A grid structure curved rod detection method based on point cloud data

Abstract

The application discloses a grid structure curved rod detection method based on point cloud data, relates to the technical field of curved rod detection, and discloses that the rod is prone to bending deformation when the space grid structure is used, but the numerous intersecting rods make it difficult for the curved rod to be found by manual point detection. The application provides a grid structure curved rod detection method based on point cloud data, utilizes point cloud features to cluster and segment the point cloud, obtains the cluster point cloud of the ball node and the end part of the nearby rod, and the effective point cloud of the rod; the effective point cloud of the rod is clustered and fitted to obtain the real axis of the rod; the approximate intersection point of the real axis of the rod associated with the center of the roughly fitted ball node is solved, the corrected ball node center is calculated and obtained, and the connecting line of the corrected ball node centers corresponding to the two ends of the rod is the initial axis of the rod; the bending degree of the rod is judged according to the vector height ratio of the real axis of the rod and the initial axis of the rod. The method can improve the accuracy of the detection of the bending deformation of the rod.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of grid structure curved bar detection, and particularly relates to a curved bar detection method based on point cloud data. BACKGROUND

[0002] The spatial grid structure is a typical structure form of large-span public and industrial buildings, and mainly characterized in that the structure is connected by a large number of bars and spherical nodes, and presents a large area of structure coverage. In the process of implementing the green environmental protection concept, the spatial grid structure is widely applied to the raw material yard closed construction of the metallurgy, power, mining and other industries, and the spatial grid structure has the characteristics of a large number of bar nodes, large structure rise and poor visual conditions. In the use process, the structural bars are prone to bending deformation, but the numerous intersecting bars make it difficult for the bending bars to be found by manual point detection.

[0003] At present, the detection of the bending bars of the spatial grid structure mainly includes contact detection and non-contact detection. The contact detection adopts the manual detection method, which is direct but low in efficiency and is easily limited by the structure accessibility. The non-contact detection adopts the three-dimensional laser scanning and digital image methods to obtain the real three-dimensional model of the structure, and the bending bars are screened on the three-dimensional model.

[0004] For the non-contact bending bar screening by the three-dimensional laser scanning, the screening process is as shown in Figure 1 , which usually first performs the segmentation of the spherical node point cloud and the accurate fitting of the spherical node center, and then performs the fitting of the bar axis and the screening of the bending bars according to the fitting results. When the spherical node point cloud is incomplete, the fitting result of the spherical node center obtained by the method is as shown in Figure 2 ; when the bar end is interfered, the fitting result of the spherical node center is as shown in Figure 3 ; when the spherical node fitting is deviated, the fitting schematic diagram of the bar axis is as shown in Figure 4 and Figure 5 . The method fully depends on the quality of the spherical node point cloud and the fitting accuracy of the spherical node center, and is easily affected by the interference of the bar end point cloud, thereby affecting the fitting accuracy. However, due to the factors such as the structure site condition and the equipment precision, it is difficult to ensure the complete scanning of the spherical node and the accurate acquisition of the point cloud, which will greatly affect the subsequent bar fitting and even the bending screening. SUMMARY

[0005] In view of the above problems, the present application aims to provide a grid structure bending bar detection method based on point cloud data, which can detect the bending deformation of the bar, save the calculation time, and improve the fitting efficiency and accuracy of the bar axis.

[0006] The main idea of the technical solution adopted by the application is: through the clustering result with relaxed parameters, the ball nodes in the point cloud of the rod member and the ball nodes are clustered and segmented to obtain the clustered point cloud of the ball nodes and the nearby rod member end and the pure rod member effective point cloud, the real axis of the rod member is fitted through the rod member effective point cloud, and the center of the roughly fitted ball node in the clustered point cloud of the ball nodes and the nearby rod member end is corrected to obtain the corrected ball node center. The method does not depend on high-quality ball node point cloud, improves the robustness of the method, reduces the calculation consumption of the accurate fitting of the ball node, and improves the calculation efficiency.

