the three-dimensional weighted covariance matrix of each point And according to the arrangement from large to small, the relationship between the three characteristic values reflects the main features of the data points and their relative importance. Set two thresholds ε1 and ε2, for the point cloud near the structure complex of the power transmission tower, set ε1=0.3, ε2=0.1;For the relatively smooth area of the power transmission line, set ε1=0.6, ε2=0.2. If the query point p q As the ISS feature point, it needs to meet

[0123]

[0124] In the application, after adaptive octree downsampling, a point cloud dataset with reasonable data volume and complete structure is obtained. Based on this dataset, the ISS algorithm is used to extract feature points. In order to improve the efficiency of the algorithm, the KD-Tree structure is combined to accelerate the neighborhood search, and an adaptive neighborhood radius adjustment strategy is adopted. In the complex area of the power transmission tower, a larger neighborhood radius is used to capture details; In the relatively smooth area of the conductor, a smaller radius is used. The optimized ISS algorithm can adapt to the complex geometric structure of different regions in the power transmission line point cloud, and significantly improve the feature extraction speed and accuracy.

[0125] S4, the direction histogram (PFH) feature description: based on the feature point set, a direction histogram feature description model is constructed combined with geometric topology information to generate a geometric topology model; the geometric topology model captures the spatial distribution direction of the feature points in the feature point set in the neighborhood, reflects the local structure characteristics of the point cloud, and provides stable description features for subsequent registration. Specifically, the following includes:

[0126] S4.1, after determining the local neighborhood points in the last step, for each feature point P and its local neighborhood point cloud set N(P), the direction vector of the neighborhood point relative to the feature point P is calculated.

[0127] S4.2, let the neighborhood point be Q i ∈N(P), the direction vector The calculation method is That is, the direction vector is obtained by subtracting the coordinates of the feature point from the coordinates of the neighborhood point. In this way, a set of vector sets reflecting the spatial distribution direction of the neighborhood points relative to the feature points is obtained.

[0128] S4.3, the vector set is divided according to the direction space; in this embodiment, considering the structural characteristics and data accuracy requirements of the power grid transmission line, the 360° direction space is divided into 24 intervals, and each interval is 15°. Then, the number of direction vectors of the neighborhood points falling in each interval is counted to construct a histogram.

[0129] S4.4, in order to further combine the geometric topology information and comprehensively reflect the geometric structure in the complex scene, a distance threshold d threshold= 0.2 meters and an angle threshold threshol = 30°. If the distance between two neighboring points is less than d threshold and the angle between them and the feature point is less than θ threshold , then it is considered that there is a connection between the two points; then the connected point pairs in the neighborhood are identified, and a geometric topology model is established. Finally, the local topology information of the point cloud is quantized into a feature descriptor.

[0130] For example, for a neighborhood point Q and a feature point P, the distance d and the angle θ between them are calculated, and these distance and angle information is encoded, with the distance and angle mapped to a fixed numerical interval, combined into a new numerical value as the geometric topology descriptor of the neighborhood point.

[0131] In the present application, the feature description of the direction histogram PFH mainly includes three steps: neighborhood point selection, direction vector calculation, and histogram quantization. The spatial distribution direction of the points in the neighborhood is captured to reflect the local geometric shape. However, relying solely on the direction histogram to describe the spatial distribution of points is insufficient to fully reflect the geometric structure in complex scenes, especially in complex areas such as the connection nodes of power transmission line towers and the intersection points of conductors. Therefore, in combination with the geometric topology information, the present application establishes a geometric topology model in the neighborhood by analyzing the connection relationship between the feature point and the neighborhood points. The model reflects the linearity and branching characteristics of the local structure, especially at the conductor connection points and tower nodes. For example, in the point cloud of the power transmission line, for the connection point of the conductor, there is a clear connection relationship between the points in its neighborhood, and this connectivity can be determined by analyzing the distance and angle information between the points.

[0132] Then the local topology information of the point cloud is quantized into a feature descriptor as a supplement to the direction histogram. For each point in the neighborhood, the connection relationship between it and the feature point is calculated, including distance, angle distribution, etc., and these information is encoded into a geometric topology descriptor. It is used to reflect the local topology structure in the neighborhood, such as straight lines, curves or branches, etc., thereby further enriching the description of the feature points.

