A high robustness reconstruction method for power towers based on airborne point clouds

Abstract

The application discloses a high-robustness reconstruction method for a power tower based on airborne point clouds, which comprises the following steps: S1, point cloud redirection; S2, point cloud data amplification, coordinates of the redirected point cloud are amplified by n times; S3, point cloud preprocessing, specifically, a symmetry neighborhood denoising method, adaptive adjustment of point cloud distribution density and symmetry enhancement of point cloud data; S4, structure decomposition, the tower is divided into an inverted triangular pyramid structure, a quadrangular prism structure and a complex structure from bottom to top; S5, inverted triangular pyramid structure reconstruction; and S6, quadrangular prism structure reconstruction, the external frame and the internal structure of the quadrangular prism are reconstructed respectively. According to the high-robustness reconstruction method for the power tower based on the airborne point clouds, the point cloud noise in the interior and the boundary of the tower can be effectively removed through the symmetry neighborhood denoising, the interference of the noise on the reconstruction result is avoided, the modeling precision is higher, the adaptive neighborhood point cloud density adjustment can solve the problem that the modeling parameters are not fixed due to the point cloud sparsity difference, and the modeling effect is more stable.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electric power, in particular to a high-robustness power tower reconstruction method based on airborne point clouds. BACKGROUND

[0002] Power towers, as a main type of building structure, exist widely in life, and fast and high-quality reconstruction of power tower structures has wide application value in virtual reality, building design and digital city, and has important scientific significance in simulation of power tower structure stress analysis (wind load, ice load, line tension, etc.) and disaster occurrence prediction (earthquake action, mountain landslide, etc.). However, due to the hole structure and type diversity of the power tower itself, it is difficult to quickly and completely model the power tower. In the face of this challenge, three main modeling methods, namely model-driven modeling method, data-driven modeling method and model-data driven combined modeling method, are applied to the power tower point cloud to reconstruct the power tower model. However, the above research focuses on the modeling of the external contour structure of the power tower, and lacks the reconstruction of the internal structure of the power tower. The rule-based modeling method can generate building models with more details by defining parameters and corresponding syntax rules. Therefore, how to use the rule-based modeling method to quickly and efficiently reconstruct the internal and external structures of the power tower is a hot issue worthy of study in the field of remote sensing.

[0003] In recent years, with the substantial improvement of computer hardware performance and the rapid development of three-dimensional modeling technology, a large amount of research has been conducted on the three-dimensional reconstruction method of power towers at home and abroad, and many valuable achievements have been made. However, basically, they can be classified into the following three categories: model-driven modeling method, data-driven modeling method and model-data driven combined modeling method. Although these modeling methods can reconstruct the external contour structure of the power tower, it is difficult to reconstruct the internal structure of the power tower with rich details.

[0004] The data-driven modeling method adopts a bottom-up strategy, first extracts basic features (such as points, lines and surfaces), then constructs the topological relationship between features, and completes reconstruction. The geometric model constructed based on data-driven has high geometric accuracy and is not constrained by the model library, but it is very dependent on data quality, such as point cloud density, point cloud structure integrity and spatial distribution uniformity. The quality of the point cloud obtained by airborne Lidar contains rich detailed information, which can compensate for the shortcomings of the data-driven method to some extent.

[0005] The closest prior art. Due to the complexity of the tower structure, there are few studies on data-driven reconstruction of towers. The existing paper "Research on 3D modeling of high-voltage transmission lines based on airborne LiDAR point cloud" and the patent "Transmission line body reverse modeling method, device, electronic equipment and storage medium" (CN115187648A) disclose a three-dimensional reconstruction method of electric power towers. The specific steps are as follows: based on the generalization structure characteristics of the tower, all tower structures are subdivided into inverted triangular pyramid structure, columnar structure, four-prism structure and complex structure, and an automatic tower reconstruction algorithm based on template structure is proposed. First, based on the existing point density characteristics, the filling rate characteristics are introduced to improve the tower segmentation position and tower type recognition accuracy, then the generalization specific template structure is used to reconstruct the columnar structure, inverted triangular pyramid structure and four-prism structure, for the four-prism structure, the internal structure reconstruction is added based on the existing reconstruction, the topology relationship of the complex structure model is determined based on the abstract template structure, the three-dimensional point cloud and two-dimensional image data processing method are fused to reconstruct the complex structure model, and the corner point coordinates are optimized through the mathematical model to solve the "expansion" error of the corner point caused by image processing. It is mentioned in the paper (5.1.3.4 section) that the data factors affecting the tower reconstruction accuracy can be divided into point cloud noise, point cloud sparsity and point cloud missing; it is mentioned in the paper (6.3 section) that there are two parameters (filling rate and complex structure boundary recognition parameter) in the automatic tower reconstruction algorithm, which are generally stable, the former has a small influence, and the latter has a large influence, and in summary, if there are structure missing and noise points in the tower point cloud, it is very difficult to use this method to reconstruct the tower structure, and if the density distribution of the point cloud is uneven, it is also easy to cause unstable modeling parameters. SUMMARY

