A highway tunnel section cloud rapid fitting method

Abstract

The present application relates to the technical field of highway tunnel rapid automatic deformation monitoring, and particularly relates to a highway tunnel section cloud rapid fitting method.S1, end surface point cloud is extracted;S2, point cloud preprocessing is carried out;S3, sampling fitting reduction is carried out;S4, progressive iteration cycle is carried out.The beneficial effects of the present application are as follows: the number of operation iterations is greatly reduced, the fitting speed is greatly improved, the efficiency of section fitting is improved.The stability of the section fitting result is improved, the number of iterations is reduced, and the efficiency is improved.Most of the noise points in the tunnel section point cloud can be removed to reduce the influence of the noise points on the tunnel section fitting.The fitting can be avoided to fall into local optimal precision, so that the fitting of the overall point cloud of the section reaches optimal precision, and the fitting precision can be improved to 2mm compared with the traditional fitting scheme.

Description

Technical Field

[0001] This invention relates to the field of rapid and automated deformation monitoring technology for highway tunnels, and in particular to a method for rapid fitting of cloud patterns on highway tunnel cross sections. Background Technology

[0002] Various types of defects may occur during tunnel operation. Rapid fitting of tunnel cross-section point clouds is a fundamental technology supporting automatic tunnel inspection. Improving the efficiency and accuracy of cross-section fitting can greatly enhance the ability to detect defects during inspection.

[0003] The essence of highway tunnel cross-section fitting is to use a specific fitting algorithm to fit the five parameters of an elliptical model to the discrete point cloud of the extracted tunnel cross-section. Existing methods mostly use the least squares method combined with the elliptic formula for elliptic fitting. The basic idea is: after scanning the tunnel with laser point cloud data, a point cloud of a certain thickness is extracted from the tunnel cross-section. This point cloud is projected onto a plane, and least squares elliptic fitting is performed on all the point clouds on the plane to fit the five parameters of the ellipse for that cross-section. However, due to the noise inherent in the scanner's point cloud data and the presence of other attached devices on the tunnel surface, the cross-section point cloud contains a large amount of noise, causing traditional techniques to fail to fit high-precision elliptic parameters effectively.

[0004] To address this issue, this application proposes a rapid cloud fitting method for highway tunnel cross-sections to solve the aforementioned problems. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention provides a method for rapid cloud fitting of highway tunnel cross-sections.

[0006] This invention is achieved through the following technical solution:

[0007] A method for rapid cloud fitting of highway tunnel cross-sections, characterized by comprising the following steps:

[0008] S1, extract the end face point cloud; S2, perform point cloud preprocessing; S3, sample fitting and reduction; S4, perform progressive iterative loop.

[0009] Furthermore, in order to better realize the present invention, S1 specifically involves obtaining the three-dimensional central axis of the tunnel based on a projection strategy, extracting the tunnel arch portion based on the elevation difference between the tunnel arch and the central axis, roaming the central axis to obtain the point cloud of the section to be monitored, and setting the section thickness to 2cm according to the experiment; projecting the extracted three-dimensional section point cloud of a certain thickness onto the plane where the section is located, compressing the three-dimensional coordinates into the two-dimensional plane, and preserving the section features of the point cloud to the greatest extent.

[0010] Furthermore, in order to better realize the present invention, S2 specifically includes: S21, voxel downsampling: if the number of cross-sectional point clouds exceeds 10,000 points, then PCL voxelGrid is used to perform voxel downsampling on the cross-sectional point clouds to obtain resampled point clouds as input cross-sectional point clouds; S22, sorted segmented sampling: the input cross-sectional point clouds are sorted according to the increasing order of the X coordinate of the point clouds, and the sorted point clouds are divided into 10 sampling intervals in order.

