Multi-station tower laser point cloud high-precision splicing method and related device

Abstract

The invention discloses a multi-station tower laser point cloud high-precision splicing method. The method comprises the following steps: extracting high-strength point cloud blocks; separating a target ball point cloud from the high-strength point cloud block; performing robust estimation on the target ball point cloud, and extracting target ball point cloud parameters; determining a one-to-one correspondence relationship of target ball point clouds based on the target ball point cloud parameters to obtain one-to-one corresponding registration key points; calculating two-dimensional normal distribution probability density according to the target ball point cloud parameters, and determining a weight matrix based on corresponding target ball point cloud registration; and constructing a Rodrigues matrix, calculating the spatial relationship of each scanning observation station by using the registration key points, and realizing the accurate splicing of the point clouds of the multi-station electric power tower. According to the method, robust automatic registration of the point cloud of the multi-station ground three-dimensional laser scanning power transmission tower is realized, the target ball fitting error influence caused by the scanning distance, the incident angle and the like is weakened, and automatic and accurate splicing of the power transmission tower is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of transmission project surveying and mapping, and particularly relates to a high-precision multi-station tower laser point cloud stitching method. Background Art

[0002] UHV transmission lines play an important role in power transmission and ensuring people's livelihood. Transmission networks are often set up in inaccessible areas with harsh field conditions. As the service life increases, the demand for abnormal detection of the structure of power towers is constantly increasing. UAVs equipped with sensors for power inspection have been widely used in line detection, tower inclination measurement, insulator detection and other fields due to their convenient and fast characteristics. However, when using airborne lidar, due to the high flight altitude of the UAV, the ground point cloud density is low, and it is restricted by environmental factors such as vegetation occlusion, so it is impossible to obtain a complete power tower point cloud. It is even more difficult to extract tower parameters based on the image-based inspection method. For the measurement task of key towers, the ground three-dimensional laser scanning technology (Terrestrial Laser Scanning, TLS) can be used to obtain high-precision three-dimensional point clouds of the target. After target extraction and multi-viewpoint cloud registration, fine tower point clouds are obtained, and then structural parameters are extracted. Although many scholars have proposed research ideas in each stage of the above process respectively, simply combining them cannot achieve the automation of obtaining fine tower point clouds. Summary of the Invention

[0003] The purpose of the present invention is to overcome the above deficiencies of the prior art and provide a high-precision multi-station tower laser point cloud stitching method to achieve precise stitching of multi-station ground three-dimensional laser scanning transmission towers.

[0004] The inventors have found through research that to obtain a complete tower target using TLS, it is necessary to complete scans from multiple perspectives and register the extracted power towers. The Iterative Closest Points (ICP) algorithm and its improved algorithms have high registration efficiency, but good initial values need to be set, and the operation results are related to the number of iterations, and the registration results are unstable; using an external target as the registration key point only requires 3 common named points in adjacent measurement stations. However, in the actual operation process, it is impossible to ensure that the distance and incident angle between the measurement system and the target ball meet the recommended range. Therefore, it is necessary to introduce a robust target ball parameter fitting method, and at the same time evaluate the suitability of each corresponding target ball as a registration key point, and use the two-dimensional normal distribution probability density of its statistical value as the right confirmation factor to improve the robustness of the fine registration of the power tower point cloud.

[0005] The technical solution adopted by the present invention is as follows: A high-precision multi-station tower laser point cloud stitching method includes:

[0006] Preprocess the point clouds of power transmission towers at multiple scanning stations collected by scanning, and extract high-intensity point cloud blocks;

[0007] Separate the target ball point cloud and the tower structure point cloud from the high-intensity point cloud blocks;

[0008] Perform robust estimation on the target ball point cloud, and extract target ball point cloud parameters (including the coordinates of the target ball center point and the target ball radius. The former is used as the key point for multi-station point cloud registration, and the latter is used to determine the key point weight value during registration);

[0009] Calculate the spatial distance and angular relationship between the target ball point clouds based on the target ball point cloud parameters, determine the one-to-one correspondence of the target ball point clouds, and obtain the one-to-one corresponding registration key points;

[0010] Calculate the two-dimensional normal distribution probability density according to the target ball point cloud parameters, and determine the weight matrix during registration based on the corresponding target ball point clouds;

[0011] Construct a Rodriguez matrix, use the registration key points to calculate the spatial relationship of multiple scanning stations, and realize the precise splicing of multi-station power transmission tower point clouds.

[0012] The preprocessing of the point clouds of power transmission towers at multiple scanning stations collected by scanning to extract high-intensity point cloud blocks is specifically as follows:

[0013] (1) Statistically calculate the intensity mean value of the point cloud {P T} within the tower range Retain the point set with intensity values greater than ; according to the height range where the target balls are arranged, set an appropriate height threshold, and screen out the point set denoted as {P F};

[0014] (2) Randomly set a point P in {P F} as the seed;

[0015] (3) Search for the point set {P th} with the seed point P as the center of the sphere and D n as the radius, and calculate the Euclidean distance D n from each point P i (i = 1,..., n) in {P i} to P, and denote the distance mean value as and the standard deviation as σ n ;

[0016] (4) Traverse P i , if is considered that P i is an outlier, otherwise it is considered that P i belongs to the same class as P;

[0017] (5) Add the seed point P and the filtered point set {P n} to the current cluster {G j}, and set the points in {P n} as new seed points, then repeat steps (3)-(5);

[0018] (6) When the number of points in {G j} no longer changes, the number of seed points also no longer changes. At this time, the set of seed points represents the boundary of the current cluster. Add all boundary points to the current cluster, and delete the points in {G j} from {P F};

[0019] (7) Repeat steps (2)-(6) until there are no new seed points;

[0020] (8) Delete the clusters with the number of points less than N min , and a total of J valid clusters are obtained.