[0006] In view of the shortcomings of the prior art, the present application provides a high-robustness reconstruction method of electric power towers based on airborne point cloud, which solves the problem that the tower model cannot be effectively reconstructed when the point cloud has noise, point cloud sparsity and point cloud missing in the prior art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a high-robustness reconstruction method of electric power towers based on airborne point cloud, comprising the following steps:

[0008] S1: point cloud reorientation;

[0009] S2: point cloud data amplification, the coordinates of the reoriented point cloud are amplified by n times;

[0010] S3: point cloud preprocessing, specifically symmetry neighborhood denoising method, adaptive adjustment of point cloud distribution density, and symmetry enhancement of point cloud data;

[0011] The symmetry neighborhood denoising method specifically comprises the following steps:

[0012] S3-1: Euclidean clustering of the point cloud with r1 as the radius, the point cloud data is divided into multiple point sets, and point sets with N1 points or less are taken for processing, and the number of point sets is n1;

[0013] S3-2: First, calculate the point set center p center1 , search the point cloud with r2 as the radius, get the number of neighborhood points n2; then calculate p center1 , the symmetry point around the Z axis p center2 , search the point cloud with r2 as the radius, get the number of neighborhood points n3;

[0014] S3-3: If n2-n1 > N2, and n3-n1 > N2, it is judged as a tower point set, and is retained to the tower point cloud; otherwise, it is judged as a noise point, and is removed;

[0015] The r1 is 0-50, the r2 value is 2 times the r1, the N1 value is 0-1000, and the N2 value is 0-1000;

[0016] The adaptive adjustment of the point cloud distribution density is specifically divided into the following steps:

[0017] S3-4: Use the nearest neighbor search algorithm to calculate the nearest 8 points of each point p, calculate the average distance d mean of the nearest points with a distance less than R, and set the distance interval T;

[0018] S3-5: If d mean > T, it means that the point cloud is sparse, then use the neighborhood-constrained up-sampling method to encrypt the point cloud, specifically calculate the neighboring points with a distance less than R as seed points p seed , generate new points in the middle of p seed to p with T as the interval, and complete the encryption of the sparse area point cloud;

[0019] S3-6: If d mean < T, it means that the point cloud is dense, then use the uniform sampling algorithm to uniformly sample in the XY direction with T as the interval, and complete the thinning of the dense point cloud;

[0020] S3-7: If d mean = T, it means that the point cloud density is moderate, then no operation is performed;

[0021] The symmetry of the point cloud data is enhanced, which is specifically divided into the following steps:

[0022] S3-8: The original point cloud C, with the XZ plane as the symmetry plane, generates a mirror image point cloud C1, the y value of C and the y1 value of C1 satisfy the relationship y1 = -y, and the x value and z value remain unchanged;

[0023] S3-9: The original point cloud C is symmetrical to the YZ plane, a mirror image point cloud C2 is generated, the x value of C meets the relationship x1 = -x with the x1 value of C2, and the y value and z value remain unchanged;

[0024] S3-10: Merge the point clouds C, C1 and C2 into C3;

[0025] S3-11: Uniformly sample the point cloud C3 at intervals of 2 to obtain C4, C4 is a symmetrical enhanced tower point cloud, which can restore the area with a large range of missing tower data;

[0026] S4: Structure decomposition, the tower is divided into an inverted triangular pyramid structure, a quadrangular prism structure and a complex structure from bottom to top;

[0027] S5: Inverted triangular pyramid structure reconstruction;

[0028] S6: Quadrangular prism structure reconstruction, the external frame and the internal structure of the quadrangular prism are reconstructed respectively;

[0029] S7: Complex structure reconstruction, divided into T-shaped tower and O-shaped tower structure, the reconstruction process includes extracting contour, contour optimization and calculating the three-dimensional coordinates of the corner point;

[0030] S8: Assemble the tower, combine the models of the inverted triangular pyramid structure, the quadrangular prism structure and the complex structure to obtain the final tower model.

[0031] Preferably, the quadrangular prism external frame reconstruction includes the following steps:

[0032] S6-1: Obtain the quadrangular prism point cloud data according to the z value range of the segmented section, and then project it on the YZ plane to obtain a two-dimensional image img;

[0033] S6-2: Image symmetry enhancement, flip the image left and right to obtain img flig , and add img flig and img enhance to obtain the enhanced image img l ;

[0034] S6-3: Perform image operations of first expansion and then corrosion on the image with S as the structure element size;

[0035] S6-4: Extract the left and right boundary points of the image, and fit two boundary straight lines L l and L r equations respectively, and calculate the y values of the four vertices of the image by substituting the z values of the two straight lines in the upper and lower segmented sections into the straight line equations respectively;

[0036] S6-5: Project the quadrangular prism point cloud data onto the XZ plane, repeat steps S6-1 to S6-4 to calculate the x values of the four vertices of the image;

[0037] S6-6: According to the vertex information of S6-4 and S6-5, the three-dimensional coordinates of the eight vertices of the quadrangular frustum are combined;

[0038] The internal structure reconstruction of the quadrangular frustum is specifically divided into the following steps:

[0039] S6-7: Project the quadrangular frustum point cloud data to the YZ plane and the XZ plane;

[0040] S6-8: Select the middle region image, calculate the upper and lower boundary points and the middle intersection points, and sort them from low to high according to the height;

[0041] S6-9: Divide the lower boundary and the adjacent intersection points into several intervals according to t pixels as an interval, then calculate the distance d of each interval to the two boundaries, and all intervals d form a set D; calculate the z value of the maximum value position of the set D; Substitute the z value of the maximum value position of the set D into the YZ plane straight line equation L l and L r , to get the y value of the boundary point of the intersection line; Substitute the z value of the maximum value position of the set D into the XZ plane straight line equation L l and L r , to get the x value of the boundary point of the intersection line;

[0042] S6-10: According to the topological relationship of each point, all points are connected to obtain the reconstruction model of the quadrangular frustum.