[0011] Furthermore, to better implement the present invention, S3 specifically comprises: S31, random sampling, randomly selecting one point cloud from each of the 10 equally divided segments, for a total of 10 points as a random sampling point set; S32, least squares ellipse fitting, using the sampling point set as the fitting sample for the ellipse reference, and using the least squares iterative algorithm to calculate the ellipse model reference of the current cross-section—the five parameters of the ellipse—as the ellipse model reference in the current iteration, wherein the five parameters of the ellipse include the ellipse center x, y; the ellipse major and minor semi-axes a, b; and the ellipse offset theta; S33, counting "interior points", traversing the input cross-section point cloud, calculating the distance from the cross-section point cloud point to the ellipse model reference in the current iteration, and when the distance between the point cloud and the fitted ellipse is less than a threshold, the point cloud point is used as an "interior point" in the current iteration, and the number of interior points in the current iteration is counted.

[0012] Furthermore, in order to better realize the present invention, S4 specifically includes:

[0013] S41, the optimal ellipse parameters are obtained through cyclic sampling and fitting. The steps of RANSAC sampling, least squares ellipse fitting, and counting "interior points" are repeated. If the number of "interior points" in the iteration is greater than the maximum number of "interior points" in the previous iteration, the five ellipse parameters calculated in this iteration are taken as the optimal RANSAC ellipse fitting parameters. The iteration ends when the number of "interior points" is greater than 95% of the input point cloud or the number of iterations reaches a set value. After the current iteration ends, the "interior points" in the iteration with the most "interior points" are output as the denoised cross-sectional point cloud. The ellipse fitting parameters in the iteration with the most "interior points" are output as the optimal ellipse fitting parameters.

[0014] S42, improve the threshold accuracy and iterate again, performing a second iteration loop. The denoised cross-sectional point cloud is used as input, and the optimal ellipse parameters from the previous iteration are inherited into the current iteration as the five ellipse parameters for the first iteration. Simultaneously, the distance threshold for "interior points" is changed to 1.5cm, and the above iteration loop is performed again. The iteration loop ends when the number of "interior points" exceeds 90% of the input point cloud or the number of iterations reaches a set value. In the output iteration, the iteration with the most "interior points" uses the "interior points" as the cross-sectional point cloud for the next iteration loop, and the ellipse fitting parameters are output as the optimal ellipse fitting parameters. Perform three iteration loops, using the cross-sectional point cloud from the previous iteration as input, and inheriting the optimal ellipse parameters from the previous iteration into the current iteration as the five ellipse parameters for the first iteration. Simultaneously, the distance threshold for "interior points" is changed to 5mm, and the above iteration loop is performed again. The iteration loop ends when the number of "interior points" exceeds 85% of the input point cloud or the number of iterations reaches a set value. In the output iteration, the ellipse fitting parameters are output as the optimal ellipse fitting parameters for this algorithm in the iteration with the most "interior points."

[0015] The beneficial effects of this invention are:

[0016] (1) Sort the cross-sectional point cloud and then perform segmented sampling according to the segment. Compared with the random sampling of traditional RANSAC, this greatly reduces the number of computation iterations and greatly improves the fitting speed and efficiency of cross-sectional fitting.

[0017] (2) Sampling point cloud for least squares ellipse fitting improves the stability of cross-section fitting results compared to traditional 5-point ellipse reduction, while reducing the number of iterations and improving efficiency.

[0018] (3) The present invention can eliminate the influence of noise points in the tunnel cross section point cloud on the tunnel cross section fitting, compared with the traditional least squares fitting scheme.

[0019] (4) The present invention can gradually increase the accuracy of tunnel fitting. Compared with the existing technology, it can avoid fitting from getting stuck in the local accuracy optimum and make the overall point cloud of the cross section fit the optimal accuracy. Its fitting accuracy can be improved to 2mm compared with the traditional fitting scheme. Attached Figure Description

[0020] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] Figure 1 This invention provides a specific embodiment of a method for rapid cloud fitting of highway tunnel cross-sections, which includes the following steps:

[0024] 1. Extract cross-sectional point cloud

[0025] The 3D central axis of the tunnel is obtained based on a projection strategy. The tunnel arch is extracted based on the elevation difference between the tunnel arch and the central axis. A point cloud of the section to be monitored is obtained by traversing the central axis, with the section thickness set to 2cm according to experimental settings. The extracted 3D section point cloud of a certain thickness is projected onto the plane containing the section, compressing the 3D coordinates into a 2D plane while preserving the section features of the point cloud to the greatest extent possible.