[0007] In order to achieve the above purpose, the technical solution adopted by the application is as follows:

[0008] The first aspect of the application provides a grid structure curved rod detection method based on point cloud data, comprising the following steps:

[0009] The original point cloud of the grid structure is obtained, and the original point cloud is preprocessed to retain the point cloud of the rod member and the ball node;

[0010] The point cloud features are extracted according to the point cloud of the rod member and the ball node, and the DBSCAN clustering and segmentation of the ball node are performed according to the point cloud features to obtain the clustered point cloud of the ball nodes and the nearby rod member end, and the point cloud of the rod member and the ball node except the clustered point cloud of the ball nodes and the nearby rod member end is the rod member effective point cloud;

[0011] The DBSCAN clustering of the rod member effective point cloud is performed, and the real axis of the rod member is obtained through axis fitting;

[0012] The ball node center is obtained through rough fitting of the clustered point cloud of the ball nodes and the nearby rod member end;

[0013] The approximate intersection of the real axis of the rod member associated with the roughly fitted ball node center is solved, the intersection centroid is calculated, the corrected ball node center is obtained, and the connecting line of the corrected ball node centers corresponding to the two ends of the rod member is the initial axis of the rod member;

[0014] The degree of bending of the rod member is judged according to the real axis of the rod member and the initial axis of the rod member.

[0015] Through the above technical solution, further, the specific steps of extracting the point cloud features according to the point cloud of the rod member and the ball node and performing the DBSCAN clustering and segmentation of the ball node according to the point cloud features include:

[0016] The neighborhood point set of the point cloud of the rod member and the ball node is constructed, the neighborhood covariance matrix is calculated, and the point cloud features are extracted;

[0017] Based on the point cloud features of the ball node, the point cloud region belonging to the ball node is identified;

[0018] The point clouds in the point cloud region belonging to the spherical node are marked as the cluster point clouds of the spherical node and the nearby rod end by using the DBSCAN clustering algorithm.

[0019] By the above technical solution, further, the neighborhood point set of the point cloud of the rod and the spherical node is constructed, the neighborhood covariance matrix is calculated, and the specific steps of extracting the point cloud feature include:

[0020] Suppose a point P i The neighborhood point set of P i ) is: ;

[0021] Wherein, is the three-dimensional coordinates of the first to the kth point in the point cloud; k represents the total number of points in the neighborhood point set, and there are k points in total;

[0022] The neighborhood point set center is: ;

[0023] The neighborhood covariance matrix is: ;

[0024] The neighborhood covariance matrix is subjected to eigenvalue decomposition, and three eigenvalues λ1≤λ2≤λ3 are obtained;

[0025] Wherein, T is a transpose matrix, is the three-dimensional coordinates of any point in the neighborhood point set, are the x-axis, y-axis and z-axis coordinates of any point in the neighborhood point set respectively; j represents the point cloud index in the neighborhood point set; λ1is the dispersion degree of the neighborhood point set in the principal direction, and the principal direction is the axis direction; λ2is the dispersion degree of the neighborhood point set in the secondary direction, and the secondary direction is the tangent direction of the plane; λ3is the dispersion degree of the neighborhood point set in the normal direction, and the normal direction is the normal direction of the plane.

[0026] Through the above technical solution, further: based on the point cloud feature of the spherical node, the specific operation of identifying the point cloud region belonging to the spherical node includes:

[0027] The linear feature L and the scattering feature S corresponding to the point P i are calculated;

[0028] When the neighborhood point set of the point P i satisfies the following conditions, it is determined that the point P i is located in the point cloud region of the spherical node:

[0029] ;

[0030] Wherein, L_threshold is a linear feature threshold, and S_threshold is a scattering feature threshold.