[0043] Preferably, the reconstruction process of the T-shaped tower structure is divided into the steps of extracting the contour, optimizing the contour, and calculating the three-dimensional coordinates of the corner points;

[0044] The extraction of the contour is specifically divided into the following steps:

[0045] S7-1: According to the z value range of the segmentation section, the complex structure point cloud data is obtained, and then a two-dimensional image img yz is obtained by projecting in the YZ plane;

[0046] S7-2: Known point enhanced image, known point enhances complex contour area, if the three-dimensional coordinates of the vertex in step S6-6 intersect with the complex structure segmentation section, it is recorded as a known point, after the known point is projected to the image img yz1 , a filled contour is drawn;

[0047] S7-3: Image symmetry enhancement, flip the image left and right to obtain img flig1 , img yz1 and img flig1 are added to obtain the enhanced image img enhance1 ;

[0048] S7-4: The image is first dilated and then eroded with E as the structural element size;

[0049] S7-5: The maximum contour of the image is extracted and fitted into a polygon contour O1;

[0050] The contour optimization, in particular, is divided into the following steps:

[0051] S7-6: The contour is optimized by judging the distance between the point in O1 and the known point. If the distance is less than 5, the known point is used to replace the point in O1. Then the known points that are not replaced are inserted into the known points in the order of the contour counterclockwise to generate a contour O2;

[0052] S7-7: The adjacent points are merged, and the distance between the contour point and the adjacent point is calculated. If the distance is less than R / 2, the adjacent points are merged;

[0053] The calculation of the three-dimensional coordinates of the corner points is as follows:

[0054] S7-8: The point cloud of the complex structure is projected onto the XY plane to obtain an image img xy1 ;

[0055] S7-9: The image is symmetrically enhanced by flipping the image upside down to obtain img flig2 , and adding img xy1 and img flig2 to obtain an enhanced image img enhance2 ;

[0056] S7-10: The image is first dilated and then eroded with E as the structural element size;

[0057] S7-11: The contour of the image is extracted, the vertical line is generated with the y value of the O2 contour point, the upper and lower intersection points of the vertical line and the contour are calculated, and the x value of the O2 contour point is obtained to finally generate a three-dimensional coordinate point;

[0058] S7-12: The topology of the final point is connected to obtain a reconstruction model of the complex structure.

[0059] Preferably, the reconstruction process of the O-type tower structure is divided into the steps of extracting the contour, optimizing the contour, and calculating the three-dimensional coordinates of the corner points;

[0060] The extraction of the contour, in particular, is divided into the following steps:

[0061] S7-1: The complex structure point cloud data is obtained according to the z value range of the segmentation section, and then projected onto the YZ plane to obtain a two-dimensional image img yz ;

[0062] S7-2: Known point enhanced image, known point enhanced complex contour region, if the vertex three-dimensional coordinate intersects with the complex structure segmentation segment in the S6-6 step, it is recorded as a known point, and the known point is projected to the image img yz2 Then, a filled contour is drawn;

[0063] S7-3: Image symmetry enhancement, the image is flipped left and right to obtain img flig3 , img yz2 and img flig3 are added to obtain the enhanced image img enhance3 ;

[0064] S7-4: The image is subjected to image operation of first inflation and then corrosion with E as the structure element size;

[0065] S7-5: The maximum contour and the second maximum contour of the image are extracted and fitted into a polygon, and are recorded as an outer contour O e and an inner contour O i ;

[0066] The contour optimization, in particular:

[0067] S7-6: Known point optimization of the outer contour, first judge the distance between the point in O e and the known point, if the distance is less than R / 2, the point in O e is replaced by the known point, then the known points not replaced are counted, and the known points are sequentially inserted into the contour counterclockwise;

[0068] S7-7: Merge adjacent points, calculate the distance between the contour point and the adjacent point, and merge if the distance is less than R / 2;

[0069] The calculation of the three-dimensional coordinates of the corner point, in particular:

[0070] S7-8: Project the point cloud of the complex structure to the XZ plane to obtain an image img xz ;

[0071] S7-9: The image is subjected to image operation of first inflation and then corrosion with E as the structure element size;

[0072] S7-10: Extract the contour of the image, generate a horizontal line with the z value of the contour point, calculate the left and right intersection points of the horizontal line and the contour, obtain the x value of the contour point, and finally generate a three-dimensional coordinate point;

[0073] S7-11: The topology of the final point connects all the points to obtain a reconstructed model of the complex structure.