[0133] S5, improved RANSAC coarse registration: an improved RANSAC algorithm is used to perform coarse registration on the feature point sets of the source point cloud set and the target point cloud set, select non-collinear feature points that meet the geometric relationship, and optimize the rigid transformation matrix. By minimizing the matching error, an initial alignment matrix of the source point cloud set to the target point cloud set is obtained. The specific steps are as follows:

[0134] S5.1, randomly select 4 corresponding data points in the source point cloud set P and the target point cloud set Q, denoted as p1, p2, p3, p4 and q1, q2, q3, q4, respectively;

[0135] S5.2, solve the rigid transformation matrix H by least square method, so that the source point cloud is subjected to rigid rotation and translation transformation to obtain a new point cloud set P'=R c P+t c , wherein R c is a rotation matrix, and t c is a translation vector;

[0136] S5.3, calculate the distance projection error of the new source point cloud set PP' of the source point cloud set P under the rigid transformation matrix H and the corresponding points of the target point cloud set Q;

[0137] S5.4, if the distance projection error of the point pair is less than the set threshold error d t (d t =d 2 (p i , Hp i ), wherein d is the distance between the corresponding points, then the point is added to the sample inlier set I, and the matching score S i , that is, the ratio of the inliers to the total number of point clouds, is calculated;

[0138] S5.5, set the number of cycles N and the matching score threshold S min , if the matching score of the point cloud is less than S min , then the loop is exited, otherwise the above steps are repeated until the iteration number reaches N;

[0139] S5.6, count the number of sample inliers under different rigid transformation matrices, and take the rigid transformation matrix with the largest number of inliers as the best rigid transformation matrix, and optimize the rigid transformation matrix to obtain the source point cloud after coarse registration.

[0140] The point cloud data after the foregoing feature description has met the requirements of the RANSAC coarse registration on the point cloud data. The present application adopts a point cloud coarse registration method based on RANSAC, randomly samples the source point cloud, obtains the corresponding transformation model, tests each random transformation model, and repeatedly cycles the process until the optimal transformation model is selected as the final result. In order to improve the robustness and running efficiency of the algorithm, the present application is improved in the following aspects:

[0141] In the selection of corresponding points, 4 corresponding point pairs are selected, and it is ensured that the 4 points are not collinear. This selection of corresponding points can more comprehensively reflect the transformation relationship between the point clouds, and compared with the traditional selection method, the accuracy and stability of the registration are improved. Then the nearest neighbor points are searched by using the KD-Tree method, the point cloud data is quickly positioned to the points near the target point by dividing the point cloud data according to certain rules, the search speed is greatly improved, and the running efficiency of the entire coarse registration algorithm is improved.

[0142] S6, improved ICP fine registration: based on the coarse registration result, an improved ICP algorithm is used for fine registration; the normal vector feature is introduced to introduce a weighting factor and combine KD-Tree acceleration search, and the rigid body transformation matrix is gradually optimized to improve the registration accuracy, and finally the accurate registration result of the source point cloud set and the target point cloud set is output. The specific steps are as follows:

[0143] S6.1, input source point cloud set P and target point cloud set Q to be registered;

[0144] S6.2, respectively, KD-Tree index is established for source point cloud set P and target point cloud set Q. For each point p (take the source point cloud set as an example), search the neighborhood point set N(P). Then, the covariance matrix C is constructed:

[0145]

[0146] Wherein, q is a point in the neighborhood, and the smallest eigenvalue of the covariance matrix corresponds to the normal vector of the point p;

[0147] S6.3, using the normal vector features of the source point cloud set and the target point cloud set, a weighted value is given to each pair of matching points in the registration process (used to control the influence of each pair of points during registration, increase the weight of the normal vector close to the point pair), according to the normal vector information, define the weighting factor ω for optimization matching, assuming And The weight ω can be expressed as:

[0148]

[0149] Wherein, σ represents the smoothing parameter.

[0150] S6.4, based on each point and its normal information, a descriptor is constructed to enhance the uniqueness of each feature point.