[0026] 2. Point cloud preprocessing

[0027] 1) Voxel downsampling

[0028] If the number of points in the cross-section point cloud exceeds 10,000, then PCL voxelGrid is used to voxel downsample the cross-section point cloud to obtain the resampled point cloud as the input cross-section point cloud.

[0029] 2) Sorted segmented sampling

[0030] The input cross-sectional point cloud is sorted in ascending order of x-coordinate, and the sorted point cloud is then divided into 10 sampling intervals in sequence.

[0031] 3. Sampling Fitting and Reduction

[0032] 1) Random sampling

[0033] In the 10 equally divided segments, a point cloud is randomly selected from each segment, and a total of 10 points are selected as the random sampling point set.

[0034] 2) Least squares elliptic fitting

[0035] The sampled point set is used as the fitting sample of the ellipse reference. The least squares iterative algorithm is used to calculate the ellipse model reference of the current section - the five parameters of the ellipse (including the ellipse center x, y; the ellipse major and minor semi-axes a, b; the ellipse offset theta) as the ellipse model reference in the current iteration.

[0036] 3) Statistics on "interior points"

[0037] Traverse the input cross-sectional point cloud, calculate the distance between the cross-sectional point cloud points and the reference of the elliptical model in the current iteration. When the distance between the point cloud and the fitted ellipse is less than the threshold, the point cloud point is regarded as the "interior point" in the current iteration, and the number of interior points in the current iteration is counted.

[0038] 4. Progressive Iterative Loop

[0039] 1) The optimal ellipse parameters are obtained by cyclic sampling and fitting.

[0040] Repeat the steps of RANSAC sampling, least-squares ellipse fitting, and counting "interior points." If the number of "interior points" in the current iteration is greater than the maximum number of "interior points" in the previous iteration, then the five parameters of the ellipse calculated in this iteration are taken as the optimal RANSAC ellipse fitting parameters. The iteration ends when the number of "interior points" is greater than 95% of the input point cloud or the number of iterations reaches a set value. After the current iteration ends, the "interior points" from the iteration with the most "interior points" are output as the denoised cross-sectional point cloud; the ellipse fitting parameters from the iteration with the most "interior points" are output as the optimal ellipse fitting parameters.

[0041] 2) Improve threshold accuracy and then iterate.

[0042] The second iteration loop is performed. The denoised cross-sectional point cloud is used as input, and the optimal ellipse parameters from the previous iteration are inherited into the current iteration, serving as the five parameters of the ellipse for the first iteration. Simultaneously, the distance threshold for "interior points" is changed to 1.5cm, and the above iteration loop is repeated. The iteration loop ends when the number of "interior points" exceeds 90% of the input point cloud or the number of iterations reaches a set value. In the output iteration, the iteration with the highest number of "interior points" uses these "interior points" as the cross-sectional point cloud for the next iteration loop, and the ellipse fitting parameters are output as the optimal ellipse fitting parameters.

[0043] The algorithm performs three iterations. The cross-sectional point cloud from the previous iteration is used as input, and the optimal ellipse parameters from the previous iteration are inherited by the current iteration, serving as the five ellipse parameters for the first iteration. The distance threshold for "interior points" is changed to 5mm, and the iteration is repeated. The iteration ends when the number of "interior points" exceeds 85% of the input point cloud or the number of iterations reaches a set value. In the output iteration, the ellipse fitting parameters from the iteration with the highest number of "interior points" are output as the optimal ellipse fitting parameters for this algorithm.