[0074] Compared with the prior art, the application provides a power tower height robust reconstruction method based on airborne point cloud, which has the following beneficial effects:

[0075] 1. The power tower high robustness reconstruction method based on airborne point cloud, which can effectively remove point cloud noise inside and on the boundary of the tower through symmetric neighborhood denoising, avoid interference of noise on the reconstruction result, has higher modeling accuracy, and can solve the problem of unstable modeling parameters caused by sparse point cloud difference through adaptive neighborhood point cloud density adjustment, so that the modeling effect is more stable.

[0076] 2. The power tower high robustness reconstruction method based on airborne point cloud, which can effectively avoid point cloud reconstruction failure caused by missing point cloud in the symmetric area through symmetric enhancement of point cloud data, and improve the robustness of the modeling algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0077] The present application Figure 1 is: a redirected result map;

[0078] The present application Figure 2 is: a point cloud noise data and missing data schematic diagram;

[0079] The present application Figure 3 is: an inverted triangular pyramid reconstruction effect diagram;

[0080] The present application Figure 4 is: a four-prism table modeling flowchart;

[0081] The present application Figure 5 is: a T-shaped tower complex structure modeling flowchart;

[0082] The present application Figure 6 is: an O-shaped tower complex structure modeling flowchart;

[0083] The present application Figure 7 is: a tower modeling effect diagram. DETAILED DESCRIPTION

[0084] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. Embodiment one

[0085] The present method discloses a power tower high robustness reconstruction method based on airborne point cloud, which specifically comprises the following steps:

[0086] S1: point cloud redirection, first, the point cloud is decentralized; then, the PCA algorithm is used to calculate the eigenvalue and eigenvector of the point cloud, the eigenvector is used to rotate the point cloud on the XY plane, so that the horizontal arm of the point cloud is perpendicular to the X axis; finally, the z coordinate of the point cloud is subtracted by the lowest point z coordinate, and the point cloud is redirected, which is specifically shown in the attachedFigure 1 .

[0087] S2: Point cloud data amplification, the coordinates of the redirected point cloud are amplified by 10 times, and the unit of the point cloud coordinates is centimeter. The accuracy of the processing is improved.

[0088] S3: Point cloud preprocessing, specifically symmetry neighborhood denoising method, adaptive adjustment of point cloud distribution density, and symmetry enhancement of point cloud data.

[0089] S4: Structure decomposition, first use local maximum point density and filling rate to identify the segmentation position S and key segmentation position S of the tower. k Then, through the segmentation position, the tower is divided into inverted triangular pyramid structure, quadrangular prism structure and complex structure from bottom to top.

[0090] S5: Inverted triangular pyramid structure reconstruction, first use the segmentation position S1 and the bottom of the tower to get the inverted triangular pyramid point cloud data, S1 is the top of the inverted triangular pyramid; then project the point cloud to the YZ and XZ planes, calculate the upper boundary points to get the upper boundary point coordinates, and project the point cloud to the XY plane to calculate the four corner point coordinates to get the lower boundary point coordinates, and finally connect all the points according to the topological relationship of each point to get the reconstruction model of the inverted triangular pyramid. The specific effect is shown in the attached Figure 3 .

[0091] S6: Quadrangular prism structure reconstruction, the external frame and internal structure of the quadrangular prism are reconstructed respectively.

[0092] S7: Complex structure reconstruction, divided into T-shaped tower and O-shaped tower structure, the reconstruction process is divided into contour extraction, contour optimization and calculation of corner point three-dimensional coordinates.

[0093] S8: Combined tower, combine the models of inverted triangular pyramid structure, quadrangular prism structure and complex structure to get the final tower model.

[0094] The modeling effect of the method is shown in the attached Figure 7 , which can support three-dimensional modeling of T-shaped tower and O-shaped tower. The method can process complex structure and tower reconstruction (attached Figure 2 -a) with noise point cloud (attached Figure 7 -b) in the tower; and can realize the tower reconstruction (attached Figure 2 -a) of a large number of point cloud missing towers (attached Figure 7 -b). Example Two

[0095] There are two kinds of noise in the point cloud data that have obvious interference on the effect of point cloud reconstruction, which are shown in the attached Figure 2-a, the first type of noise distributed in the interior of the quadrangular prism, will affect the reconstruction of the cross line in the interior of the quadrangular prism; the second type of noise distributed near the complex structure affects the reconstruction of the complex structure, and at present, a manual removal method is adopted, the noise point cloud data has the characteristics of small number and randomness, and there is no data in the Y-axis symmetrical position in the neighborhood of the noise data.

[0096] Point cloud preprocessing, the steps of which include symmetrical neighborhood denoising, adaptive neighborhood point cloud density adjustment, and point cloud data symmetry enhancement.

[0097] Symmetrical neighborhood denoising, which is specifically divided into the following steps:

[0098] S3-1: Euclidean clustering of point cloud with 5 as radius, point cloud data is divided into multiple point sets, and point sets with less than 100 points are taken for processing, and the number of point sets is n1;

[0099] S3-2: first, calculate the center p of the point set center1 , search the point cloud with 10 as radius, get the neighborhood point number n2; then calculate p center1 , the symmetrical point around the Z axis p center2 , search the point cloud with 10 as radius, get the neighborhood point number n3;

[0100] S3-3: if n2-n1 > 10 and n3-n1 > 10, it is judged as a tower point set, and is reserved to the tower point cloud; otherwise, it is judged as a noise point, and is removed.