[0151] In order to enhance the robustness of point pair matching, first calculate the covariance matrix M of a point with the help of point cloud normal i , assuming that the number of neighborhood points of a point is n, the components of the normal vector of the point in x, y and z directions are respectively n x , n y , n z , the calculation formula of the covariance matrix M i is:

[0152]

[0153] Calculate the weighted function f i of the point:

[0154] f i =10detM i -0.2(traceMi ) 2

[0155] Using a KD-Tree for neighborhood search, the search criteria for the nearest point are changed to the feature distance equation:

[0156]

[0157] Where c i This is a descriptor for the point cloud, and it can also be used to filter out matching point pairs based on feature distance.

[0158] S6.5, Traverse all points p in the source point cloud P. i Using KD-Tree to search for each p in the target point cloud Q i The nearest neighbor q i , forming matching point pairs (p i ,q i The set s;

[0159] S6.6 Calculate the average distance d between matching points in the matching point set s. m With respect to the standard deviation σ, if a point is relative to d si The distance satisfies the 3σ criterion, d si <d m If the value is +3σ, the matching pair is retained; otherwise, it is discarded.

[0160] S6.7, according to d si Sort the matching point set s from smallest to largest, and select the top 70% of matching point pairs (p i ,q i The coordinates of ) are used to solve for the rotation matrix R and translation vector t using the weighted least squares method;

[0161] S6.8. Solve for the rigid transformation matrix [RT] using the SVD method, satisfying q i ≈R·p i +t;

[0162] S6.9 Calculate the registration error evaluation function E for this registration. k (RT):

[0163]

[0164] Where, ω i Indicates the weighting factor; n represents the number of valid matching pairs; k represents the k-th transformation, if E is equal to or greater than E in the previous two transformations. k Difference D k =E k -E k-1 If the set threshold is met, the process is considered converged; otherwise, the iteration continues.

[0165] On the basis of coarse registration, the improved ICP fine registration method is used to further improve the registration accuracy. By introducing the weighted factor based on the normal feature of the point cloud and the new feature descriptor, the robustness of the point pair matching is enhanced, and the neighborhood search and registration error evaluation mechanism are optimized to realize more accurate point cloud registration.

[0166] Firstly, the covariance matrix M of a point is calculated by means of point cloud normal i Then, the weighted function f of a point is calculated i By calculation, the weighted values of all points are obtained, which will be used to adjust the importance of different points in the subsequent fine registration process. Then, according to the distance between the normal vector and the point, the neighborhood normal angle invariant property of the point cloud is calculated, and the coordinate value of the point is used to form the feature descriptor of the point. This method makes up for the defects of using only coordinate feature to describe the point cloud, and improves the speed and accuracy of feature calculation and construction. At the same time, the weighted factor f i is used to synchronize the weighting of all points, which increases the feature description difference between the feature-rich points and the feature-lacking points, and further improves the iterative convergence speed.

[0167] The technical scheme of the present application is aimed at large-scale complex power transmission line point cloud data. Firstly, the adaptive octree downsampling is used to reduce data redundancy and retain key geometric features. Then, the ISS feature point extraction algorithm is used in combination with KD-Tree acceleration to dynamically adjust the neighborhood radius and extract stable geometric feature points. In the feature description stage, the direction histogram is used in combination with geometric topological information to enhance the description ability of the feature points, and ensure good discrimination in complex structures (such as towers, conductor connection points, etc.). The coarse registration step realizes the fast alignment of feature points by using the improved RANSAC method, and combines with KD-Tree to optimize the search efficiency. In the fine registration process, the improved ICP algorithm is used to further improve the registration accuracy and robustness of the point cloud through weighted error calculation, dynamic step control and KD-Tree accelerated search. This method can efficiently and accurately process large-scale power transmission line point cloud data, and is particularly suitable for power inspection, three-dimensional modeling and maintenance scenarios.

[0168] The present application adopts adaptive octree downsampling, dynamic ISS feature point extraction, direction histogram (PFH) feature description combined with geometric topological information, improved RANSAC coarse registration and improved ICP fine registration technology, to form an efficient and accurate point cloud registration scheme, which is particularly suitable for three-dimensional modeling and inspection analysis of power transmission lines.

[0169] Example 2

[0170] Please refer to Figure 2 , Figure 2is a structural schematic diagram of a large-scale point cloud registration system based on a power grid transmission line disclosed by an embodiment of the application. The system can realize three-dimensional modeling and inspection analysis of the transmission line, and specifically comprises:

[0171] A data acquisition module is configured to acquire point cloud data of the transmission line under different perspectives, to form a source point cloud set P and a target point cloud set Q.