[0031] By the above technical solution, further, the approximate intersection point associated with the ball node center of the rough fitting is solved, the intersection point mass center is calculated, and the specific steps of obtaining the corrected ball node center include:

[0032] Traverse the distance from the end point of all real axes of the rod to any one of the rough fitting ball node centers, if the distance is less than the sum of the diameter of the ball node and the length of the cone head at the proximal end of the current rod, it is determined that the real axis of the rod is associated with the ball node;

[0033] Traverse the real axis pair of the rod associated with the same ball node, that is, the real axis A and the real axis B of the rod associated with the same ball node, calculate the midpoint of the shortest connecting line of the real axis pair as the approximate intersection point, and obtain the approximate intersection point Q= ;

[0034] = + , = + , , are the intersection points of the shortest connecting line of the real axis pair and the real axis A and the real axis B of the rod;

[0035] Wherein, is the passing point of the real axis A of the rod, is the passing point of the real axis B of the rod, is the direction vector of the real axis A of the rod, is the direction vector of the real axis B of the rod, , is the direction vector of the shortest connecting line of the real axis pair;

[0036] = , = , = , = , = ; , , , , are the dot products of the defined direction vectors;

[0037] approximate intersection point of all bar real axes of the ball node The cluster with the largest number of samples is obtained by using DBSCAN clustering, and the corrected ball node center is calculated as:

[0038] ;

[0039] wherein, is the n th approximate intersection point, and n is the total number of approximate intersection points.

[0040] By the above technical solution, the formula for calculating the rise ratio is:

[0041] ;

[0042] wherein, is the length of the initial axis of the bar, is the bending rise, = , is the distance from the point on the real axis of the bar to the initial axis of the bar is the rise ratio of the bar.

[0043] The second aspect of the application provides a grid structure curved bar detection system based on point cloud data, comprising: a computer readable storage medium and a processor.

[0044] The computer readable storage medium is used to store executable instructions.

[0045] The processor is used to read the executable instructions stored in the computer readable storage medium and execute the grid structure curved bar detection method based on point cloud data as described above.

[0046] The third aspect of the application provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the grid structure curved bar detection method based on point cloud data as described above.

[0047] The beneficial effects of the application are:

[0048] The application clusters and segments the ball nodes in the point cloud of the bar and the ball node through the clustering result with relaxed parameters, obtains the clustered point cloud of the ball node and the nearby end of the bar and the pure bar effective point cloud, and fits the real axis of the bar through the bar effective point cloud. In this way, the influence of the point cloud of the end of the bar such as the cone head on the fitting of the bar axis is avoided, the calculation time is greatly saved, and the fitting efficiency of the bar axis is improved.

[0049] The intersection of the real axis of the rod is used to correct the center of the roughly fitted ball node, which is more in line with the actual construction on site from the aspects of field installation and node connection, and avoids the problems of low fitting accuracy of the ball node center, deviation of the fitting center from the rod and the like through correction, improves the fitting accuracy of the ball node center, and accurately positions the initial axis of the rod.

[0050] In the existing bending degree judgment of the rod, the bending height is often used to distinguish and judge the bending degree of the rod. This method cannot consider the influence of the length of the rod on the bending degree, and under this judgment method, the error conclusion that the bending degrees of rods with large length difference are the same is easily obtained. The rod height ratio is used as the bending judgment index in the application, so that rods of different lengths can be compared and screened fairly with the same standard.

[0051] The application obtains the point cloud of the spatial grid structure through the three-dimensional laser scanning technology, and pre-processes the data. Through the unique "ball node influence domain isolation-rod real axis fitting-ball node center correction" collaborative technical route, the full-automatic and high-precision analysis of the point cloud data of the spatial grid structure is realized, and the rod height ratio is used as the bending judgment index, which greatly improves the automation level of the structure health detection and the reliability of the evaluation results. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The existing non-contact bending rod screening process schematic diagram in the background art;

[0053] Figure 2 The fitting result schematic diagram of the ball node center when the ball node point cloud is incomplete in the background art;

[0054] Figure 3 The fitting result schematic diagram of the ball node center when the rod end is disturbed in the background art;

[0055] Figure 4 The fitting schematic diagram of the rod axis after the ball node fitting deviation in the background art;

[0056] Figure 5 The fitting schematic diagram of another rod axis after the ball node fitting deviation in the background art;

[0057] Figure 6 The principle flow schematic diagram of the application;

[0058] Figure 7 The ball node influence domain schematic diagram of the application;

[0059] Figure 8 The schematic diagram of obtaining the effective point cloud of the rod of the application;

[0060] Figure 9Fig. 1 is a schematic diagram of the center of the rough fitting ball node of the present application;