[0101] Adaptive neighborhood point cloud density adjustment, which is specifically divided into the following steps:

[0102] S3-4: use the nearest neighbor search algorithm to calculate the nearest 8 points of each point p, and calculate the average distance d of the nearest points with a distance less than 10 mean , take 2 as the threshold T;

[0103] S3-5: if d mean > T, it means that the point cloud is sparse, then use the neighborhood constraint upsampling method to encrypt the point cloud. Specifically, calculate the neighboring points with a distance less than 10 as seed points p seed , generate new points with 2 as interval between p seed and p to complete the encryption of the sparse area point cloud;

[0104] S3-6: if d mean < T, it means that the point cloud is dense, then use the uniform sampling algorithm to uniformly sample in the XY direction with 2 as interval to complete the thinning of the dense point cloud;

[0105] S3-7: if d mean = T, it means that the point cloud density is moderate, then no operation is performed.

[0106] For a large range of point cloud data missing, it is impossible to use the neighborhood point cloud processing method to supplement, and the tower point cloud has very strong structural symmetry. The structural symmetry is used to fill the large range of missing point cloud.

[0107] The point cloud data symmetry enhancement is specifically divided into the following steps:

[0108] S3-8: The original point cloud C is generated with the XZ plane as the symmetry plane, the mirror image point cloud C1 is generated, the y value of C meets the relationship y1 = -y, and the x value and z value remain unchanged;

[0109] S3-9: The original point cloud C is generated with the YZ plane as the symmetry plane, the mirror image point cloud C2 is generated, the x value of C meets the relationship x1 = -x, and the y value and z value remain unchanged;

[0110] S3-10: The point clouds C, C1 and C2 are merged into C3;

[0111] S3-11: The point cloud C3 is uniformly sampled with 2 as the interval to obtain C4, and the point cloud C4 is the symmetry enhanced tower point cloud. The point cloud C4 can restore the large range of missing areas of the tower data. Embodiment three

[0112] The four-pyramid external frame reconstruction is specifically divided into the following steps:

[0113] S6-1: Obtain the four-pyramid point cloud data according to the z value range of the segmented section, and then project in the YZ plane to obtain a two-dimensional image img;

[0114] S6-2: Image symmetry enhancement, flip left and right to obtain img flig , and add img flig to obtain the enhanced image img enhance ;

[0115] S6-3: With 3 as the structural element size, the image is first expanded and then eroded, and the final image is shown in the following Figure 4 -a;

[0116] S6-4: Extract the left boundary point and the right boundary point of the image, and fit two boundary straight lines L l and L r equations, respectively. The z values of the two straight lines in the upper and lower segmented sections are substituted into the straight line equations, respectively, and the y values of the four vertices of the image are calculated;

[0117] S6-5: Project the four-pyramid point cloud data to the XZ plane, repeat steps S6-1 to S6-4, and calculate the x values of the four vertices of the image;

[0118] S6-6: According to the vertex information of S6-4 and S6-5, the three-dimensional coordinates of the eight vertices of the quadrangular frustum are combined.

[0119] The internal structure reconstruction of the quadrangular frustum is specifically divided into the following steps:

[0120] S6-7: Project the quadrangular frustum point cloud data to the YZ plane and the XZ plane;

[0121] S6-8: Select the middle region image, calculate the upper and lower boundary points and the middle intersection points, and sort them from bottom to top according to height;

[0122] S6-9: Divide the lower boundary and the adjacent intersection points into several intervals with 5 pixels as the interval, then calculate the distance d of each interval to the two boundaries, and all intervals d form a set D; calculate the z value of the maximum value position of the set D; Substitute the z value of the maximum value position of the set D into the YZ plane straight line equation L l and L r , to get the y value of the boundary point of the intersection line; Substitute the z value of the maximum value position of the set D into the XZ plane straight line equation L l and L r , to get the x value of the boundary point of the intersection line, see attached Figure 4 -b.

[0123] S6-10: According to the topological relationship of each point, all points are connected to obtain the reconstruction model of the quadrangular frustum, see attached Figure 4 -c.

[0124] This method has the characteristics of high precision and high robustness in reconstructing the quadrangular frustum, and uses image symmetry enhancement to supplement the internal and external missing of the quadrangular frustum. Extracting intersection points and all boundary points can avoid interference caused by other noise. Example four

[0125] Due to the diversity of complex structures and the extremely low density of internal structure point cloud and the existence of a large number of structural missing.

[0126] Complex structure fine modeling is very difficult, this method mainly reconstructs the main contour of the complex structure, and the reconstruction core is to determine the three-dimensional coordinates of the connecting points and their topological relationship. Specifically divided into the following steps: extracting contour, contour optimization, calculating corner three-dimensional coordinates.

[0127] For the reconstruction process of T-shaped tower structure.