[0021] Separating the target ball point cloud and the tower structure point cloud from the high-intensity point cloud block specifically includes: estimating the curvature of the point cloud data in the high-intensity point cloud block, and judging by comparing with the corresponding curvature of the known target ball radius to distinguish the target ball point cloud and the tower structure point cloud.

[0022] The estimating the curvature of the point cloud data in the high-intensity point cloud block and judging by comparing with the corresponding curvature of the known target ball radius to distinguish the target ball point cloud and the tower structure point cloud specifically includes:

[0023] In the current cluster {G k}, randomly select t points, fit the local quadratic surface of their neighborhood points, and calculate the local differential geometric properties through formula (1), including the mean curvature Gaussian curvature K and principal curvatures K1 and K2, where K1 is the maximum curvature of a point on the quadratic surface in each direction, and K2 is the minimum curvature. When K1≈K2≠0, it is considered that the point is on the sphere;

[0024]

[0025] Among them, L = r xx n, M = r xy n, N = r yy n, E = r x r x F = r x r y G = r y r y r xx r xy r yy r x ry is the partial differential of the curved surface;

[0026] When more than 90% of the t local curved surfaces in the cluster {G k} satisfy the curvature constraint condition of formula (2), the current cluster can be determined as the target ball, where ε is the curvature threshold;

[0027]

[0028] The robust estimation of the target ball point cloud and the extraction of the target ball point cloud parameters are specifically as follows:

[0029] The criterion for robust estimation is expressed by formula (3), where is the estimated vector of the unknown parameter, w i is the i-th diagonal element of the weight matrix W, a i is the i-th row of the full column rank design matrix A, V i is the residual vector, L i is the observed value;

[0030]

[0031] Take the derivative of X in formula (3) and set it to 0, and denote Then there is:

[0032]

[0033] Let be the weight factor, be the diagonal element of the equivalent weight matrix, then the solution form of the least squares can be expressed by formula (5);

[0034]

[0035] The weight function of the IGGⅢ method is as shown in formula (6), where k0 and k1 are the weight factor adjustment thresholds, is the standardized residual corresponding to the observed value L i ;

[0036]

[0037] For the target ball model (x - a) 2 +(y - b) 2 +(z - c) 2 = r 2 , construct -2ax - 2yb - 2zc + d = x 2 + y 2 + z 2 , where d = a 2 + b 2 + c 2 - r2 The least squares method in matrix form can be obtained as follows:

[0038]

[0039] The target ball point cloud parameters are calculated from Equation (7), including the target ball diameter The center of the target ball where j = 1, 2, …, N s is the serial number of the scanning station, and N s is the total number of scanning stations, k = 1, 2, …, N t is the serial number of the target ball, and N t is the total number of target balls.

[0040] Based on the target ball point cloud parameters, calculate the spatial distance and angular relationship between the target ball point clouds, determine the one-to-one correspondence relationship of the target ball point clouds, and obtain the registered key points in one-to-one correspondence, specifically:

[0041] Based on the target ball diameter The center of the target ball Calculate the relative position relationship of the corresponding target balls, and calculate the mutual relationship between the target balls within the same station j through Equation (8);

[0042]

[0043] where k + m ≤ N t , and m is a natural number greater than 0; is the spatial distance between the k-th target ball and the (k + m)-th target ball, is the included angle; for the target balls with the same corresponding relationship, the relative position and angular relationship in different stations are the same. Compare the difference between the binary groups When both the distance and the angle are less than the thresholds ε d and ε θ , it is considered that the target ball point clouds are in one-to-one correspondence.

[0044] Calculate the two-dimensional normal distribution probability density according to the target ball point cloud parameters, and determine the weight matrix for registration based on the corresponding target ball point clouds, specifically:

[0045] The statistical values of the corresponding target ball radii for registration conform to the bivariate normal distribution and are independent of each other. The probability density can be used to determine the weights, that is, the more the fitting radius of any target ball pair deviates from the mean value, the smaller its probability density, and the probability density value conforms to the normal distribution; according to the target ball diameter The center of the target ball and the registered key points, calculate the binary probability density according to Formula (9), where r1 and r2 are the fitting radii of the corresponding target balls, is the standard deviation of the fitted target ball, r2 is the base of the natural logarithm, is the mean of the fitted radii of all target spheres.

[0046]

[0047] All corresponding target balls are calculated to obtain a set of weights, and the matrix W with these values as diagonal elements is fit It is the weight matrix during registration.

