Bridge linear measurement method based on vehicle-mounted three-dimensional laser scanning

Abstract

The application provides a bridge linear measurement method based on vehicle-mounted three-dimensional laser scanning, which comprises the processes of original point cloud data acquisition, original data preprocessing, horizontal approximate surface point cloud extraction, bridge deck surface point cloud extraction and bridge linear feature point extraction; the application extracts complete bridge deck surface point cloud based on original data segments, and then obtains bridge deck outer boundary point cloud based on the bridge deck surface point cloud. The application acquires bridge linear based on three-dimensional laser scanning, and the method belongs to non-contact measurement, the field measurement time is short, the linear data extraction automation degree is high, the measurement result resolution is high, the measurement linear is continuous, and the method can be applied to long-term linear monitoring of large-scale bridge groups and long bridges.

Description

Technical Field

[0001] This invention relates to the field of bridge alignment measurement technology, and specifically to a bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning. Background Technology

[0002] Bridge beam alignment measurement is an important part of bridge operation and maintenance. It can reflect problems that will be encountered during the bridge operation phase, such as beam deformation and pier settlement. It is a key parameter for bridge geometric condition detection and plays a vital role in evaluating the bridge's operational status.

[0003] Traditional bridge alignment measurement techniques mainly include geometric leveling and total station surveying. These methods obtain bridge alignment by measuring a few discrete points. The smoothness of the fit depends on the density of the measuring points. Limited by measurement time and ease of implementation, the number of measuring points is generally small, thus failing to accurately reflect the continuous alignment of the bridge and easily overlooking structural defects. Furthermore, most traditional measurement methods are contact-based, requiring station relocation and setup, making the process cumbersome and time-consuming. For long-term health monitoring of large-scale bridge structures, traditional techniques struggle to achieve rapid and continuous alignment detection, both in terms of technological advancement and ease of measurement.

[0004] Vehicle-mounted mobile measurement technology is one of the cutting-edge technologies in the surveying and mapping industry today. Belonging to the mobile 3D laser measurement system, it combines Global Navigation Satellite System (GNSS) positioning technology, Inertial Navigation (INS) technology, digital photogrammetry technology, and laser scanning technology. Its field scanning measurement process is simple, the measurement time is short, the degree of automation is high, and it acquires abundant spatial coordinate data of object surfaces, exhibiting significant advantages. Leveraging these advantages, this application aims to provide a bridge alignment measurement method based on vehicle-mounted 3D laser scanning, achieving continuous, rapid, and high-precision detection of bridge alignment, thereby addressing the shortcomings of existing bridge alignment measurement methods. Summary of the Invention

[0005] The purpose of this invention is to provide a bridge alignment measurement method based on vehicle-mounted 3D laser scanning, aiming to solve the problems of long measurement time, low automation, and discontinuous measurement results in current bridge alignment measurement methods. The specific technical solution is as follows:

[0006] A method for measuring bridge alignment based on vehicle-mounted 3D laser scanning includes the following steps:

[0007] Step S1: Acquire raw point cloud data using a vehicle-mounted 3D laser scanner;

[0008] Step S2: Segment the original point cloud data to obtain the original bridge deck point cloud data; downsample the original bridge deck point cloud data to obtain the downsampled bridge deck point cloud data.

[0009] Step S3: Using the same data segmentation method, divide both the downsampled bridge surface point cloud data and the original bridge surface point cloud data into multiple data segments; and execute steps S4-S6 with each downsampled data segment and its corresponding original data segment as the processing object.

[0010] Step S4: Apply point cloud normal filtering to the downsampled data segment to remove data points whose deviation between the normal direction and the vertical direction is greater than the normal deviation threshold K1; then apply the region growing algorithm to cluster the discrete point cloud data to obtain multiple clustered region sets; merge the same horizontal approximate surface point clouds according to the Z-axis coordinates of the centroid points of each region set to obtain multiple horizontal approximate surface point clouds.

[0011] Step S5: Select the horizontal approximate surface point cloud with the largest number of interior points, and use least squares fitting to obtain the planar mathematical model of the horizontal approximate surface point cloud. Based on this planar mathematical model, extract the complete bridge deck surface point cloud P from the original data segment. fi ;

[0012] Step S6: Assess the point cloud P on the upper surface of the bridge deck. fi Bridge deck boundary point cloud acquisition is performed to identify all boundary point clouds along the longitudinal direction of the bridge; based on the positional differences of all boundary point clouds along the longitudinal direction of the bridge in the width direction, the outer boundary line is identified as the bridge linear feature point.

[0013] Step S7: Integrate the bridge alignment feature point extraction results of the point cloud on the upper surface of each bridge deck segment to obtain the complete bridge alignment feature points.

[0014] In the preferred embodiment of the above technical solution, the data segmentation in step S3 specifically involves: performing principal component analysis on the X and Y axis coordinates of the point cloud data to obtain the directions of the first principal axis and the second principal axis; transforming the X and Y coordinates of the point cloud data along the principal axis directions, with the first principal axis direction as the X direction and the second principal axis direction as the Y direction; calculating the difference between the maximum and minimum values ​​of the transformed point cloud data along the second principal axis and using the difference as the bridge deck width; and segmenting the transformed point cloud data along the first principal axis direction, with the segment length set to k times the bridge deck width, where k = 2.0 to 2.5.