[0128] Extracting contour, specifically divided into the following steps:

[0129] S7-1: According to the z value range of the segmentation section, the complex structure point cloud data is obtained, and then projected to the YZ plane to obtain a two-dimensional image img yz ;

[0130] S7-2: Known point enhanced image, known point enhanced complex contour region, if the vertex three-dimensional coordinate intersects with the complex structure segmentation segment in S6-6 step, it is recorded as a known point. The known point is projected to the image img yz1 , and a filled contour is drawn, see attached Figure 5 -a;

[0131] S7-3: Image symmetry enhancement, the image is flipped left and right to obtain img flig1 , img yz1 and img flig1 are added to obtain the enhanced image img enhance1 ;

[0132] S7-4: The image is operated by first expanding and then eroding with a structure element size of 3;

[0133] S7-5: The maximum contour of img enhance is extracted and fitted into a polygon contour O1, see attached Figure 5 -b.

[0134] Contour optimization, specifically:

[0135] S7-6: Known point optimized contour, first judge the distance between the point in O1 and the known point, if the distance is less than 5, use the known point to replace the point in O1, then count the known points that have not been replaced, and insert the known points in the contour counterclockwise order to generate a contour O2, see attached Figure 5 -c.

[0136] S7-7: Merge adjacent points, calculate the distance between the contour point and the adjacent point, if the distance is less than R / 2, merge.

[0137] Calculate the three-dimensional coordinates of the corner points, specifically:

[0138] S7-8: Project the point cloud of the complex structure to the XY plane to obtain the image img xy1 , see attached Figure 5 -d;

[0139] S7-9: Image symmetry enhancement, the image is flipped up and down to obtain img flig2 , img xy1 and img flig2 are added to obtain the enhanced image img enhance2 ;

[0140] S7-10: The image is operated by first expanding and then eroding with a structure element size of 3;

[0141] S7-11: Extract the contour of the image, generate a perpendicular line with the y value of the O2 contour point, calculate the intersection point of the perpendicular line and the upper and lower contour, get the x value of the O2 contour point, and finally generate a three-dimensional coordinate point, see attached Figure 5 -f;

[0142] S7-12: The topology of the final point connects all the points to get the reconstruction model of the complex structure, see attached Figure 5 -g;

[0143] For O-shaped tower structure reconstruction process.

[0144] Extract the contour, which is divided into the following steps:

[0145] S7-1: Obtain the complex structure point cloud data according to the z value range of the segmentation section, and then project it in the YZ plane to get a two-dimensional image img yz , see attached Figure 6 -a;

[0146] S7-2: Known point enhanced image, known point enhanced complex contour area, if the three-dimensional coordinates of the vertex in step S6-6 intersect with the complex structure segmentation section, it is recorded as a known point. After the known point is projected to the image img yz , draw a filled contour;

[0147] S7-3: Image symmetry enhancement, flip the image left and right to get img flig3 , img yz2 and img flig3 add to get enhanced image img enhance3 ;

[0148] S7-4: Take 3 as the structure element size, perform image operation of inflation first and then corrosion on the image, see attached Figure 6 -b;

[0149] S7-5: Extract the largest contour and the second largest contour of the image, and fit them into a polygon, denoted as outer contour O e and inner contour O i , see attached Figure 6 -c.

[0150] Contour optimization, specifically:

[0151] S7-6: Known point optimization outer contour, first judge the distance between the point in O e and the known point, if the distance is less than 5, use the known point to replace the point in O e , then count the known points that have not been replaced, and insert them into the known points in the order of contour counterclockwise.

[0152] S7-7: Merge adjacent points, calculate the distance between the contour point and the adjacent point, if the distance is less than 5, merge, see attachedFigure 6 d and 6-e.

[0153] The three-dimensional coordinates of the corner points are calculated, and specifically:

[0154] S7-8: Project the point cloud of the complex structure to the XZ plane to obtain an image img xz , see attached Figure 6 -f;

[0155] S7-9: Perform image operations of first inflation and then corrosion on the image with a structure element size of 3;

[0156] S7-10: Extract the contour of the image, generate a horizontal line with the z value of the contour point, calculate the left and right intersection points of the horizontal line and the contour, obtain the x value of the contour point, and finally generate a three-dimensional coordinate point, see attached Figure 6 -h;

[0157] S7-11: The topological relationship of the final point connects all the points to obtain a reconstruction model of the complex structure, see attached Figure 6 -i;

[0158] The neighborhood parameter R, R ranges from 0 to 50; the optimal value of R is 10.

[0159] The distance interval parameter T, T ranges from 0 to 10, and the optimal value of T is 2; the parameter S, S=2T+1.

[0160] The interval interval parameter t, t ranges from 0 to 100, and the optimal value of t is 5.

[0161] The beneficial effects of the present application are:

[0162] 1) Symmetrical neighborhood denoising method.

[0163] 2) Self-adaptive adjustment of point cloud distribution density.

[0164] 3) Symmetrical enhancement of point cloud data.

[0165] 4) S6-9 of internal structure reconstruction, the lower boundary and the adjacent intersection point are divided into several intervals with t pixels as the interval, then the distance d of each interval to the two boundaries is calculated, and all the intervals d form a set D; the z value of the maximum value position of the set D is calculated; the z value is substituted into the YZ plane straight line equation L l and L r , to obtain the y value of the boundary point of the intersection line; the z value is substituted into the XZ plane straight line equation L l and L r , to obtain the x value of the boundary point of the intersection line.

[0166] 5) The reconstruction process of the T-shaped tower complex structure, the known point enhancement image of the contour and the image symmetry enhancement step.