[0061] Figure 10 Fig. 2 is a schematic diagram of the center of the modified ball node of the present application;

[0062] Figure 11 Fig. 3 is a schematic diagram of the accurate connection of the initial axis of the member after the center of the modified ball node of the present application;

[0063] Figure 12 Fig. 4 is a schematic diagram of the accurate connection of the initial axis of another member after the center of the modified ball node of the present application;

[0064] Figure 13 Fig. 5 is a schematic diagram of the real axis of the member and the initial axis of the member of the present application; the red line is the initial axis of the member, and the blue line is the real axis of the member. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0066] The present application proposes a mesh structure curved member detection method based on point cloud data. The ball nodes in the point clouds of the members and the ball nodes are clustered and segmented to obtain the clustered point clouds of the ball nodes and the nearby member end parts and the pure member effective point clouds. The real axis of the member is fitted through the member effective point clouds, and the center of the rough fitting ball node in the clustered point clouds of the ball nodes and the nearby member end parts is corrected to obtain the center of the modified ball node. The method does not depend on high-quality ball node point clouds, improves the robustness of the method, reduces the calculation consumption of the accurate fitting of the ball nodes, and improves the calculation efficiency.

[0067] Embodiment one: refer to Figures 6-13 The present application discloses a mesh structure curved member detection method based on point cloud data, and the main technical scheme is as follows:

[0068] Step one: obtaining the original point cloud of the mesh structure, and pre-processing the original point cloud to retain the point clouds of the members and the ball nodes.

[0069] The original point cloud of the mesh structure is obtained, and the original point cloud is voxel down-sampled and statistically filtered to remove noise and reduce the calculation amount. The irrelevant points (noise points) in the original point cloud, including the point clouds of the roof panels, equipment and other objects irrelevant to the detection, are deleted, and only the point clouds of the members and the ball nodes are retained.

[0070] Step 2: Extract point cloud features from the point clouds of the rods and ball nodes, and perform DBSCAN clustering and segmentation on the ball nodes based on the point cloud features to obtain the clustered point clouds of the ball nodes and the nearby rod ends. The point clouds of the rods and ball nodes, excluding the clustered point clouds of the ball nodes and the nearby rod ends, are the valid point clouds of the rods.

[0071] The specific steps are as follows:

[0072] (1) Construct the neighborhood point set of the point cloud of the rod and ball node, calculate the neighborhood covariance matrix, and extract the point cloud features;

[0073] Let point P be... i The neighborhood point set is

[0074] N(P i )= ;

[0075] in, The coordinates of the first to the kth points in the point cloud are given; k represents the total number of points in the neighborhood set, which is k points in total.

[0076] Then the center of the neighborhood point set: ;

[0077] Neighborhood covariance matrix: ;

[0078] Eigenvalue decomposition of the neighborhood covariance matrix yields three eigenvalues: λ1≤λ2≤λ3;

[0079] Where T is the transpose matrix. Let be the three-dimensional coordinates of any point in the neighborhood point set. , λ1 represents the x-axis, y-axis, and z-axis coordinates of any point in the neighborhood point set, respectively; j represents the point cloud index in the neighborhood point set, ranging from 1 to k; λ1 represents the degree of dispersion of the corresponding neighborhood point set in the principal direction, which is the axis direction; λ2 represents the degree of dispersion of the corresponding neighborhood point set in the secondary direction, which is the tangent direction of the plane; λ3 represents the degree of dispersion of the corresponding neighborhood point set in the normal direction, which is the normal direction of the plane.

[0080] (2) Based on the point cloud features of the sphere nodes, identify the point cloud regions belonging to the sphere nodes;

[0081] Based on the point cloud features of the sphere nodes, the point cloud regions belonging to the sphere nodes are initially identified using the linear feature threshold L_threshold and the scattering feature threshold S_threshold, which are the sphere node influence domains.

[0082] Calculate point P i The corresponding linear feature L= and scattering characteristics S= .

[0083] When the neighborhood point set of point P i satisfies the following conditions, it is determined that point P i is located in the point cloud region of the spherical node:

[0084] .

[0085] Generally, 0.4 can be taken, 0.25 can be taken.