[0044] 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 it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A method for rapid cloud fitting of highway tunnel cross-sections, characterized in that, Includes the following steps: S1, extract the end face point cloud; S1 specifically involves obtaining the three-dimensional central axis of the tunnel based on a projection strategy, extracting the tunnel arch portion based on the elevation difference between the tunnel arch and the central axis, roaming the central axis to obtain the point cloud of the section to be monitored, and setting the section thickness to 2cm according to the experiment; projecting the extracted three-dimensional section point cloud of the above thickness onto the plane where the section is located, compressing the three-dimensional coordinates into a two-dimensional plane, and preserving the section features of the point cloud to the greatest extent. S2, perform point cloud preprocessing: S3, sampling fitting regression; specifically... S31, Random sampling: In the 10 equally divided segments, a point cloud is randomly selected from each segment, for a total of 10 points as the random sampling point set; S32, Least square ellipse fitting: The sampling point set is used as the fitting sample for the ellipse reference. The least square iterative algorithm is used to calculate the ellipse model reference of the current section - the five parameters of the ellipse as the ellipse model reference in the current iteration. The five parameters of the ellipse include the ellipse center x and y. Ellipse major and minor semi-axes a and b; ellipse offset theta; S33, count "interior points", traverse the input cross-sectional point cloud, calculate the distance from the cross-sectional point cloud point to the ellipse model reference in the current iteration. When the distance between the point cloud point and the fitted ellipse is less than the threshold, the point cloud point is regarded as the "interior point" in the current iteration, and the number of interior points in the current iteration is counted. S4 involves a progressive iterative loop; specifically, S41, the optimal ellipse parameters are obtained through cyclic sampling and fitting. The steps of RANSAC sampling, least squares ellipse fitting, and counting "interior points" are repeated. If the number of "interior points" in the iteration is greater than the maximum number of "interior points" in the previous iteration, the five ellipse parameters calculated in this iteration are taken as the optimal RANSAC ellipse fitting parameters. The iteration ends when the number of "interior points" is greater than 95% of the input point cloud or the number of iterations reaches a set value. After the current iteration ends, the "interior points" in the iteration with the most "interior points" are output as the denoised cross-sectional point cloud. The ellipse fitting parameters in the iteration with the most "interior points" are output as the optimal ellipse fitting parameters. S42, improve the threshold accuracy and iterate again, perform the second iteration loop, take the denoised cross-sectional point cloud as input, and inherit the optimal ellipse parameters from the previous iteration to the current iteration as the five parameters of the ellipse in the first iteration; at the same time, change the distance judgment threshold of "interior points" to 1.5cm, and perform the above iteration loop again. When the number of "interior points" is greater than 90% of the number of input point clouds or the number of iterations reaches the set value, the iteration loop ends. In the output iteration, the iteration with the most "interior points" is used as the cross-sectional point cloud for the next iteration cycle, and the ellipse fitting parameters are used as the optimal ellipse fitting parameters for output. The algorithm performs three iterations, using the cross-sectional point cloud from the previous iteration as input and inheriting the optimal ellipse parameters from the previous iteration into the current iteration as the five ellipse parameters for the first iteration. Simultaneously, the distance threshold for "interior points" is changed to 5mm, and the iteration is repeated. The iteration ends when the number of "interior points" exceeds 85% of the input point cloud or the number of iterations reaches a set value. In the output iteration, the ellipse fitting parameters from the iteration with the highest number of "interior points" are output as the optimal ellipse fitting parameters for this algorithm.

2. The method for rapid cloud fitting of highway tunnel cross-sections according to claim 1, characterized in that: Specifically, S2 consists of: S21, voxel downsampling: if the number of points in the cross-sectional point cloud exceeds 10,000, then PCLvoxelGrid is used to perform voxel downsampling on the cross-sectional point cloud to obtain a resampled point cloud as the input cross-sectional point cloud; S22, sorted segmented sampling: the input cross-sectional point cloud is sorted according to the ascending order of the X coordinate of the point cloud, and the sorted point cloud is divided into 10 sampling intervals in sequence.

Images