[0167] 6) The reconstruction process of the complex structure of the T-tower, the known points which are not replaced in the profile optimization are inserted into the known points in anticlockwise order according to the profile.

[0168] 7) The reconstruction process of the complex structure of the T-tower, the adjacent points are merged in the profile optimization.

[0169] While the embodiments of the application have been shown and described, it is to be understood that the embodiments proposed are only by way of example and various changes, modifications, substitutions and alterations can be made thereto without departing from the spirit and scope of the application as defined in the appended claims and their equivalents.

Claims

1. A highly robust reconstruction method for power poles based on airborne point clouds, characterized in that, Includes the following steps: S1: Point cloud redirection; S2: Point cloud data magnification; the coordinates of the point cloud are magnified n times after redirection. S3: Point cloud preprocessing, specifically symmetric neighborhood denoising method, adaptive adjustment of point cloud distribution density, and symmetric enhancement of point cloud data; The symmetric neighborhood denoising method specifically comprises the following steps: S3-1: Perform Euclidean clustering on the point cloud with radius r1. The point cloud data is divided into multiple point sets. Take the point set with N1 points or less for processing. The number of point sets is n1. S3-2: First calculate the center p of the point set. center1 Search the point cloud with radius r2 to obtain the number of neighborhood points n2; then calculate p. center1 The point p that is symmetric about the Z-axis center2 Search the point cloud with radius r2 to obtain the number of neighborhood points n3; S3-3: If n2-n1 > N2 and n3-n1 > N2, then it is determined to be a tower point set and retained in the tower point cloud; otherwise, it is determined to be a noise point and removed. r1 is 0-50, r2 is twice the value of r1, N1 is 0-1000, and N2 is 0-1000. The adaptive adjustment of point cloud distribution density is specifically divided into the following steps: S3-4: Using the nearest neighbor search algorithm, calculate the 8 nearest neighbors of each point p, and calculate the average distance d between the nearest points whose distance is less than R. mean Set the distance interval T; S3-5: If d mean If T indicates that the point cloud is sparse, then a neighborhood-constrained upsampling method is used to densify the point cloud. Specifically, neighboring points with a distance less than R are calculated as seed points p. seed , in p seed New points are generated at intervals of T from point p to complete the encryption of the sparse point cloud. S3-6: If d mean If T < T, it indicates that the point cloud is dense. In this case, a uniform sampling algorithm is used to perform uniform sampling in the XY direction at intervals of T to thin out the dense point cloud. S3-7: If d mean = T indicates that the point cloud density is moderate, so no operation is performed; The symmetry-enhanced point cloud data is specifically processed in the following steps: S3-8: The original point cloud C is used as the symmetric plane of the XZ plane to generate a mirror point cloud C1. The y-values ​​of C and C1 satisfy the relationship y1 = -y, while the x and z values ​​remain unchanged. S3-9: The original point cloud C is used as the symmetric plane of the YZ plane to generate a mirror point cloud C2. The x value of C and the x1 value of C2 satisfy the relationship that x1 = -x, while the y and z values ​​remain unchanged. S3-10: Merge point clouds C, C1, and C2 into C3; S3-11: The point cloud C3 is uniformly sampled at intervals of 2 to obtain C4. C4 is a symmetrically enhanced tower point cloud. This point cloud C4 can restore the large-scale missing areas of tower data. S4: Structural decomposition, dividing the tower into inverted triangular pyramid structure, square frustum structure and complex structure from bottom to top; S5: Reconstruction of the inverted triangular pyramid structure; S6: Reconstruction of the quadrangular frustum structure, reconstructing both the external frame and the internal structure of the quadrangular frustum; S7: Complex structure reconstruction, divided into T-type tower and O-type tower structures. The reconstruction process is divided into contour extraction, contour optimization, and calculation of the three-dimensional coordinates of corner points. S8: Composite tower, which combines inverted triangular pyramid structure, square truncated pyramid structure and complex structure models to obtain the final tower model.

2. The method for high robustness reconstruction of power poles based on airborne point clouds according to claim 1, characterized in that, The reconstruction of the external frame of the frustum is specifically divided into the following steps: S6-1: Obtain the point cloud data of the quadrangular frustum based on the z-value range of the segment, and then project it onto the YZ plane to obtain the two-dimensional image img; S6-2: Image symmetry enhancement, flipping the image horizontally to obtain the img file. flig img and img flig Adding them together yields the enhanced image (img). enhance ; S6-3: Using S as the structuring element size, perform image operations on the image by first dilating and then eroding; S6-4: Extract the left and right boundary points of the image, and fit two boundary lines L to each. l and L r The equations are obtained by substituting the z-values ​​of the two lines in the upper and lower segments into the equations of the lines, and then calculating the y-values ​​of the four vertices of the image. S6-5: Project the quadrangular frustum point cloud data onto the XZ plane, repeat steps S6-1 to S6-4, and calculate the x-values ​​of the four vertices of the image. S6-6: Based on the vertex information of S6-4 and S6-5, the three-dimensional coordinates of the eight vertices of the quadrangular frustum are obtained by combining them. The reconstruction of the internal structure of the truncated pyramid is specifically divided into the following steps: S6-7: Project the quadrangular frustum point cloud data onto the YZ plane and XZ plane; S6-8: Select the middle region image, calculate the upper and lower boundary points and the middle intersection point, and sort them from bottom to top according to height; S6-9: Divide the lower boundary and its adjacent intersection points into several intervals with an interval of t pixels. Then calculate the distance d from each interval to the two boundaries. The distances d from all intervals form a set D. Calculate the z-value of the position of the maximum value in set D. Substitute the z-value of the position of the maximum value in set D into the equation L of the line in the YZ plane. l and L r In the equation, the y-values ​​of the boundary points of the intersecting lines are obtained; the z-value of the location of the maximum value in set D is substituted into the equation of the line L in the XZ plane. l and L r In the middle, the x-values ​​of the boundary points of the intersecting lines are obtained; S6-10: Connect all points according to their topological relationships to obtain the reconstructed model of the quadrangular frustum.