[0172] A downsampling module is configured to perform adaptive octree downsampling on the point cloud data of the source point cloud set P and the target point cloud set Q respectively, to dynamically adjust the interval of the selected points in different density regions, to obtain representative point cloud data from the source point cloud set P to generate a first point cloud data set, and to obtain representative point cloud data from the target point cloud set Q to generate a second point cloud data set.

[0173] A feature point extraction module is configured to extract point cloud feature points from the first point cloud data set and the second point cloud data set respectively by using an improved ISS algorithm. In the process of extracting the point cloud feature points, the KD-Tree structure is used to accelerate the neighborhood search, and the neighborhood search radius is adaptively adjusted according to the local geometric characteristics of the point cloud, so that a feature point set suitable for complex scene structures is obtained.

[0174] A model construction module is configured to construct a direction histogram feature description model based on the feature point set and in combination with geometric topological information, to generate a geometric topological model. The geometric topological model captures the spatial distribution direction of the feature points in the feature point set within the neighborhood, reflects the local structure characteristics of the point cloud, and provides stable description features for subsequent registration.

[0175] A coarse registration module is configured to perform coarse registration on the feature point sets of the source point cloud set and the target point cloud set by using an improved RANSAC algorithm, to select non-collinear feature points that meet the geometric relationship, and to optimize the rigid body transformation matrix. By minimizing the matching error, an initial alignment matrix of the source point cloud set to the target point cloud set is obtained.

[0176] A fine registration module is configured to perform fine registration by using an improved ICP algorithm based on the coarse registration result. By introducing a weighting factor using the normal vector feature and combining KD-Tree accelerated search, the rigid body transformation matrix is gradually optimized to improve the registration accuracy, and finally the accurate registration result of the source point cloud set and the target point cloud set is output.

[0177] In an optional embodiment, the large-scale point cloud registration method based on power grid transmission line includes: a) collecting point cloud data of the transmission line at different angles to form a source point cloud set P and a target point cloud set Q; b) using an adaptive octree to downsample the point cloud data, dynamically adjusting the sampling granularity of different density regions; c) using an improved ISS algorithm to extract point cloud feature points, accelerating neighborhood search through KD-Tree and adaptively adjusting the neighborhood radius; d) combining geometric topology information to describe the direction histogram (PFH) feature, and constructing a geometric topology model of the local structure of the point cloud; e) performing coarse registration based on an improved RANSAC algorithm, selecting four non-collinear feature points, and optimizing the transformation model; and f) performing fine registration using an improved ICP algorithm, and introducing a weighting factor using the normal vector feature to optimize the registration accuracy.

[0178] Embodiment 3

[0179] Referring to Figure 3 , Figure 3 is a structural schematic diagram of a large-scale point cloud registration system based on a power grid transmission line according to an embodiment of the present application. In the figure, Figure 3 The device described can be applied to a power system, such as three-dimensional modeling and inspection analysis of a transmission line, and the like, and the present application is not limited thereto.

[0180] As Figure 3 shown, the device can include a processor and a memory, the memory storing computer instructions, and the processor being configured to execute the computer instructions stored in the memory, so that the electronic device implements the steps of the method according to the above embodiment when the computer instructions are executed by the processor, and achieves the same technical effects as the above method.

[0181] The memory can include a computer system readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device can further include other removable / non-removable, volatile / non-volatile computer system storage media. For example, the memory can be used to read and write to a non-removable, non-volatile magnetic media, commonly referred to as a "hard disk drive". Programs / utilities with a set of (at least one) program modules can be stored in, for example, the memory, such program modules including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof can include the implementation of a network environment. The program modules generally perform the functions and / or methods described in the embodiments of the present application.

[0182] The processor performs various functional applications and data processing by running the programs stored in the memory, such as implementing the method provided by the first embodiment of the present application.

[0183] Embodiment 4

[0184] The embodiment 4 of the present application further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps of the method in the above embodiment and achieve the same technical effects.

[0185] The computer readable storage medium of the embodiment of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device.