[0015] In the preferred embodiment of the above technical solution, step S4, merging the point clouds of the same horizontal approximate surface based on the Z-axis coordinates of the centroid of the region set, specifically involves:

[0016] Step A1: Set the distance threshold K3, calculate the Z-axis coordinate of the centroid of each region after clustering; and number each region sequentially starting from 1; let n = 1, m = n + 1, and MAX be the number of regions after clustering.

[0017] Step A2: Calculate the absolute value of the difference between the Z-axis coordinates of the centroids of regions n and m;

[0018] If the absolute value is greater than or equal to the distance threshold K3, proceed to step A4;

[0019] If the absolute value is less than the distance threshold K3, then merge the m region set into the n region set, delete the m region set, update MAX = MAX-1, reduce the number of the region set with the number greater than m by 1, and then proceed to step A3.

[0020] Step A3: If m is less than MAX+1, proceed to step A2; if m is not less than MAX+1, proceed to step A5.

[0021] Step A4: If m is less than MAX, make m = m + 1 and return to step A2; if m is not less than MAX, proceed to step A5.

[0022] Step A5: If n is less than MAX-1, set n = n+1 and m = n+1 and return to step A2; if n is not less than MAX-1, then end the loop.

[0023] In the preferred embodiment of the above technical solutions, the X and Y coordinates of the point cloud data are transformed along the principal axis according to equation (2);

[0024]

[0025] In equation (2): (x c ,y c ) T The coordinates of the point cloud data after coordinate transformation along the principal axis direction are (x p ,y p ) T To transform the coordinates of the point cloud data, Let A be the coordinate transformation matrix, where (A) F1 A F2 ) T For the first principal axis direction, (A) S1 A S2 ) T This is the direction of the second principal axis.

[0026] In the preferred embodiment of the above technical solution, in step S5, a point cloud set P is first coarsely extracted from the original data segment using a planar mathematical model. co Then, in the point cloud collection P co The complete point cloud P on the upper surface of the bridge deck was extracted.fi .

[0027] The preferred technical solution among the above is the coarse extraction point cloud collection P. co Specifically, a distance threshold L1 is set, and the distance from the data point in the original data segment to the plane mathematical model p is calculated according to equation (9). Data points with a distance less than the distance threshold L1 are considered to belong to the upper surface of the bridge deck, so as to achieve coarse extraction.

[0028]

[0029] In equation (9), d(a,p) is the distance from data point a to the planar mathematical model p, and X a Y a Z a These are the coordinates of data point a after the transformation.

[0030] The preferred technical solution above is the one that precisely extracts the complete point cloud P from the upper surface of the bridge deck. fi Specifically: Calculate the point cloud set P co The normal and curvature of the data points are set, the normal evaluation weight W is set, and the comprehensive distance evaluation index D(a,p) is calculated using equations (10)-(12). If D(a,p) is less than the distance threshold L1, the data points are considered to belong to the upper surface of the bridge deck, thus achieving accurate extraction.

[0031] W a =W(1-C) a (10),

[0032] d N (a,p)=cos -1 (|N a ·N P |) (11),

[0033] D(a,p)=|(1-W a ')d(a,p)+W a 'd N (a,p)| (12),

[0034] In equations (10)-(12), C a Let W be the curvature of data point a. a 'To evaluate the normal of data point a considering the effect of curvature, d N (a, p) represents the deviation between the normal to data point a and the normal to the plane mathematical model p, where N is the deviation. a N P These are the unit normal vectors of data point a and the plane mathematical model p, respectively.

[0035] In the preferred embodiment of the above technical solutions, in step S6: the point cloud P on the upper surface of the bridge deck is processed using an angle criterion algorithm.fi The process involves obtaining the boundary point cloud of the bridge deck; predefining a straight line model; and applying a random sampling consensus algorithm to identify all boundary line point clouds with an angle less than K5 with the X-axis and a number of interior points greater than the minimum interior point threshold of the longitudinal boundary point cloud as the boundary point cloud along the longitudinal direction of the bridge, where K5 is the allowable angle of deviation from the X-axis; and identifying the outer boundary line as the bridge linear feature point based on the positional differences of the boundary point cloud along the longitudinal direction of the bridge in the Y-axis direction.

[0036] In the preferred embodiment of the above technical solutions, the linear model is shown in equation (13):

[0037]

[0038] In equation (13), (A,B,C) T is the unit direction vector of the line; (a,b,c) are the coordinates of a point on the line model.

[0039] In the preferred embodiment of the above technical solution, the y-coordinate value of the intersection point of the boundary point cloud along the longitudinal direction of the bridge and the plane X = 0 is calculated according to formula (14), and the longitudinal boundary point cloud corresponding to the maximum and minimum values ​​is taken as the bridge linear feature point.

[0040]

[0041] In equation (14), Y X＝0 The y-coordinate of the intersection point of the longitudinal boundary point cloud and the plane X = 0.