[0048] The Rodriguez matrix is constructed, and the spatial relationship of each scanning station is calculated using the registration key points to achieve accurate splicing of multi-station power tower point clouds, specifically:

[0049] Use the Rodrigues matrix to describe the transformation relationship of the registration station:

[0050]

[0051] Among them, (X f ,Y f ,Z f ) is the registration point in the first coordinate system, (X s ,Y s ,Z s ) is the corresponding registration point in the second coordinate system, λ is the scale factor, R is the rotation matrix, [ΔX, ΔY, ΔZ] T is the translation vector;

[0052] The Rodriguez matrix is used to describe the rotation change and is calculated as:

[0053]

[0054] Substituting equation (11) into equation (10), the resulting transformation relationship includes seven unknown parameters: λ, α, β, γ, ΔX, ΔY, and ΔZ. Under the premise of scale factor λ≈1, the translation parameters can be eliminated by difference, and the following simple relationship can be further obtained:

[0055]

[0056] Combining the weight matrix, equation (11) and equation (12), we can get the following error equation:

[0057]

[0058] in,

[0059]

[0060] Formula (14) is a singular coefficient matrix, where n ≥ 3;

[0061] After solving equations (12) and (13) to obtain the Rodriguez matrix, substitute a pair of corresponding registered key points into formula (10) to solve the translation vector [ΔX, ΔY, ΔZ]. T to achieve high-precision automatic and accurate splicing of transmission towers.

[0062] The present invention also provides a high-precision splicing device for multi-station tower laser point clouds, which is characterized by including:

[0063] An intensity clustering module for preprocessing the power tower point clouds of multiple scanning stations collected by scanning to extract high-intensity point cloud blocks;

[0064] A point cloud separation module for separating the target ball point cloud and the tower structure point cloud from the high-intensity point cloud blocks;

[0065] A robust estimation module for robustly estimating the target ball point cloud and extracting target ball point cloud parameters (including the coordinates of the target ball center point and the target ball radius, the former is used as the key points for multi-station point cloud registration, and the latter is used to determine the key point weights during registration);

[0066] A target ball relationship correspondence module for calculating the spatial distance and angular relationship between target ball point clouds based on the target ball point cloud parameters, determining the one-to-one correspondence of target ball point clouds, and obtaining one-to-one corresponding registration key points;

[0067] A corresponding target ball weight determination module for calculating the two-dimensional normal distribution probability density according to the target ball point cloud parameters and determining the weight matrix during registration based on the corresponding target ball point cloud;

[0068] An accurate splicing module for constructing a Rodriguez matrix, calculating the spatial relationship of each scanning station using the registration key points, and realizing the accurate splicing of multi-station power tower point clouds.

[0069] On the other hand, the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the method for accurately splicing multi-station ground three-dimensional laser scanning transmission tower point clouds described in the first aspect.

[0070] On the other hand, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for accurately splicing multi-station ground three-dimensional laser scanning transmission tower point clouds described in the first aspect.

[0071] The beneficial effects of the present invention are as follows: (1) The related technologies do not consider the noise problem in the target ball fitting process and the problem of determining the rights of the target ball pair. The present invention uses a robust estimation method to fit the target ball parameters and determines the weights by statistical means, significantly improving the registration accuracy; (2) The related technologies mostly use airborne laser scanning technology to obtain the point cloud of transmission towers. The main inspection targets are power lines, insulators, etc., and less attention is paid to the transmission towers themselves. Only parameters such as the longitudinal and lateral inclination of the towers can be extracted, and it is impossible to cope with the key tower monitoring tasks. The present invention uses ground three-dimensional laser scanning technology to obtain high-precision point cloud of transmission towers, significantly increasing the point cloud density of transmission towers and realizing precise measurement of key structures; (3) To achieve high-precision stitching of transmission towers in the related technologies, manual intervention is mostly required to interactively extract key points. The present invention realizes the full automation of the registration process by constraining the reflection intensity between the target ball and the tower, the geometric attributes between the target ball and the tower, and the spatial relationship between the target balls.

[0072] The present invention realizes robust automatic registration for the point cloud of transmission towers scanned by multi-station ground three-dimensional laser, weakens the influence of target ball fitting errors caused by scanning distance, incident angle, etc., and fully utilizes the intensity information of the target ball and the tower structure, the geometric attribute differences between the target ball and the tower structure, and the spatial relationship between multi-station target balls to realize automatic and precise stitching of transmission towers. Brief Description of the Drawings

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0074] Figure 1 It is the flowchart of the method of the present invention;

[0075] Figure 2 It is the schematic diagram of intensity clustering of the present invention;

[0076] Figure 3 It is the registered key point pairs of the present invention;

[0077] Figure 4 It is the schematic diagram of the original point cloud of the present invention;

[0078] Figure 5 It is the schematic diagram of the registered tower foot point cloud of the present invention;

[0079] Figure 6 It is the schematic diagram of the registered tower body point cloud of the present invention;

[0080] Figure 7Schematic diagram of the tower head point cloud after registration for the present invention. Specific implementation manners

[0081] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0082] The first aspect of the present invention provides a method for high-precision stitching of multi-station transmission tower laser point clouds, aiming to solve the problem that the prior art cannot achieve automatic and precise stitching of high-density and high-precision point clouds of transmission towers. The present invention aims to accurately stitch the multi-station ground three-dimensional laser transmission tower point clouds scanned multiple times and from multiple angles. By arranging target balls with high reflectivity characteristics and isotropy at appropriate positions on the transmission tower, key points for calculating the registration transformation relationship are used to achieve point cloud registration. The target balls, like the steel structure of the power tower, have the characteristic of high reflectivity and appear as points with higher intensity in the point cloud. Setting an intensity threshold can quickly filter out the target ball points and tower structure points; there are obvious geometric feature differences between the target balls and the tower structure, and curvature can be used as a judgment basis to distinguish them; using a robust estimation method can weaken the influence of noise points on the fitting of target ball parameters, establish the corresponding relationship of multi-station target balls according to the spatial relationship between the target balls in the same measurement station, and then use the two-dimensional normal distribution probability density of the fitting statistical value of the target balls to determine the weight matrix in the registration process; finally, the Rodriguez matrix is used to describe the coordinate transformation relationship of the transmission tower.