[0186] The computer readable signal medium can include a data signal propagated in a baseband or propagated as a carrier wave in a propagated data signal, in which the computer readable program code is embodied. Such propagated data signal can take many forms, including but not limited to electro-magnetic, optical or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport program for use by or in connection with an instruction execution system, apparatus or device.

[0187] The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire line, optical cable, RF, etc., or any suitable combination thereof.

[0188] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0189] Of course, the storage medium provided by the embodiments of the present application includes computer executable instructions, and the computer executable instructions are not limited to the method operations described above, but can also perform related operations in the method provided by any embodiment of the present application.

[0190] The above specific embodiments further illustrate the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only for specific embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.​

Claims

1. A method for large-scale point cloud registration based on power grid transmission lines, characterized in that, Comprise: The point cloud data of the power transmission line under different visual angles is collected to form a source point cloud set and a target point cloud set ​ Adaptive octree down-sampling is respectively performed on point cloud data of a source point cloud set and a target point cloud set , and intervals of selected points in different density regions are dynamically adjusted, representative point cloud data is obtained from the source point cloud set , a first point cloud data set is generated, representative point cloud data is obtained from the target point cloud set , and a second point cloud data set is generated; Utilizing the improved The algorithm extracts point cloud feature points from the first and second point cloud datasets respectively. During the point cloud feature point extraction process, it uses... The structure accelerates neighborhood search and adaptively adjusts the neighborhood search radius based on the local geometric characteristics of the point cloud, thereby obtaining a set of feature points that adapt to complex scene structures. Based on the feature point set, a direction histogram feature description model is constructed in combination with geometric topology information to generate a geometric topology model; the geometric topology model captures the spatial distribution direction of the feature points in the feature point set within the neighborhood, reflects the local structure characteristics of the point cloud, and provides stable description features for subsequent registration; An improved Algorithm is used to roughly register the feature point sets of the source point cloud set and the target point cloud set, select non-collinear feature points that meet geometric relations, and optimize the rigid transformation matrix. By minimizing the matching error, an initial alignment matrix of the source point cloud set to the target point cloud set is obtained, comprising: In the source point cloud set and the target point cloud set , 4 corresponding data points are randomly selected, respectively denoted as and ; Solving the rigid body transformation matrix by least square method So that the source point cloud is subjected to a rigid rotation translation transformation to obtain a new point cloud set Wherein is a rotation matrix, is a translation vector; A computed source point cloud set Under a rigid body transformation matrix A new source point cloud set A target point cloud set The distance projection error of corresponding points; If the distance projection error of the point pair is less than a set threshold error , wherein, is the distance between the corresponding points, the point is added to the set of inliers I and a matching score is calculated i.e. the ratio of inliers to the total number of points in the point cloud; Setting the number of cycles and a matching score threshold If the matching score of the point cloud is less than then break the loop, otherwise repeat the above steps until the number of iterations reaches ; Counting the number of inliers under different rigid transformation matrices, taking the rigid transformation matrix with the most inliers as the best rigid transformation matrix, optimizing the rigid transformation matrix, and obtaining the source point cloud after coarse registration; Based on the coarse registration result, the improved algorithm is used for fine registration; the weighted factor is introduced by using normal vector features and combined with accelerated search to gradually optimize the rigid transformation matrix to improve registration accuracy, and finally output the accurate registration result of the source point cloud set and the target point cloud set.

2. The method for large-scale point cloud registration based on power grid transmission lines of claim 1, wherein, Adaptive octree downsampling is performed on the point cloud data of the source point cloud set and the target point cloud set respectively, and the interval of the selected points in different density regions is dynamically adjusted, representative point cloud data is obtained from the source point cloud set to generate a first point cloud data set, and representative point cloud data is obtained from the target point cloud set to generate a second point cloud data set, comprising: Initialization of octree: determining the source point cloud set of the power transmission line and the target point cloud set occupies in the three-dimensional space, defined as ; Splitting octree nodes: For the root node of the octree, it is recursively split into eight child nodes using the following formula: For example, the direction is and similar to the direction ； ； wherein , denotes the index of the eight child nodes in direction. Judgment adaptive subdivision: calculate the point cloud density within each sub-node , as follows: ； wherein, represents the number of points within a node; represents the spatial volume of a node, for a cuboid shaped node, the volume is: ； A density threshold of point cloud data in a power transmission tower region and a density threshold of point cloud data in an empty region are set, and when the node is further subdivided; wherein the density threshold of point cloud data in the power transmission tower region is greater than the density threshold of point cloud data in the empty region. Selecting representative points after completing the construction of the octree, comprising: Through the above steps, a part of representative points are selected from the original large-scale point cloud data, and adaptive octree downsampling is completed.