[0042] The application of the technical solution of the present invention has the following beneficial effects:

[0043] This invention is based on three-dimensional laser scanning to obtain bridge alignment. This method is a non-contact measurement method with short field measurement time, high degree of automation in alignment data extraction, high resolution of measurement results, and continuous measurement alignment. For a long bridge of one kilometer, the measurement time of the vehicle-mounted laser scanner is within 12 minutes. It can be applied to long-term alignment monitoring of large-scale bridge groups and long bridges.

[0044] Based on the characteristics of point cloud data on the upper surface of bridge deck, this invention proposes a novel method for extracting bridge alignment feature points. This method can extract bridge alignment based on vehicle-mounted 3D laser point cloud data. In practical applications, corresponding programs can be developed based on this method to achieve rapid extraction of bridge alignment data.

[0045] This invention utilizes downsampled bridge deck point cloud data to extract the horizontal approximate surface point cloud with the largest number of interior points, and uses least squares fitting to obtain the planar mathematical model of the horizontal approximate surface point cloud. Based on this planar mathematical model, coarse extraction of the point cloud on the upper surface of the bridge deck is achieved, which has higher computational efficiency and can realize bridge alignment data extraction more quickly.

[0046] After establishing a planar mathematical model, this invention extracts the complete point cloud of the upper surface of the bridge deck based on the original data segment, and then obtains the point cloud of the bridge deck boundary based on the point cloud of the upper surface of the bridge deck. This invention obtains linear feature points based on the original data, resulting in better smoothness of the bridge alignment fitting and better reflection of the bridge alignment.

[0047] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0048] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0049] Figure 1 This is a flowchart of the bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to the present invention;

[0050] Figure 2 This is the original point cloud data map of the bridge;

[0051] Figure 3a This is a schematic diagram of the original point cloud data of the bridge before segmentation;

[0052] Figure 3b This is a schematic diagram of the segmented original point cloud data of the bridge;

[0053] Figure 4a This is a schematic diagram of the original point cloud data of the bridge before downsampling;

[0054] Figure 4b This is a schematic diagram of the bridge's original point cloud data after downsampling;

[0055] Figure 5 This is a schematic diagram of point cloud data segmentation;

[0056] Figure 6a This is a schematic diagram before the extraction of the horizontal approximate surface point cloud;

[0057] Figure 6b This is a schematic diagram after extracting the point cloud of the approximate horizontal surface;

[0058] Figure 7a This is a schematic diagram before point cloud extraction on the bridge surface;

[0059] Figure 7b This is a schematic diagram after the point cloud of the bridge surface has been extracted;

[0060] Figure 8 This is a diagram showing the results of extracting the feature points of the bridge's alignment.

[0061] Figure 9 This is a diagram showing the results of the full bridge alignment extraction. Detailed Implementation

[0062] To facilitate understanding of the present invention, a more comprehensive description is provided below, along with preferred embodiments. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the present invention.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0064] Example 1:

[0065] See Figure 1 A method for measuring bridge alignment based on vehicle-mounted 3D laser scanning includes the following steps:

[0066] Step S1: Acquire raw point cloud data using a vehicle-mounted 3D laser scanner;

[0067] Preferably, step S1 specifically involves: the vehicle-mounted 3D laser integrated device acquiring the vehicle's GPS location information, vehicle's IMU spatial attitude information, and 3D laser scanning data, and using equation (1) to calculate the original point cloud data.

[0068]

[0069] In equation (1): (x p ,y p ,z p ) T This provides the geodetic coordinate data of the three-dimensional laser point. This is the rotation matrix from the local horizontal coordinate system to the geodetic coordinate system. This is the rotation matrix from the IMU coordinate system to the local horizontal coordinate system. Let x be the rotation matrix from the laser scanner coordinate system to the IMU coordinate system. L y L , z L ) T Let be the coordinates of the laser point in the laser scanner coordinate system. These are the coordinates of the laser scanner center in the IMU coordinate system. The coordinates of the IMU center in the geodetic coordinate system.

[0070] Step S2: Segment the original point cloud data to obtain the original bridge deck point cloud data; downsample the original bridge deck point cloud data to obtain the downsampled bridge deck point cloud data.

[0071] Preferably, the original point cloud data is manually segmented using open-source point cloud processing software to remove a large amount of point cloud data that is not related to the bridge, including ground, trees, and irrelevant buildings, leaving complete original bridge surface point cloud data. This segmentation process is a prerequisite for subsequent operations. The segmentation operation does not require high precision and takes little time, which helps subsequent steps to accurately extract the bridge linear data points.

[0072] Preferably, the original bridge deck point cloud data downsampling adopts the mature voxelized mesh downsampling technology, which divides the three-dimensional space of the point cloud into voxels, calculates the centroid of each non-empty voxel to replace all points within the voxel, and realizes point cloud downsampling.