[0083] The method for accurately stitching the multi-station ground three-dimensional laser scanning transmission tower point clouds provided by the present invention has achieved centimeter-level registration accuracy and includes the following steps:

[0084] S10. Since the target balls and the tower have relatively high reflectivity and the intensity values of the corresponding point clouds are relatively large, an intensity-based region growing clustering method is used to extract high-intensity point cloud blocks.

[0085] The layout principle of the target balls in the present invention is that any three target balls are not collinear and multiple target balls are not coplanar.

[0086] The scanner in the present invention is a ground three-dimensional laser scanner, which is installed at an open location with as much visibility as possible to the tower and the target balls.

[0087] The present invention keeps the installation height of the three-dimensional laser scanner between 1.5 m and 1.8 m, and the horizontal field of view angle of the three-dimensional laser scanner is 360°, and the vertical field of view angle is 300°.

[0088] The target ball is usually designed as a white diffuse reflection surface with a high intensity in the point cloud; the main and auxiliary materials that make up the power transmission tower are mostly made of light steel, angle steel, and aluminum alloy, which also have the characteristic of high strength like the target ball. Therefore, by setting an intensity threshold for the input original point cloud, the point clouds of other materials with low reflection intensity can be removed. Refer to Figure 1 - Figure 2 , and its clustering process is as follows:

[0089] (1) Calculate the mean intensity of the point cloud {P T} within the range of the tower; Retain the point set with intensity values greater than ; According to the height range where the target ball is arranged, set an appropriate height threshold. Since the target ball is generally arranged at the tower bottom, the points within 1 / 4 of the tower height can be retained. Denote the filtered point set as {P F};

[0090] (2) Randomly set a point P in {P F} as the seed;

[0091] (3) Search for the point set {P th} with the seed point P as the center and D n as the radius, and calculate the Euclidean distance D n from each point P i (i = 1,..., n) in {P i} to P. Denote the mean distance as and the standard deviation as σ n ;

[0092] (4) Traverse P i . If , consider P i as an outlier, otherwise consider P i to belong to the same class as P;

[0093] (5) Add the seed point P and the filtered point set {P n} to the current cluster {G j}. Set the points in {P n} as new seed points, and repeat steps (3)-(5);

[0094] (6) When the number of points in {G j} no longer changes, the number of seed points also no longer changes. At this time, the seed point set represents the boundary of the current cluster. Add all boundary points to the current cluster, and delete the points in {G j} from {P F};

[0095] (7) Repeat steps (2)-(6) until there are no new seed points;

[0096] (8) Delete the clusters with the number of points less than Nmin Clustering is performed, and a total of J effective clusters are obtained. Among them, N min is related to the point density, and it is recommended to take N min > 200.

[0097] The target ball is connected to the pole tower through a connecting piece, which is discontinuous in space, and there are obvious curvature characteristic differences between the target ball and the pole tower material. Therefore, the target ball can be extracted by the method of curvature constraint clustering. For the point cloud within a certain height range after being screened by the strength threshold, clustering is performed based on the Euclidean distance, and the local curvature of each clustering point is calculated by sampling.

[0098] S20, the target ball and the pole tower structure have obvious geometric differences. The method of sampling point curvature estimation is used to separate the target ball point cloud and the pole tower structure point cloud.

[0099] The steps of the sampling point curvature estimation method described in the present invention are as follows: In the current cluster {G k}, randomly select t points, fit the local quadratic surface of their neighborhood points, and calculate the local differential geometric properties through formula (1), including the mean curvature Gaussian curvature K and principal curvatures K1 and K2, where K1 is the maximum curvature of a certain point on the quadratic surface in each direction, and K2 is the minimum curvature. When K1≈K2≠0, it is considered that the point is on the sphere.

[0100]

[0101] Among them, L = r xx n, M = r xy n, N = r yy n, E = r x r x , F = r x r y , G = r y r y , r xx 、r xy 、r yy 、r x 、r y are the partial differentials of the surface.

[0102] When more than 90% of the t local surfaces in the cluster {G k} satisfy the curvature constraint condition of formula (2), it can be determined that the current cluster is the target ball, where ε is the curvature threshold, which is determined according to the radius of the target ball used, and ε = 1 / the curvature radius of the target ball.

[0103]

[0104] Due to environmental restrictions during the layout of the measurement stations, it is impossible to ensure that the distance between the scanner and the target ball is within the recommended range. A long distance means a larger laser spot on the surface of the target ball. Therefore, points on the edge of the target ball will be deformed due to the "trailing" phenomenon, that is, there are gross errors in the target ball points participating in the fitting.