3. The method for large-scale point cloud registration based on power grid transmission lines of claim 2, wherein, After completing the construction of the octree, representative points are selected from each node, comprising: The method of node barycenter point is adopted for a point set in an octree node The calculation formula of the barycenter point is as follows: ； ； ； wherein, represents the number of points within the node.

4. The method of claim 1, wherein, Utilizing the improved The algorithm extracts point cloud feature points from the first and second point cloud datasets respectively. During the point cloud feature point extraction process, it uses... The structure accelerates neighborhood search and adaptively adjusts the neighborhood search radius based on the local geometric characteristics of the point cloud, thereby obtaining a set of feature points adapted to complex scene structures, including: Utilizing a structure determines a local neighborhood for each point in the point cloud data , determining a local neighborhood point cloud set within a radius of the point ; For local neighborhood point clusters Each point in Establish a local coordinate system and set a search radius, where the neighborhood search radius of the point cloud near the transmission tower is set to be larger than the neighborhood search radius of the point cloud in the smooth area of ​​the transmission line. Find all points within the search radius and calculate the weight of each point using the following formula: ； wherein, representative points weight inversely proportional to the number of points in its local neighborhood; representative search radius; representative points within the neighborhood, the sparser the points in the sample, the higher the weight; Calculate the three-dimensional weighted covariance matrix of each point, the formula is: ； Eigenvalues of each point are solved according to three-dimensional weighted covariance matrix of each point , and arranged in descending order, the relationship among the three eigenvalues reflects the main characteristics of the data points and their relative importance; two thresholds and are set for the point cloud near the structure complex transmission tower and the point cloud of smooth area of transmission line respectively; if the query point is the feature point, it needs to meet ； 5. The method for large-scale point cloud registration based on power grid transmission lines of claim 1, wherein, Based on the feature point set, a direction histogram feature is described in combination with geometric topology information, a geometric topology model reflecting local structure characteristics of the point cloud is constructed, including: for each feature point and a local neighborhood point cloud set , a direction vector of the neighborhood point relative to the feature point is calculated; The neighborhood point is , the direction vector is calculated as , that is, the direction vector is obtained by subtracting the coordinates of the feature point from the coordinates of the neighborhood point, so as to obtain a set of vector sets reflecting the spatial distribution direction of the neighborhood point relative to the feature point; Divide the vector set according to the direction space, count the number of direction vectors of the neighborhood points falling in each interval, and construct a histogram; Setting distance threshold and angle threshold If the distance between two neighboring points is less than and the included angle formed by them and the feature point is less than , it is considered that there is a connection between the two points, the connected point pairs in the neighborhood are identified, and the local topological information of the point cloud is quantified as a feature descriptor by establishing a geometric topological model.

6. The method for large-scale point cloud registration based on power grid transmission lines of claim 1, wherein, Based on the coarse registration result, the improved algorithm is used for fine registration; the weighted factor is introduced by using normal vector features, combined with accelerated search, and the rigid transformation matrix is gradually optimized to improve the registration accuracy, and finally the accurate registration result of the source point cloud set and the target point cloud set is output, including: Input source point cloud set Target point cloud set to be registered ; Respectively, the source point cloud set , the target point cloud set Establish Index; taking the source point cloud set as an example, for each point , search the neighborhood point set , construct the covariance matrix : ； wherein, is a point in the neighborhood, the eigenvector corresponding to the smallest eigenvalue of the covariance matrix is the normal vector of the point ; The normal vector features of the source point cloud set and the target point cloud set are used to assign a weighted value to each pair of matching points in the registration process, and a weighted factor is defined according to the normal vector information For optimizing matching, it is assumed that And The normal vector of the corresponding point pair, the weight Can be expressed as: ； wherein represents a smoothing parameter; Based on each point and its normal information, a descriptor is constructed to enhance the uniqueness of each feature point; Traversing the source point cloud collection All points ,use Gathering at the target point Search each nearest neighbor , forming matching point pairs set ; Compute the set of matching points The average distance between the matching points The standard deviation If the distance between a pair of points satisfies the criterion , then the pair of points is kept, otherwise it is discarded.​ According to Sort the matching point set from small to large Sort, filter out the top 70% matching point pairs Coordinates to solve the rotation matrix and translation vector ; Utilizing The rigid transformation matrix is obtained by solving the method , satisfying ; calculating the registration error evaluation function for this iteration : ； wherein, represents a weighting factor; represents the number of valid matching point pairs; represents the th conversion, if the difference between the two consecutive values satisfies a set threshold, then it is convergent, otherwise iteration continues.