[0073] Step S3: Using the same data segmentation method, both the downsampled bridge surface point cloud data and the original bridge surface point cloud data are divided into N data segments; Steps S4-S6 are executed with each downsampled data segment and its corresponding original data segment as the processing object; those skilled in the art will understand that here the downsampled data segment is a data segment in the bridge surface point cloud data, and the original data segment is a data segment in the original bridge surface point cloud data;

[0074] Preferably, when segmenting the data, principal component analysis is performed on the X and Y axis coordinates of the downsampled bridge surface point cloud data or the original bridge surface point cloud data to obtain the first principal axis (the eigenvector corresponding to the large eigenvalue) and the second principal axis direction (the eigenvector corresponding to the small eigenvalue). The X and Y coordinates of the point cloud data are transformed along the principal axis direction. After transformation, the first principal axis direction is taken as the X direction and the second principal axis direction is taken as the Y direction. The difference between the maximum and minimum values ​​of the transformed point cloud data on the second principal axis is calculated and the difference is taken as the bridge surface width. The transformed point cloud data is segmented along the first principal axis direction, and the segment length is set to k times the bridge surface width, where k = 2.0 to 2.5.

[0075] The detailed operation of the data segmentation is shown in equations (2)-(6):

[0076]

[0077] W = y c max -y c min (3),

[0078] L=kW(k=2.0~2.5) (4),

[0079]

[0080]

[0081] In equations (2)-(6): (x c ,y c ) T The coordinates of the point cloud data after coordinate transformation along the principal axis direction are (x p ,y p ) T To transform the coordinates of the point cloud data, Let A be the coordinate transformation matrix, where (A) F1 A F2 ) T For the first principal axis direction, (A) S1 A S2 ) T The second principal axis direction, W is the bridge deck width, y c max y c min These represent the maximum and minimum values ​​of the transformed point cloud data along the Y-axis, respectively, where L is the segment length and S is the minimum value. N For the Nth segment of data after segmentation, x c max x c min These represent the maximum and minimum values ​​of the transformed point cloud data along the X-axis, respectively. Let n be the X-axis coordinate of Pi after conversion, n be the number of segments, and [ ] be the rounding operation. When segmenting the data, if the length of the Nth segment of point cloud data in the X-axis direction is less than 0.5 times the width of the bridge deck, then the Nth segment of point cloud data will be merged into the (N-1)th segment of point cloud data, and the Nth segment of data will be deleted.

[0082] Preferably, since the downsampled point cloud data has a small data volume, in order to improve computational efficiency, the downsampled bridge surface point cloud data can be segmented first, and the segmentation of the original bridge surface point cloud data can be based on the segmentation results of the downsampled bridge surface point cloud data.

[0083] Those skilled in the art will understand that steps S2 and S3 are preprocessing of the original point cloud data.

[0084] Step S4: Apply point cloud normal filtering to the downsampled data segment to remove data points whose deviation between the normal direction and the vertical direction is greater than the normal deviation threshold K1; then apply the region growing algorithm to cluster the discrete point cloud data to obtain multiple clustered region sets; merge the same horizontal approximate surface point clouds according to the Z-axis coordinates of the centroid points of each region set to obtain multiple horizontal approximate surface point clouds.

[0085] The specific steps of step S4 are as follows:

[0086] Step S4.1: Pre-set the normal deviation threshold K1, which is used to remove data points whose normal direction is not vertical;

[0087] Step S4.2: Calculate the unit normal and curvature of all data points, use formula (7) to calculate the degree of deviation between the normal of the data points and the vertical direction and compare it with the normal deviation threshold K1, remove data points that are greater than the normal deviation threshold K1, and set the point cloud after removal as P;

[0088]

[0089] In equation (7), D(N) pi N represents the degree of deviation between the unit normal and the vertical direction. pi The unit normal.

[0090] Step S4.3: Build a K-dimensional tree representation for the point cloud set P. Pre-set the normal deviation threshold K2, curvature threshold, region point cloud quantity threshold, and region set R. The normal deviation threshold K2 and curvature threshold are used to control the region growth process, the region point cloud quantity threshold is used to remove non-planar point clouds, and the region set R is used to accommodate the region set extracted by the algorithm.

[0091] Step S4.4: Set a subset S of points and a set R of regions. c Select the point with the smallest curvature and add it to the seed point set;

[0092] Step S4.5: Select a seed point and use the nearest neighbor search algorithm based on K-dimensional tree to find the neighbor points of the current seed point. Check whether the neighbor points have been processed in step S4.6. If there are unprocessed neighbor points, proceed to step S4.6. If all neighbor points have been processed, proceed directly to step S4.8.

[0093] Step S4.6: Take an unprocessed neighboring point, calculate the angle between its normal and the normal of the current seed point. If the angle is less than the normal deviation threshold K2, then merge the neighboring point into the current region set R. c Then, continue to check the curvature of the neighboring point. If the curvature of the neighboring point is less than the curvature threshold, then the neighboring point is added to the seed point set S.

[0094] Step S4.7: Repeat step S4.6 until all unprocessed neighboring points have been processed by step S4.6.

[0095] Those skilled in the art will understand that, under the processing of steps S4.6 and S4.7, the region set R c Points in the region set R are not necessarily in the seed point set S, but points in the seed point set S are necessarily in the region set R.c ;

[0096] Step S4.8: Delete the current seed point from the seed point set;

[0097] Step S4.9: Select points from the seed point set S as seed points, and repeat steps S4.5-S4.8 until the seed point set S is empty; at this point, the growth of a region is complete, and the region set R is... c Add it to the region set R, and delete the region set R from the point cloud set P. c The point in the middle;

[0098] Step S4.10: Repeat steps S4.4-S4.9 for the remaining data points in the point cloud P until all points have been traversed;

[0099] Step S4.11: Check the number of interior points in each region set and remove the region sets whose number of interior points is less than the threshold of the number of region point clouds; after step S4.11, the clustered region sets are obtained.