3. The method for high robustness reconstruction of power poles based on airborne point clouds according to claim 2, characterized in that, The reconstruction process of the T-shaped tower structure is divided into three steps: contour extraction, contour optimization, and calculation of the three-dimensional coordinates of the corner points. The extraction of the contour is specifically divided into the following steps: S7-1: Obtain complex structure point cloud data based on the z-value range of the segmented segments, and then project it onto the YZ plane to obtain a two-dimensional image (img). yz ; S7-2: Enhance the image with known points. Enhance complex contour regions with known points. If the 3D coordinates of a vertex in step S6-6 intersect with a segment of a complex structure, it is recorded as a known point. The known point is projected onto the image (img). yz1 Then, draw a filled outline; S7-3: Image symmetry enhancement, flipping the image horizontally to obtain the img file. flig1 img yz1 With img flig1 Adding them together yields the enhanced image (img). enhance1 ; S7-4: Using E as the structuring element size, perform image operations of first dilation and then erosion on the image; S7-5: Extract the maximum contour of the image and fit it into a polygonal contour O1; The contour optimization is specifically divided into the following steps: S7-6: Optimize the contour with known points. First, determine the distance between the points in O1 and the known points. If the distance is less than 5, replace the points in O1 with the known points. Then count the known points that have not been replaced. Insert the known points in the counterclockwise order of the contour. The generated contour is called O2. S7-7: Merge adjacent points. Calculate the distance between the contour point and adjacent points. If the distance is less than R / 2, merge them. The calculation of the three-dimensional coordinates of the corner point is specifically as follows: S7-8: Projecting a point cloud with a complex structure onto the XY plane to obtain an image (img) xy1 ; S7-9: Image symmetry enhancement, flipping the image vertically to obtain the img file. flig2 img xy1 With img flig2 Adding them together yields the enhanced image (img). enhance2 ; S7-10: Using E as the structuring element size, perform image operations of first dilation and then erosion on the image; S7-11: Extract the contour of the image, generate a perpendicular line based on the y-value of the O2 contour point, calculate the upper and lower intersection points of the perpendicular line and the contour, obtain the x-value of the O2 contour point, and finally generate the three-dimensional coordinate point. S7-12: Connect all points using the topological relationships of the final point to obtain a reconstruction model of the complex structure.

4. The method for high robustness reconstruction of power poles based on airborne point clouds according to claim 2, characterized in that, The reconstruction process of the O-shaped tower structure is divided into three steps: contour extraction, contour optimization, and calculation of the three-dimensional coordinates of the corner points. The extraction of the contour is specifically divided into the following steps: S7-1: Obtain complex structure point cloud data based on the z-value range of the segmented segments, and then project it onto the YZ plane to obtain a two-dimensional image (img). yz ; S7-2: Enhance the image with known points. Enhance complex contour regions with known points. If the 3D coordinates of a vertex in step S6-6 intersect with a segment of a complex structure, it is recorded as a known point. The known point is projected onto the image (img). yz2 Then, draw a filled outline; S7-3: Image symmetry enhancement, flipping the image horizontally to obtain the img file. flig3 img yz2 With img flig3 Adding them together yields the enhanced image (img). enhance3 ; S7-4: Using E as the structuring element size, perform image operations of first dilation and then erosion on the image; S7-5: Extract the largest and second largest contours of the image and fit them into a polygon, denoted as the outer contour O. e and inner contour O i ; The contour optimization specifically includes: S7-6: Optimize the outer contour using known points, first determine O e If the distance between the midpoint and the known point is less than R / 2, then the known point is used to replace O. e Then, count the known points that have not been replaced, and insert the known points in a counter-clockwise order according to the installation contour; S7-7: Merge adjacent points. Calculate the distance between the contour point and adjacent points. If the distance is less than R / 2, merge them. The calculation of the three-dimensional coordinates of the corner point is specifically as follows: S7-8: Projecting a point cloud with a complex structure onto the XZ plane to obtain an image (img) xz ; S7-9: Using E as the structuring element size, perform image operations of first dilation and then erosion on the image; S7-10: Extract the contour of the image, generate a horizontal line using the z-value of the contour point, calculate the left and right intersection points of the horizontal line and the contour, obtain the x-value of the contour point, and finally generate the three-dimensional coordinate point. S7-11: Connect all points using the topological relationships of the final point to obtain a reconstruction model of the complex structure.

Images