[0086] (3) The DBSCAN clustering algorithm is used to divide the point cloud of the spherical node into independent spherical node candidate clusters. The key is that the clustering parameters in this stage are relatively loose, and are intended to ensure that all spherical node regions (including the point cloud of the end part of the rod that may be adhered) are completely included, forming a "spherical node influence domain".

[0087] (4) Based on the spherical node clustering result, all points in the "spherical node influence domain" in the point cloud are marked as spherical node point clouds, and are divided. This operation is equivalent to dividing the point cloud into two parts, obtaining the spherical node influence domain and the pure rod effective point cloud, and the spherical node influence domain is the clustering point cloud of the spherical node and the end part of the rod nearby, so that the rod effective point cloud can exclude the end part interference.

[0088] Step three: DBSCAN clustering is performed on the rod effective point cloud, and the real axis of the rod is obtained through axis fitting.

[0089] DBSCAN clustering is performed on the rod effective point cloud, and a single rod is separated. For the point cloud of each rod, a segmented center fitting and spline interpolation algorithm is used for axis fitting:

[0090] (1) The projection of the point cloud of each rod along its main direction is segmented.

[0091] (2) In each segment, the point cloud is projected onto a plane perpendicular to the main direction, and the least square method is used to fit the circular cross section to obtain the center and radius of the segment.

[0092] (3) The center points of each segment fitted by the circle are fitted by a three-dimensional spline curve based on the cumulative chord length parameterization, and a smooth and accurate real axis of the rod is finally generated.

[0093] Step four: coarse fitting is performed on the clustering point cloud of the spherical node and the end part of the rod nearby, and a coarsely fitted spherical node center is obtained.

[0094] For each neighborhood point set center of the preliminary detection of each spherical node, it is taken as the coarsely fitted spherical node center.

[0095] Step five: solving the approximate intersection point of the real axis of the bar associated with the ball node center of the rough fitting, calculating the intersection point mass center, obtaining the corrected ball node center, and the connecting line of the corrected ball node centers corresponding to the two ends of the bar is the initial axis of the bar.

[0096] Since the ball node center of the rough fitting has errors due to the influence of the point cloud data, the real axis of the bar should intersect at the center of the ball node, so the intersection point of the real axis of the bar is calculated to correct the ball node center of the rough fitting:

[0097] First, determine the associated real axis of the bar, traverse the distance from the end point of all fitted real axes of the bar to If the distance is less than the connection threshold, which is the sum of the diameter of the ball node and the length of the cone head at the near end of the current bar, it is considered that the real axis of the bar is associated with the ball node. Among them, the near end of the current bar refers to the end of the two ends of the current bar closer to the ball node.

[0098] The cone head is a conical structure, the large diameter end is welded with the steel pipe, and the small diameter end is connected with the bolt ball through high-strength bolts and sleeves. The bar refers to the state after the steel pipe is welded with the cone head at both ends.

[0099] Traverse the real axis pair of the bar associated with the same ball node, i.e. the real axis A of the bar associated with the same ball node and the real axis B of the bar, calculate the midpoint of the shortest connecting line of the real axis pair as the approximate intersection point. Assuming that the passing point S A of the real axis A of the bar is , the direction vector of the real axis A of the bar is B , the passing point S A of the real axis B of the bar is , and the direction vector of the real axis B of the bar is ;

[0100] ; The direction vector of the shortest connecting line of the real axis pair is

[0101] The definition coefficient = , = , = , = , = ; , , , , are the dot products of the defined direction vectors, respectively.

[0102] Then the two points on the shortest connecting line are P A =SA + , P B =S B + , , are the shortest line of the real axis pair of the rod and the intersection point of the real axis A of the rod and the real axis B of the rod, respectively;

[0103] The approximate intersection point Q of all real axes of the ball node is .

[0104] The approximate intersection point Q of all real axes of the ball node is , and the cluster with the largest number of samples is obtained by DBSCAN clustering, and the centroid thereof is calculated as the corrected ball node center:

[0105] .

[0106] wherein, is the nth approximate intersection point, and n is the total number of approximate intersection points.