[0100] Those skilled in the art will understand that steps S4.3-S4.11 are the operation method in this embodiment of applying the region growing algorithm to cluster discrete point cloud data and obtain multiple regions after clustering.

[0101] Step S4.12: Set the distance threshold K3 and calculate the R of each region set after clustering. c The Z-axis coordinates of the centroid; and the regions are numbered sequentially starting from 1; let n = 1, m = n + 1, and MAX be the number of regions after clustering;

[0102] Step S4.13: Calculate the absolute value of the difference between the Z-axis coordinates of the centroids of regions n and m;

[0103] If the absolute value is greater than or equal to the distance threshold K3, proceed to step S4.15;

[0104] If the absolute value is less than K3, then merge the m region set into the n region set, delete the m region set, update MAX = MAX-1, reduce the number of the region set with the number greater than m by 1, and then proceed to step S4.14.

[0105] Step S4.14: If m is less than MAX+1, proceed to step S4.13; if m is not less than MAX+1, execute S4.16.

[0106] Step S4.15: If m is less than MAX, make m = m + 1 and return to step S4.13; if m is not less than MAX, proceed to step S4.16.

[0107] Step S4.16: If n is less than MAX-1, make n = n+1 and m = n+1 and return to step S4.13; if n is not less than MAX-1, then end the loop.

[0108] Those skilled in the art will understand that steps S4.12-S4.16 are the operation method in this embodiment for merging the point clouds of the same horizontal approximate surface based on the Z-axis coordinates of the centroids of each region set. The merging logic here is:

[0109] (1) First, number the clustered regions starting from 1;

[0110] (2) Select the region set with the smallest number as the comparison region set, and the rest as the merged region set. Calculate the absolute value of the difference between the centroid Z-axis coordinates of each merged region set and the comparison region set. Merge the merged region sets with the difference less than the distance threshold K3 into the comparison region set. Use the merged region set and the comparison region set as the horizontal approximate surface. If there is no merged region set that can be merged with the comparison region set, then the comparison region set is used as the horizontal approximate surface.

[0111] (3) In the region set after removing the horizontal approximation surface, continue to perform the above operation (2) in the order of numbering; if there is no comparison region set and merged region set that can be merged in the remaining region set, the remaining region set is used as the horizontal approximation surface respectively.

[0112] Step S5: Select the horizontal approximate surface point cloud with the largest number of interior points, and use least squares fitting to obtain the planar mathematical model of the horizontal approximate surface point cloud. Based on this planar mathematical model, extract the complete bridge deck surface point cloud P from the original data segment. fi ;

[0113] The specific process of step S5 is as follows:

[0114] Step S5.1: Predefine a planar mathematical model, select the horizontal approximate surface point cloud with the largest number of interior points, and use least squares fitting to obtain the planar mathematical model p of this horizontal approximate surface point cloud as follows:

[0115] p:AX+BY+CZ=D (8),

[0116] In equation (8), (A,B,C) T D is the unit normal vector of the plane mathematical model, and D / C is the Z-intercept of the plane mathematical model (C≠0).

[0117] Step S5.2: Coarsely extract the point cloud collection P co : Set a distance threshold L1, calculate the distance from the data point in the original data segment to the plane mathematical model p according to formula (9), and consider the data point with a distance less than the distance threshold L1 to belong to the upper surface of the bridge deck, so as to achieve coarse extraction;

[0118]

[0119] In equation (9), d(a,p) is the distance from data point a to the planar mathematical model p, and X a Y a Z a These are the coordinates of data point a after the transformation.

[0120] Step S5.3: Extract the complete point cloud P from the upper surface of the bridge deck. fi : Calculate the point cloud set P co The normal and curvature of the data points are set, the normal evaluation weight W is set, and the comprehensive distance evaluation index D(a,p) is calculated using equations (10)-(12). If D(a,p) is less than the distance threshold L1, the data points are considered to belong to the upper surface of the bridge deck, thus achieving accurate extraction.

[0121] W a =W(1-C) a (10),

[0122] d N (a,p)=cos -1 (|N a ·N P |) (11),

[0123] D(a,p)=|(1-W a ')d(a,p)+W a 'd N (a,p)| (12),

[0124] In equations (10)-(12), C a Let W be the curvature of data point a, and W be the normal evaluation weight. a 'To evaluate the normal of data point a considering the effect of curvature, d N (a, p) represents the deviation between the normal to data point a and the normal to the plane mathematical model p, where N is the deviation. a N P , respectively, are the unit normal vectors of data point a and plane mathematical model p, and D(a,p) is the comprehensive distance evaluation index.