[0105] S30. The separated target ball point cloud contains some noise points. The parameters of the target ball point cloud are robustly extracted by the IGGⅢ robust estimation method, including the coordinates of the target ball in the three-dimensional space coordinate system and the radius of the target ball.

[0106] The steps for the IGGⅢ robust estimation method of the present invention to robustly extract the parameters of the target ball point cloud are as follows: The criterion for robust estimation is represented by formula (3), where is the estimated vector of unknown parameters, w i is the i-th diagonal element of the weight matrix W, a i is the i-th row of the full column rank design matrix A, V i is the residual vector, L i is the observed value.

[0107]

[0108] Take the derivative of X in formula (3) and set it to 0, and denote Then there is:

[0109]

[0110] Let be the weight factor, be the diagonal element of the equivalent weight matrix, then the solution form of the least squares can be represented by formula (5).

[0111]

[0112] The IGGⅢ method has strong resistance to outliers and is easy to construct. Its weight function is shown in formula (6), where k0 and k1 are weight factor adjustment thresholds, which are related to the probability of the distribution of the mean error, is the observed value L i corresponding standardized residual.

[0113]

[0114] For the target ball model (x - a) 2 +(y - b) 2 +(z - c) 2 = r 2 , construct -2ax - 2yb - 2zc + d = x 2 + y 2 + z 2 , where d = a 2 + b 2+c 2 -r 2 , the least squares matrix form can be obtained as follows:

[0115]

[0116] The target ball point cloud parameters are solved from Equation (7), including the target ball diameter the center of the target ball where j = 1, 2, …, N s is the serial number of the scanning station, and N s is the total number of scanning stations, k = 1, 2, …, N t is the serial number of the target ball, and N t is the total number of target balls.

[0117] Calculate the distances and angles between the fitting centers of the target balls in the measured point cloud of the same station. If the positions of the target balls do not change during the measurement process, the relative positions of the target balls in different scanning stations do not change.

[0118] Even if some target balls cannot be scanned due to occlusion or other reasons, or are not successfully extracted due to incomplete target ball point clouds, it will not affect the spatial relationship between the target balls with clear visibility, and thus the corresponding target ball point clouds of different scanning stations.

[0119] S40. Based on the spatial distance and angle relationships between the multi-station ground three-dimensional laser target ball point clouds extracted, determine the one-to-one correspondence relationship of the target ball point clouds.

[0120] The step for determining the one-to-one correspondence relationship of the ball point clouds described in the present invention is as follows: Refer to Figure 3 , the positions of the target balls do not change during multi-view scanning. Therefore, after obtaining the parameters of multiple target balls in a single scanned point cloud through robust fitting, the correspondence of the target balls is judged based on the distance and included angle relationships between the target balls, which is used as the key points for point cloud registration.

[0121] Using the target ball parameters of each scanning station obtained by the above S30, the target ball diameter the center of the target ball calculate the relative position relationship of the corresponding target balls, where j = 1, 2, …, N s is the serial number of the scanning station, and N s is the total number of scanning stations, k = 1, 2, …, N t , N t is the total number of target balls. Calculate the mutual relationship between the target balls within the same scanning station j through Equation (8).

[0122]

[0123] where k + m is the serial number of all subsequent target balls whose serial number is larger than that of target ball k, and k + m ≤ N t , is the spatial distance between the k target ball and the k+m target ball, and is the included angle. For the target balls with the same corresponding relationship, their relative positions and angular relationships in different measuring stations are the same. Comparing the binary group the difference between them. When both the distance and the angle are less than the thresholds ε d and ε θ , it is considered that the target ball point clouds are in one-to-one correspondence. Determine the thresholds ε d and ε θ according to the requirements of the splicing accuracy. It is recommended that ε d <0.1m, ε θ <1°.

[0124] The closer the fitting result of the target ball is to the actual radius of the target ball, the higher the quality of the point cloud participating in the fitting operation of this target ball can be considered, the closer its fitting radius value is to the true value, the smaller the standard deviation, and the greater the probability density.

[0125] S50. Calculate the two-dimensional normal distribution probability density through the statistical parameters of the corresponding target ball point cloud, and determine the weight matrix for registration based on the corresponding target ball point cloud.

[0126] The steps of the present invention for determining the weight matrix for registration based on the corresponding target ball point cloud are as follows:

[0127] For the corresponding target ball, its radius fitting results are independently and identically distributed. Use the probability density formula (9) of the two-dimensional normal distribution to describe its confidence level, and determine the weight value of the corresponding target ball as the key point for coordinate transformation calculation through the confidence level.

[0128] The statistical values of the radii of the corresponding target balls for registration conform to the bivariate normal distribution and are independent of each other. The weight can be determined using the probability density, that is, the more the fitting radius of any pair of target balls deviates from the mean, the smaller its probability density, and the probability density value conforms to the normal distribution. According to the target ball parameters obtained from the S30, including the target ball diameter the center of the target ball and the corresponding relationship of the target balls obtained from the S40, calculate the binary probability density according to formula (9), where r1 and r2 are the fitting radii of the corresponding target balls, is the standard deviation of the fitting target ball, r2 is the base of the natural logarithm, is the mean of the fitting radii of all target balls.

[0129]

[0130] A set of weight values are calculated for all corresponding target balls, and the matrix W fit with these values as diagonal elements is the weight matrix for registration described in S50.