[0107] Step six: bending rod discrimination. The present application proposes a bending evaluation index independent of the length of the rod, i.e., the rise ratio, which normalizes the bending degree to the length of the rod, so that rods of different lengths can be compared and screened fairly by the same standard.

[0108] For a single rod, the initial axis of the undeformed rod is the line connecting the corrected ball node centers at both ends, and the length of the rod is the length of the initial axis of the rod .

[0109] Correspondingly, the distance of the points on the real axis of the rod to the initial axis of the rod is calculated , and the bending rise of the rod is ; . The rise ratio of the rod is the rise ratio of the rod.

[0110] The calculation formula of the rise ratio of the rod is:

[0111] .

[0112] By setting a unified rise ratio threshold, rods of different lengths can be screened by a unified standard.

[0113] ​The application does not need to accurately fit the center of the spherical node, and the effective point cloud of the rod can be quickly obtained by removing the "spherical node influence domain" around the spherical node, and the real axis of the rod can be obtained by clustering and fitting the rod point cloud. In this way, the influence of the rod end point cloud such as the cone head on the fitting of the rod axis is avoided, and the calculation time is greatly saved, and the fitting efficiency of the rod axis is improved.

[0114] The intersection of the real axis of the rod is used to correct the center of the spherical node, which is more in line with the actual construction from the aspects of field installation and node connection, and the problems of low fitting accuracy of the spherical node center and deviation of the fitting center from the rod are avoided through correction, thereby improving the fitting accuracy of the spherical node center and accurately positioning the initial axis of the rod.

[0115] The existing method mainly uses the bending vector height to distinguish and judge the bending degree of the rod when judging the bending degree of the rod, which cannot consider the influence of the length of the rod on the bending degree, and under this judgment method, the error conclusion that the bending degrees of rods with large length difference are the same is easily obtained. The rod vector height ratio is used as the bending judgment index in the application, so that rods of different lengths can be compared and screened fairly with the same standard.

[0116] Table 1 Comparison of evaluation of bending vector height by traditional method and evaluation of vector height ratio by the application

[0117]

[0118] The application obtains the point cloud of the spatial grid structure by the three-dimensional laser scanning technology, and pre-processes the data. Through the unique "spherical node region isolation-real axis fitting of the rod-spherical node center correction" collaborative technical route, the full-automatic and high-precision analysis of the point cloud data of the spatial grid structure is realized, and the rod vector height ratio is used as the bending judgment index, which is scientific and unified, and greatly improves the automation level of the structure health detection and the reliability of the evaluation results.

[0119] The basic principles, main features and advantages of the application are shown and described. Those skilled in the art should understand that the application is not limited to the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the application. Without departing from the spirit and scope of the application, various changes and improvements can be made to the application, and these changes and improvements all fall within the scope of the claimed application. The scope of protection of the application is defined by the appended claims and their equivalents.

Claims

1. A method for detecting bent members in a mesh structure based on point cloud data, characterized in that, Includes the following steps: Obtain the original point cloud of the mesh structure and preprocess the original point cloud to retain the point clouds of the rod and ball nodes; Point cloud features are extracted from the point clouds of the rods and ball nodes, and DBSCAN clustering and segmentation are performed on the ball nodes based on the point cloud features to obtain the clustered point clouds of the ball nodes and the ends of the nearby rods. The point clouds of the rods and ball nodes, excluding the clustered point clouds of the ball nodes and the ends of the nearby rods, are the effective point clouds of the rods. DBSCAN clustering is performed on the effective point cloud of the rod, and the true axis of the rod is obtained by axis fitting. A coarse fit is performed on the clustered point cloud of the ball node and the ends of nearby members to obtain the coarsely fitted ball node center; Find the approximate intersection point of the true axis of the member associated with the coarsely fitted ball node center, calculate the centroid of the intersection point, obtain the corrected ball node center, and the line connecting the corrected ball node centers at both ends of the member is the initial axis of the member. Specifically, the distance from the endpoint of the true axis of all members to the center of any coarsely fitted spherical node is traversed. If the distance is less than the sum of the diameter of the spherical node and the length of the cone at the near end of the current member, then the true axis of the member is determined to be associated with the spherical node. Traverse the pairs of true axis lines of the members associated with the same ball node, i.e., true axis lines A and B of the members associated with the same ball node, and calculate the midpoint of the shortest line connecting the pairs of true axis lines. Use this midpoint as the approximate intersection point to obtain the approximate intersection point Q. ; = + , = + , , These are the intersection points of the shortest line connecting the true axes of the members with the true axis A and the true axis B of the members, respectively. in, The point through which the true axis A of the rod passes. The point through which the true axis B of the member passes. Let A be the direction vector of the true axis A of the member. Let B be the direction vector of the true axis of the member. , The direction vector of the shortest line connecting the true axes of the member; = , = , = , = , = ; , , , , These are the dot products of the defined direction vectors; The approximate intersection of the true axes of all the members of the ball joint. DBSCAN clustering was used to obtain the cluster with the largest number of samples, and the corrected sphere node center was calculated as: C refined = ; in, This is the nth approximate intersection point, where n is the total number of approximate intersection points; The degree of bending of the member is determined by the sag ratio obtained from the true axis and the initial axis of the member.