[0125] Step S6: Assess the point cloud P on the upper surface of the bridge deck. fi Bridge deck boundary point cloud acquisition is performed to identify all boundary point clouds along the longitudinal direction of the bridge; based on the positional differences of all boundary point clouds along the longitudinal direction of the bridge in the width direction, the outer boundary line is identified as the bridge linear feature point.

[0126] The specific process of step S6 is as follows:

[0127] Step S6.1: Process the point cloud P on the upper surface of the bridge deck using the Angle Criterion algorithm. fi Obtain the boundary point cloud of the bridge deck (the specific method of using the Angle Criterion algorithm to obtain the boundary point cloud of the bridge deck is common knowledge in the field and will not be described in detail in this embodiment);

[0128] Step S6.2: Predefine a straight line model, pre-setting the minimum interior point threshold, distance threshold K4, and allowable deviation angle K5 from the X-axis for the longitudinal boundary point cloud; the straight line model is shown in equation (13):

[0129]

[0130] In equation (13), (A,B,C) T Let (a, b, c) be the unit direction vector of the line, and (a, b, c) be the coordinates of a point on the line model.

[0131] Step S6.3: Based on the predefined distance threshold K4 and the allowable deviation angle K5 from the X-axis, apply the random sampling consensus algorithm to identify a set of boundary line point clouds in the bridge deck boundary point cloud that have an angle less than K5 with the X-axis (i.e., the longitudinal direction of the bridge). (How to use the random sampling consensus algorithm to obtain boundary line point clouds with an angle less than K5 with the X-axis is common knowledge in the field and will not be described in detail in this embodiment.)

[0132] Step S6.4: Check the number of points inside the boundary line point cloud identified in S6.3. If the number of points inside is greater than the minimum threshold of the longitudinal boundary point cloud, then the boundary line point cloud is regarded as the boundary point cloud along the longitudinal direction of the bridge, and this part of the point cloud is removed from the bridge deck boundary point cloud before proceeding to step S6.3. If the number of points inside the identified boundary line point cloud is less than or equal to the minimum threshold of the longitudinal boundary point cloud, then all boundary point clouds along the longitudinal direction of the bridge are considered to have been identified.

[0133] It should be noted that, as those skilled in the art know, due to the inherent characteristics of the random sample consensus algorithm, each execution of the algorithm identifies the boundary line point cloud with the largest number of interior points in the current identification object;

[0134] Step S6.5: Based on the positional differences of the boundary point cloud along the longitudinal direction of the bridge in the Y-axis direction, identify the outer boundary line as the bridge's linear feature point. Specifically:

[0135] According to formula (14), calculate the Y coordinate value of the intersection point between the straight line model of each group of longitudinal boundary point clouds and the plane X=0, and take the longitudinal boundary point clouds corresponding to the maximum and minimum values ​​as the bridge linear feature points;

[0136]

[0137] In equation (14), Y X＝0 The Y-coordinate of the intersection point of the longitudinal boundary point cloud and the plane X=0.

[0138] Step S7: Integrate the bridge alignment feature point extraction results from the point cloud of each bridge deck surface to obtain the complete bridge alignment feature points, specifically:

[0139] According to Equation (15), the bridge alignment feature points of the point cloud on the upper surface of each bridge deck are restored to the original coordinate system, and then the restored bridge alignment feature points are merged to obtain the complete full bridge alignment feature points.

[0140]

[0141] In equation (15): (x c ,y c ) T The coordinates of the point cloud data after coordinate transformation along the principal axis direction are (x r ,y r ) T To reconstruct the coordinates of the point cloud data, This is the coordinate transformation matrix in step S3.

[0142] This embodiment also provides a specific application example of the bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning, as follows:

[0143] This application case requires the rapid measurement of the alignment of a continuous railway beam bridge with a total length of 834m. The relevant point cloud data processing technologies were all implemented using the C++ programming language combined with the Point Cloud Library (PCL) open-source programming library. The rapid bridge alignment measurement method based on vehicle-mounted 3D laser scanning technology is as follows:

[0144] Field data collection: A GPS base station for a vehicle-mounted 3D laser scanner was set up under the bridge, and a small railcar integrating the vehicle-mounted 3D laser scanner was mounted on the bridge tracks. The vehicle was set to a speed of 5 kilometers per hour to acquire point cloud data of the entire bridge. The equipment used in this case is the Rail Moving 3D Laser Measurement System (RMMS) from Wuhan Xirui Optoelectronic Technology Co., Ltd. The raw point cloud data was processed as follows: Figure 2 As shown.

[0145] Raw data preprocessing: The raw point cloud data was preprocessed using CloudCompare point cloud data processing software. The raw point cloud was manually segmented to obtain the full bridge surface point cloud data. A comparison before and after manual segmentation is shown below. Figure 3a and Figure 3b The VoxelGrid class of PCL is used to implement a voxelized mesh downsampling algorithm to process bridge surface point cloud data, reducing the amount of point cloud data. A comparison of data before and after downsampling is shown in [link to documentation]. Figure 4a and Figure 4b A principal component analysis (PCA) algorithm was developed to perform PCA on the X and Y coordinates of the downsampled bridge surface point cloud, obtaining the first and second principal axis directions. After transforming the X and Y coordinates of the bridge surface point cloud along the principal axis directions, the difference between the maximum and minimum values ​​of the bridge surface data along the second principal axis direction was calculated and used as the bridge surface width. The downsampled bridge surface data was then segmented along the first principal axis direction, with the segment length set to twice the bridge surface width. Simultaneously, the bridge surface point cloud before downsampling was also segmented with the same principal axis direction and segment length. See step S3 for specific operation details, and see the segmentation results below. Figure 5 For each segment of the bridge deck point cloud, perform horizontal approximate surface point cloud extraction, bridge deck point cloud extraction, and bridge linear feature point extraction operations.