[0131] S60. Using no less than 3 groups of corresponding target ball point clouds, construct a Rodriguez matrix to achieve precise stitching of the point clouds of transmission towers in multi-station terrestrial three-dimensional laser scanning. The specific steps are as follows:

[0132] The general conversion relationship in a spatial rectangular coordinate system is:

[0133]

[0134] Among them, (X f , Y f , Z f ) is the registration point in the first coordinate system, (X s , Y s , Z s ) is the corresponding registration point in the second coordinate system, λ is the scale factor, R is the rotation matrix, [ΔX, ΔY, ΔZ] T is the translation vector, and the Rodriguez matrix is used to describe the rotational change.

[0135] The expanded form of the Rodrigues rotation matrix is:

[0136]

[0137] Substituting Equation (11) into Equation (10), the obtained conversion relationship contains seven unknown parameters: λ, α, β, γ, ΔX, ΔY, and ΔZ. On the premise that the scale factor λ≈1, the translation parameters can be eliminated by taking the difference, and the following simple relationship can be further obtained:

[0138]

[0139] Combining the S50 weight matrix, Equation (11), and Equation (12) gives the following error equation:

[0140]

[0141] Among them,

[0142]

[0143] Equation (14) is a singular coefficient matrix, and only 2 are independent in each group of equations. Therefore, it is necessary to satisfy n≥3 for the number of common points. After obtaining the Rodriguez matrix by solving Equations (12) and (13), substitute a pair of corresponding registered key points into Formula (10) to solve the translation vector [ΔX, ΔY, ΔZ] T , to achieve high-precision automatic and precise stitching of transmission towers, refer to Figure 4 - Figure 7 .

[0144] The present invention solves the following deficiencies in the prior art: When airborne lidar is used for power inspection, the point cloud range is large, but the resolution and density are low. Moreover, due to vegetation and environmental occlusion, it is impossible to completely obtain the point cloud of the power pole tower, making it difficult to analyze the abnormal structure of the power pole tower. When terrestrial three-dimensional laser scanning is used for fine three-dimensional reconstruction of power pole towers and extraction of structural parameters, the differences in the scanning distance and incident angle of the target ball and the noise problems caused by other interferences are not considered, resulting in a large error in the target ball fitting result and unable to achieve high-precision registration of multi-station point clouds. The iterative closest point algorithm without using a spherical target requires a large overlapping area between adjacent measurement stations, and the optimal accuracy after registration can only reach the decimeter level. The prior art means requires manual intervention to achieve high-precision registration of the point cloud of transmission power pole towers. Affected by the proficiency of the operator, the registration results vary greatly, and the manual intervention method is time-consuming and laborious.

[0145] The present invention realizes robust automatic registration of the point cloud of multi-station terrestrial three-dimensional laser scanning of transmission power pole towers, weakens the influence of target ball fitting errors caused by scanning distance, incident angle, etc., and makes full use of the intensity information of the target ball and the power pole tower structure, the geometric property differences between the target ball and the power pole tower structure, and the spatial relationship between multi-station target balls to achieve automatic and precise splicing of transmission power pole towers.

[0146] The embodiment of the present invention also provides a precise splicing system for the point cloud of multi-station terrestrial three-dimensional laser scanning of transmission power pole towers, including:

[0147] Intensity clustering unit: used for preprocessing the input point cloud of power pole towers, separating the high-reflection target ball point cloud and the power pole tower structure point cloud, removing the low-reflection point cloud, and clustering each high-reflection point cloud according to the Euclidean distance;

[0148] Curvature estimation unit: Estimate the curvature of the clustered point cloud data, and distinguish the target ball point cloud and the power pole tower structure point cloud by comparing with the corresponding curvature of the known target ball radius;

[0149] Robust estimation unit for target ball parameters: Robustly estimate the separated target ball point cloud containing noise, use the IGGⅢ robust estimation method to reduce the weight of noise points, and thus weaken the influence of noise;

[0150] Target ball relationship correspondence unit: Calculate the distances and angles between the target ball point clouds extracted in each scanning station, and determine the corresponding target balls according to the spatial relationship of the fitting centers of each target ball, which are used as the key points for point cloud registration;

[0151] Corresponding target ball right confirmation unit: Calculate the two-dimensional normal distribution probability density of the corresponding target ball according to the statistical relationship of the target ball fitting values, and determine the weight with this probability density as the diagonal value in the weight matrix to improve the robustness of the system;

[0152] Precise stitching unit: Use at least 3 groups of corresponding target ball points to calculate the Rodriguez matrix, substitute the observed values to solve the translation matrix, and achieve precise stitching of the power pole tower point cloud.

[0153] An embodiment of the present invention further provides a multi-station tower laser point cloud high-precision stitching device, which is characterized by including:

[0154] An intensity clustering module for preprocessing the power pole tower point cloud of multiple scanning stations collected by scanning, and extracting high-intensity point cloud blocks;

[0155] A point cloud separation module for separating the target ball point cloud and the tower structure point cloud from the high-intensity point cloud blocks;

[0156] A robust estimation module for robustly estimating the target ball point cloud and extracting target ball point cloud parameters (including the coordinates of the target ball center point and the target ball radius, the former is used as the key point for multi-station point cloud registration, and the latter is used to determine the key point weight value during registration);

[0157] A target ball relationship corresponding module for calculating the spatial distance and angular relationship between target ball point clouds based on the target ball point cloud parameters, determining the one-to-one correspondence of the target ball point clouds, and obtaining the one-to-one corresponding registration key points;

[0158] A corresponding target ball right confirmation module for calculating the two-dimensional normal distribution probability density according to the target ball point cloud parameters and determining the weight matrix during registration based on the corresponding target ball point cloud;

[0159] A precise stitching module for constructing the Rodriguez matrix, using the registration key points to calculate the spatial relationship of each scanning station, and achieving precise stitching of the multi-station power pole tower point cloud.