2. The method for detecting bent members in a mesh structure based on point cloud data according to claim 1, characterized in that, The specific steps for extracting point cloud features from the point clouds of the members and ball nodes, and performing DBSCAN clustering and segmentation on the ball nodes based on the point cloud features to obtain the clustered point cloud of the ball nodes and the ends of nearby members include: Construct neighborhood point sets for the point clouds of rods and ball nodes, calculate the neighborhood covariance matrix, and extract point cloud features; Based on the point cloud features of sphere nodes, point cloud regions belonging to sphere nodes are identified. The DBSCAN clustering algorithm is used to mark the point cloud regions belonging to the ball nodes as clustered point clouds of the ball nodes and the ends of nearby members.

3. The method for detecting bent members in a mesh structure based on point cloud data according to claim 2, characterized in that, The specific steps for constructing the neighborhood point set of the point cloud for rods and spheres, calculating the neighborhood covariance matrix, and extracting point cloud features include: Let point P be... i The neighborhood set of points is: N(P) i )= ; in, The coordinates of the first to the kth points in the point cloud are 3D coordinates; k represents the total number of points in the neighborhood point set. Then the center of the neighborhood point set: ; Neighborhood covariance matrix: ; Eigenvalue decomposition of the neighborhood covariance matrix yields three eigenvalues: λ1≤λ2≤λ3; Where T is the transpose matrix. Let be the three-dimensional coordinates of any point in the neighborhood point set. λ1 represents the x-axis, y-axis, and z-axis coordinates of any point in the neighborhood point set, respectively; j represents the point cloud index in the neighborhood point set; λ1 represents the degree of dispersion of the corresponding neighborhood point set in the principal direction, which is the axis direction; λ2 represents the degree of dispersion of the corresponding neighborhood point set in the secondary direction, which is the tangent direction of the plane; λ3 represents the degree of dispersion of the corresponding neighborhood point set in the normal direction, which is the normal direction of the plane.

4. The method for detecting bent members in a mesh structure based on point cloud data according to claim 3, characterized in that, Based on the point cloud features of sphere nodes, the specific operations for identifying point cloud regions belonging to sphere nodes include: Calculate point P i The corresponding linear feature L= and scattering characteristics S= ; When point P i A point P is determined when its neighborhood set satisfies the following conditions. i Point cloud region located at the sphere node: ； Where L_threshold is the linear feature threshold and S_threshold is the scattering feature threshold.

5. The method for detecting bent members in a mesh structure based on point cloud data according to claim 4, characterized in that, The formula for calculating the sag-to-height ratio is: ； in, The length of the initial axis of the rod. For the curvature, = , Points on the true axis of the rod Distance to the initial axis of the member; Let be the sag ratio of the member.

6. A grid structure bending member detection system based on point cloud data, characterized in that, include: Computer-readable storage media and processors; Computer-readable storage media are used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the mesh structure bending member detection method based on point cloud data according to any one of claims 1-5.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for detecting bent members in a mesh structure based on point cloud data as described in any one of claims 1-5.

Images