[0146] Horizontal approximate surface point cloud extraction: The PCL's NormalEstimation class is used to calculate the point cloud normals and curvature. The deviation between the normal direction and the vertical direction of all point clouds is checked, and data points with deviations greater than the normal deviation threshold K1 are removed. The normal filtering operation is detailed in steps S4.1-S4.2. The PCL's RegionGrowing class is used to implement a region growing algorithm, clustering the discrete point cloud data. The PCL's compute3DCentroid function is used to calculate the Z-axis coordinates of the centroid of each clustered region. After merging, multiple segments of horizontal approximate surface point clouds are obtained. A comparison before and after horizontal approximate surface point cloud extraction is shown below. Figure 6a and Figure 6b ;

[0147] Bridge panel point cloud extraction: According to the predefined planar mathematical model in equation (8), the point cloud of the horizontal approximate surface with the largest number of internal points is fitted by least squares to obtain the point cloud model parameters; coarse extraction of bridge panel upper surface point cloud: set a distance threshold L1, calculate the distance from the data points in the original data segment without downsampling to the planar mathematical model according to equation (9), and take the data points whose distance is less than the distance threshold L1 as belonging to the upper surface of the bridge panel; fine extraction of bridge panel upper surface cloud: calculate the normal and curvature of the coarsely extracted point cloud set, set the normal evaluation weight W, calculate the comprehensive distance evaluation index D(a,p) according to equations (10)-(12), and if D(a,p) is less than the distance threshold L1, the data points are considered to belong to the upper surface of the bridge panel. The comparison of the bridge panel upper surface point cloud before and after extraction is shown in the figure. Figure 7a and Figure 7b .

[0148] Bridge linear feature point extraction: The Angle Criterion algorithm is implemented using the BoundaryEstimation class of PCL to extract the bridge deck boundary point cloud; according to the predetermined straight line model according to Equation (13), the random sampling consistency algorithm is applied to identify all boundary straight line point clouds with an angle less than K5 with the X-axis and a number of interior points greater than the minimum interior point threshold of the longitudinal boundary point cloud as the boundary point cloud along the longitudinal direction of the bridge; according to Equation (14), the Y coordinate value of the intersection point of each group of longitudinal boundary point cloud straight line model and the plane X=0 is calculated; the boundary point cloud corresponding to the maximum and minimum values ​​is taken as the bridge linear feature point; the linear feature point extraction effect of each bridge deck point cloud is shown in the figure. Figure 8 .

[0149] Data Integration: After coordinate system restoration, the linear feature point extraction results of the point cloud on the upper surface of each bridge deck segment are integrated to obtain the complete bridge alignment extraction results. The complete bridge alignment extraction results are shown below. Figure 9 .

[0150] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for measuring bridge alignment based on vehicle-mounted three-dimensional laser scanning, characterized in that, Includes the following steps: Step S1: Acquire raw point cloud data using a vehicle-mounted 3D laser scanner; Step S2: Segment the original point cloud data to obtain the original bridge deck point cloud data; downsample the original bridge deck point cloud data to obtain the downsampled bridge deck point cloud data. Step S3: Using the same data segmentation method, divide both the downsampled bridge surface point cloud data and the original bridge surface point cloud data into multiple data segments; and execute steps S4-S6 with each downsampled data segment and its corresponding original data segment as the processing object. Step S4: Apply point cloud normal filtering to the downsampled data segment to remove data points whose deviation between the normal direction and the vertical direction is greater than the normal deviation threshold K1; then apply the region growing algorithm to cluster the discrete point cloud data to obtain multiple clustered region sets; merge the same horizontal approximate surface point clouds according to the Z-axis coordinates of the centroid points of each region set to obtain multiple horizontal approximate surface point clouds. Step S5: Select the horizontal approximate surface point cloud with the largest number of interior points, and use least squares fitting to obtain the planar mathematical model of the horizontal approximate surface point cloud. Based on this planar mathematical model, the complete point cloud P on the upper surface of the bridge deck is extracted from the original data segment. fi Specifically, it involves using a planar mathematical model to first roughly extract the point cloud set P from the original data segment. co Then, in the point cloud collection P co The complete point cloud P on the upper surface of the bridge deck was extracted. fi ; Extracting complete point cloud P from the upper surface of the bridge deck fi Specifically: Calculate the point cloud set P co The normal and curvature of the data points are set, the normal evaluation weight W is set, and the comprehensive distance evaluation index D(a,p) is calculated using equations (10)-(12). If D(a,p) is less than the distance threshold L1, the data points are considered to belong to the upper surface of the bridge deck, thus achieving accurate extraction. W a '=W(1-C a ) (10), d N (a,p)=cos -1 (|N a ·N P |) (11), D(a,p)=|(1-W a ')d(a,p)+W a 'd N (a,p)| (12), In equations (10)-(12), C a Let W be the curvature of data point a. a 'To evaluate the normal of data point a considering the effect of curvature, d N (a, p) represents the deviation between the normal to data point a and the normal to the plane mathematical model p, where N is the deviation. a N P Let a, p be the unit normal vectors of data point a and plane mathematical model p, respectively, and d(a, p) be the distance from data point a to plane mathematical model p. Step S6: Assess the point cloud P on the upper surface of the bridge deck. fi Bridge deck boundary point cloud acquisition is performed to identify all boundary point clouds along the longitudinal direction of the bridge; based on the positional differences of all boundary point clouds along the longitudinal direction of the bridge in the width direction, the outer boundary line is identified as the bridge linear feature point. Step S7: Integrate the bridge alignment feature point extraction results of the point cloud on the upper surface of each bridge deck segment to obtain the complete bridge alignment feature points.