[0160] On the other hand, the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the multi-station ground three-dimensional laser scanning transmission tower point cloud precise stitching method described in the first aspect.

[0161] On the other hand, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the multi-station ground three-dimensional laser scanning transmission tower point cloud precise stitching method described in the first aspect.

[0162] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0163] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0164] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. A high-precision stitching method for laser point clouds of multi-station poles and towers, characterized in that, include: Pre-process the point clouds of power towers collected by scanning at multiple scanning stations to extract high-intensity point cloud blocks; Separating the target sphere point cloud and the tower structure point cloud from the high-intensity point cloud block; Robust estimation is performed on the target sphere point cloud to extract target sphere point cloud parameters; Based on the target point cloud parameters, the spatial distance and angle relationship between the target point clouds are calculated, a one-to-one correspondence relationship between the target point clouds is determined, and a one-to-one corresponding registration key point is obtained; Calculate the two-dimensional normal distribution probability density according to the target point cloud parameters, and determine the weight matrix based on the corresponding target point cloud registration; The Rodriguez matrix is constructed, and the spatial relationship of multiple scanning stations is calculated using the registration key points to achieve accurate splicing of multi-station power tower point clouds.

2. The multi-station pole tower laser point cloud high-precision stitching method according to claim 1, wherein, The point clouds of power pole towers collected by scanning at multiple scanning stations are preprocessed to extract high-intensity point cloud blocks, specifically: (1) Statistically calculate the average intensity of the point cloud {P T} Retain the point set with intensity values greater than ; According to the height range of the target ball layout, set an appropriate height threshold, and screen out the point set denoted as {P F}; (2) Randomly set a point P in {P F} as the seed; (3) Search for the point set {P} with the seed point P as the center of the sphere and D as the radius, and calculate the Euclidean distance D from each point P (i = 1, …, n) in {P} to P. Denote the mean of the distances as and the standard deviation as σ; th} and calculate the Euclidean distance D from each point P (i = 1, …, n) in {P} to P. Denote the mean of the distances as and the standard deviation as σ; n}, and calculate the Euclidean distance D from each point P (i = 1, …, n) in {P} to P. Denote the mean of the distances as and the standard deviation as σ; n}, i (i = 1, …, n) to P. Denote the mean of the distances as and the standard deviation as σ; i , and the standard deviation as σ n ; (4) Traverse P i If considers P i to be an outlier, otherwise consider P i to belong to the same class as P; (5) Add the seed point P and the filtered point set {P n} to the current cluster {G j}, set the points in {P n} as new seed points, and repeat steps (3)-(5); (6) When the number of points in {G j} no longer changes, the number of seed points also stops changing. At this time, the set of seed points represents the boundary of the current cluster. Add all boundary points to the current cluster and delete the points in {G j} from {P F}; (7) Repeat steps (2)-(6) until there are no new seed points; (8)Delete clusters with less than N points min A total of J valid clusters are obtained.

3. The multi-station pole tower laser point cloud high-precision stitching method according to claim 1, characterized in that The method of separating the target sphere point cloud and the pole tower structure point cloud from the high-intensity point cloud block specifically includes: estimating the curvature of the point cloud data in the high-intensity point cloud block, and distinguishing the target sphere point cloud from the pole tower structure point cloud by judging the curvature corresponding to the known target sphere radius.

4. The multi-station pole tower laser point cloud high-precision stitching method according to claim 3, wherein, The curvature estimation of the point cloud data in the high-intensity point cloud block is performed, and the target sphere point cloud and the tower structure point cloud are distinguished by judging the curvature corresponding to the known target sphere radius. Specifically, the method includes: In the current cluster {G k}, randomly select t points, fit the local quadratic surface of their neighborhood points, and calculate the local differential geometric properties through formula (1), including the mean curvature Gaussian curvature K and principal curvatures K1 and K2, where K1 is the maximum curvature of a point on the quadratic surface in all directions, K2 is the minimum curvature, and when K1≈K2≠0, it is considered that the point is on the sphere; where L = r xx n, M = r xy n, N = r yy n, E = r x r x , F = r x r y , G = r y r y , r xx , r xy , r yy , r x , r y are the partial differentials of the surface; When more than 90% of the t local surfaces in the cluster {G k} satisfy the curvature constraint condition of formula (2), it can be determined that the current cluster is a target sphere, where ε is the curvature threshold; 5. The multi-station pole and tower laser point cloud high-precision stitching method according to claim 1, characterized in that, The robust estimation of the target sphere point cloud and the extraction of target sphere point cloud parameters are specifically as follows: The criterion for robust estimation is expressed by Equation (3), where is the estimated vector of unknown parameters, w i is the i-th diagonal element of the weight matrix W, a i is the i-th row of the column full-rank design matrix A, V i is the residual vector, L i is the observed value; Differentiate X in Equation (3) and set it to 0, and denote Then we have: Let be the weight factor, be the diagonal element of the equivalent weight matrix, then the solution form of the least squares can be expressed by formula (5); The IGGⅢ method weight function is shown in Equation (6), where k0 and k1 are weight factor adjustment thresholds, is the observed value L i corresponding standardized residual; For the target ball model (x - a) 2 +(y - b) 2 +(z - c) 2 =r 2 , constructing -2ax - 2yb - 2zc + d = x 2 +y 2 +z 2 , where d = a 2 +b 2 +c 2 -r 2 , the least squares matrix form can be obtained as follows: AX=L The target ball point cloud parameters are calculated from Equation (7), including the target ball diameter The center of the target ball where j = 1, 2, …, N s is the serial number of the scanning station, and N s is the total number of scanning stations, k = 1, 2, …, N t is the serial number of the target ball, and N t is the total number of target balls.