7. A system for large-scale point cloud registration based on power grid transmission lines, characterized in that, Comprise: A data acquisition module is configured to acquire point cloud data of the power transmission line under different viewing angles to form a source point cloud set and a target point cloud set . A downsampling module is configured to respectively perform adaptive octree downsampling on point cloud data of a source point cloud set and a target point cloud set , and dynamically adjust intervals of selected points in different density regions, to obtain representative point cloud data from the source point cloud set , generate a first point cloud data set, and obtain representative point cloud data from the target point cloud set , generate a second point cloud data set. The feature point extraction module is configured to extract feature points of the point cloud from the first point cloud data set and the second point cloud data set respectively by using an improved algorithm, and in the process of extracting the feature points of the point cloud, a structure accelerated neighborhood search is performed, and a neighborhood search radius is adaptively adjusted according to local geometric characteristics of the point cloud, so that a feature point set adapted to a complex scene structure is obtained. algorithm, and in the process of extracting the feature points of the point cloud, a structure accelerated neighborhood search is performed, and a neighborhood search radius is adaptively adjusted according to local geometric characteristics of the point cloud, so that a feature point set adapted to a complex scene structure is obtained. The model construction module is used for constructing a direction histogram feature description model based on the feature point set in combination with geometric topology information to generate a geometric topology model; the geometric topology model captures the spatial distribution direction of the feature points in the feature point set within the neighborhood, reflects the local structure characteristics of the point cloud, and provides stable description features for subsequent registration; A coarse registration module is used to employ an improved The coarse registration module is used to employ an improved algorithm to perform coarse registration on the feature point sets of the source point cloud set and the target point cloud set, select non-collinear feature points that meet geometric relations, and optimize a rigid transformation matrix. By minimizing the matching error, an initial alignment matrix of the source point cloud set to the target point cloud set is obtained, comprising: In the source point cloud set and the target point cloud set , 4 corresponding data points are randomly selected, respectively denoted as and ; Solving the rigid body transformation matrix by least square method So that the source point cloud is subjected to a rigid rotation translation transformation to obtain a new point cloud set Wherein is a rotation matrix, is a translation vector; A computed source point cloud set Under a rigid body transformation matrix A new source point cloud set The distance projection error of the corresponding points of the target point cloud set ​ If the distance projection error of the point pair is less than a set threshold error , wherein, is the distance between the corresponding points, the point is added to the set of inliers I and a matching score is calculated i.e. the ratio of inliers to the total number of points in the point cloud; Setting the number of cycles and a matching score threshold , if the matching score of the point cloud is less than , then break the loop, otherwise repeat the above steps until the number of iterations reaches ; Counting the number of inliers under different rigid transformation matrices, taking the rigid transformation matrix with the most inliers as the best rigid transformation matrix, optimizing the rigid transformation matrix, and obtaining the source point cloud after coarse registration; A fine registration module is used to perform fine registration based on the coarse registration result by using an improved algorithm; a weighting factor is introduced by using normal vector features and combined with accelerated search, step-by-step optimization of the rigid body transformation matrix to improve registration accuracy, and finally output the accurate registration result of the source point cloud set and the target point cloud set.

8. A power grid transmission line based large scale point cloud registration device, characterized in that, The electronic device comprises a processor and a memory, and the memory stores computer instructions; the processor is used to execute the computer instructions stored in the memory; when the computer instructions are executed by the processor, the electronic device realizes the steps of the large-scale point cloud registration method based on the power grid transmission line in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, which is executed by the processor to realize the steps of the large-scale point cloud registration method based on the power grid transmission line in any one of claims 1 to 6.

Images