2. The bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to claim 1, characterized in that, The data segmentation in step S3 specifically involves: performing principal component analysis on the X and Y axis coordinates of the point cloud data to obtain the directions of the first and second principal axes; transforming the X and Y coordinates of the point cloud data along the principal axis directions, with the first principal axis direction as the X direction and the second principal axis direction as the Y direction; calculating the difference between the maximum and minimum values ​​of the transformed point cloud data along the second principal axis and using this difference as the bridge deck width; and segmenting the transformed point cloud data along the first principal axis direction, with the segment length set to k times the bridge deck width, where k = 2.0 to 2.

5.

3. The bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to claim 1, characterized in that, In step S4, the point clouds of the same horizontal approximate surface are merged based on the Z-axis coordinates of the centroid of the region set. Specifically: Step A1: Set the distance threshold K3, calculate the Z-axis coordinate of the centroid of each region after clustering; and number each region sequentially starting from 1; let n = 1, m = n + 1, and MAX be the number of regions after clustering. Step A2: Calculate the absolute value of the difference between the Z-axis coordinates of the centroids of regions n and m; If the absolute value is greater than or equal to the distance threshold K3, proceed to step A4; If the absolute value is less than the distance threshold K3, then merge the m region set into the n region set, delete the m region set, update MAX = MAX-1, reduce the number of the region set with the number greater than m by 1, and then proceed to step A3. Step A3: If m is less than MAX+1, proceed to step A2; if m is not less than MAX+1, proceed to step A5. Step A4: If m is less than MAX, make m = m + 1 and return to step A2; if m is not less than MAX, proceed to step A5. Step A5: If n is less than MAX-1, set n = n+1 and m = n+1 and return to step A2; if n is not less than MAX-1, then end the loop.

4. The bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to claim 2, characterized in that, According to equation (2), the X and Y coordinates of the point cloud data are transformed along the principal axis; In equation (2): (x c ,y c ) T The coordinates of the point cloud data after coordinate transformation along the principal axis direction are (x p ,y p ) T To transform the coordinates of the point cloud data, Let A be the coordinate transformation matrix, where (A) F1 A F2 ) T In the direction of the first principal axis, (A) S1 A S2 ) T This is the direction of the second principal axis.

5. The bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to claim 1, characterized in that, Coarse Extraction Point Cloud Collection P co Specifically, a distance threshold L1 is set, and the distance from the data point in the original data segment to the plane mathematical model p is calculated according to equation (9). Data points with a distance less than the distance threshold L1 are considered to belong to the upper surface of the bridge deck, so as to achieve coarse extraction. In equation (9), d(a,p) is the distance from data point a to the planar mathematical model p, and X a Y a Z a These are the coordinates of data point a after the transformation.

6. The bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to claim 1, characterized in that, In step S6: the point cloud P on the upper surface of the bridge deck is processed using an angle criterion algorithm. fi The process involves obtaining the boundary point cloud of the bridge deck; predefining a straight line model; and applying a random sampling consensus algorithm to identify all boundary line point clouds with an angle less than K5 with the X-axis and a number of interior points greater than the minimum interior point threshold of the longitudinal boundary point cloud as the boundary point cloud along the longitudinal direction of the bridge, where K5 is the allowable angle of deviation from the X-axis; and identifying the outer boundary line as the bridge linear feature point based on the positional differences of the boundary point cloud along the longitudinal direction of the bridge in the Y-axis direction.

7. The bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to claim 6, characterized in that, The linear model is shown in equation (13): In equation (13), (A,B,C) T is the unit direction vector of the line; (a,b,c) are the coordinates of a point on the line model.

8. The bridge alignment measurement method based on vehicle-mounted three-dimensional laser scanning according to claim 6, characterized in that, Calculate the y-coordinate value of the intersection point of the boundary point cloud along the longitudinal direction of the bridge and the plane X = 0 according to formula (14), and take the longitudinal boundary point cloud corresponding to the maximum and minimum values ​​as the bridge linear feature points. In equation (14), Y X＝0 The y-coordinate of the intersection point of the longitudinal boundary point cloud and the plane X = 0.

Images