6. The multi-station pole tower laser point cloud high-precision stitching method according to claim 5, wherein, The spatial distance and angle relationship between the target point clouds are calculated based on the target point cloud parameters, and the one-to-one correspondence between the target point clouds is determined to obtain the one-to-one corresponding registration key points, specifically: Based on the diameter of the target ball Center of the target ball Calculate the relative position relationship of the corresponding target balls, and calculate the mutual relationship between the target balls within the same measuring station j through Equation (8); where k + m ≤ N t , and m is a natural number greater than 0; is the spatial distance between the k target balls and the k + m target balls, is the included angle; for the target balls with the same corresponding relationship, the relative positions and angular relationships in different measuring stations are the same. Compare the difference between the binary groups and . When both the distance and the angle are less than the thresholds ε d and ε θ , it is considered that the target ball point clouds are in one-to-one correspondence.

7. The multi-station tower laser point cloud high-precision stitching method according to claim 6, wherein The two-dimensional normal distribution probability density is calculated according to the target point cloud parameters, and the weight matrix based on the corresponding target point cloud registration is determined, specifically: The statistical values of the corresponding target ball radii for registration conform to a bivariate normal distribution and are independent of each other. Probability density weighting can be used, that is, the more the fitting radius of any target ball pair deviates from the mean, the smaller its probability density, and the probability density values conform to a normal distribution; according to the target ball diameter Target ball center and the registration key points, calculate the bivariate probability density according to formula (9), where r1 and r2 are the fitting radii of the corresponding target balls, and is the standard deviation of the fitted target ball, r2 is the base of the natural logarithm, is the mean of the fitting radii of all target balls. A set of weights is calculated for all corresponding target balls, and the matrix W with these values as diagonal elements fit is the weight matrix during registration.

8. The high-precision stitching method for laser point clouds of multi-station poles and towers according to claim 1, characterized in that The Rodriguez matrix is constructed, and the spatial relationship of each scanning station is calculated using the registration key points to achieve accurate splicing of multi-station power tower point clouds, specifically: Use the Rodrigues matrix to describe the transformation relationship of the registration station: Among them, (X f , Y f , Z f ) is a registration point in the first coordinate system, (X s , Y s , Z s ) is the corresponding registration point in the second coordinate system, λ is the scale factor, R is the rotation matrix, [ΔX, ΔY, ΔZ] T is the translation vector; The Rodriguez matrix is used to describe the rotation change and is calculated as: Substituting equation (11) into equation (10), the resulting transformation relationship includes seven unknown parameters: λ, α, β, γ, ΔX, ΔY, and ΔZ. Under the premise of scale factor λ≈1, the translation parameters can be eliminated by difference, and the following simple relationship can be further obtained: Combining the weight matrix, equation (11) and equation (12), we can get the following error equation: in, Formula (14) is a singular coefficient matrix, where n ≥ 3; After obtaining the Rodriguez matrix by solving equations (12) and (13), substitute a pair of corresponding registered key points into formula (10) to solve the translation vector [ΔX, ΔY, ΔZ] T , to achieve high-precision automatic and accurate splicing of transmission towers.

9. A high-precision splicing device for laser point clouds of multi-station poles and towers, characterized in that, include: The intensity clustering module is used to pre-process the point clouds of power towers collected by scanning at multiple scanning stations and extract high-intensity point cloud blocks; A point cloud separation module, used to separate the target sphere point cloud and the tower structure point cloud from the high-strength point cloud block; A robust estimation module performs robust estimation on the target sphere point cloud and extracts target sphere point cloud parameters; A target-ball relationship corresponding module is used to calculate the spatial distance and angle relationship between target-ball point clouds based on the target-ball point cloud parameters, determine the one-to-one correspondence between the target-ball point clouds, and obtain one-to-one corresponding registration key points; The corresponding target ball rights confirmation module is used to calculate the two-dimensional normal distribution probability density according to the target ball point cloud parameters, and determine the weight matrix during the registration based on the corresponding target ball point cloud; The precise stitching module is used to construct a Rodriguez matrix, calculate the spatial relationship of each scanning station using the registration key points, and realize the precise stitching of the multi-station power transmission tower point cloud.

10. An electronic device, characterized in that: It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the multi-station ground three-dimensional laser scanning power transmission tower point cloud precise stitching method according to any one of claims 1-8.

Images