Tunnel lining segmentation and cross-section deformation detection method and system

Abstract

The application discloses a tunnel lining segmentation and section deformation detection method and system, and relates to the fields of tunnel structure detection, three-dimensional laser scanning and point cloud processing. The method acquires three-dimensional point cloud data of a target section of a tunnel, determines a reference line and divides local area point clouds according to longitudinal positions and annular angles; interference point clouds are removed through local plane fitting to obtain tunnel lining point clouds; the lining point clouds are coarsely classified through clustering, a point cloud scalar field is constructed, and an inter-ring demarcation interval is determined by using a Gaussian mixture model; intra-ring segmentation is performed according to a pipe piece block design arrangement rule; section point clouds are extracted based on a single-ring central fitting plane, and convergence deformation parameters and / or staggered bench deformation parameters are calculated. A corresponding system comprises point cloud acquisition, area division, point cloud denoising, inter-ring segmentation, intra-ring segmentation, section extraction and deformation analysis modules, and can be used for shield tunnel lining segmentation, section extraction and deformation detection.

Description

Technical Field

[0001] This invention relates to the fields of tunnel structure inspection, three-dimensional laser scanning, point cloud data processing and intelligent identification of tunnel defects, and in particular to a method and system for tunnel lining segmentation and cross-sectional deformation detection. Background Technology

[0002] With the continuous expansion of urban rail transit and shield tunnel projects, a large number of existing tunnels are gradually entering the long-term operation and maintenance phase. Affected by factors such as ground pressure, groundwater erosion, train vibration, disturbance from nearby construction, and structural aging, tunnel lining structures may exhibit defects such as convergence deformation, local misalignment, and abnormal segment joints. Rapid, accurate, and continuous inspection of tunnel structures is a crucial technical foundation for ensuring safe tunnel operation.

[0003] Traditional tunnel structure inspections often rely on total stations, convergence meters, rangefinders, and manual inspections. While these methods can obtain deformation data from some sections, they typically suffer from low inspection efficiency, high reliance on manual labor, limited sampling sections, and difficulty in comprehensively reflecting the spatial state of the tunnel.

[0004] With the development of 3D laser scanning technology, mobile 3D laser scanning equipment can quickly acquire high-density point cloud data of the tunnel's inner surface, providing a new data foundation for automated tunnel structure inspection. However, the original tunnel point cloud usually contains a large number of interfering point clouds that affect lining segmentation and cross-section extraction, such as tunnel pipelines, tracks, supports, grouting holes, connecting holes, pipe joints, circumferential joints, and discrete noise points. These point clouds can affect the accuracy of lining surface fitting, segment identification, cross-section extraction, and deformation calculation.

[0005] Existing tunnel point cloud processing methods mainly have the following shortcomings:

[0006] First, some methods rely on two-dimensional images, intensity images, or image segmentation results, and then map the image recognition results onto a three-dimensional point cloud. These methods are greatly affected by image quality, lighting conditions, scanning intensity, image unfolding method, and the accuracy of image-point cloud registration. They are prone to segmentation errors when there is partial occlusion in the tunnel, uneven point cloud density, or poor image quality.

[0007] Second, some methods rely on template matching or manual selection of boundary features to achieve segmentation. These methods are highly dependent on the regularity of the tunnel structure and the setting of manual parameters, making them difficult to adapt to point clouds of long-distance tunnels, curved sections, or tunnels with significant local deformation.

[0008] Third, traditional cross-section extraction methods are usually based on the design axis or the fitted centerline, generating normal cross-sections with a fixed step size, and then extracting point clouds of a certain thickness as cross-section point clouds. This type of method is sensitive to the accuracy of the centerline extraction and the extraction step size. When the tunnel is a curved section, a locally deformed section, or there is a deviation in the centerline extraction, it is easy to cause the cross-section position to shift, which in turn affects the accuracy of convergence deformation and misalignment deformation calculation. Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for tunnel lining segmentation and cross-sectional deformation detection, which solves the problems in the prior art such as numerous interference points in tunnel point cloud, difficulty in automatic segmentation of lining structure, reliance on the accuracy of the centerline for cross-sectional extraction, and insufficient precision in segment block misalignment analysis.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for tunnel lining segmentation and cross-sectional deformation detection, comprising:

[0011] Acquire 3D point cloud data of the target tunnel segment and establish a tunnel point cloud coordinate dataset;

[0012] Based on a reference line that can characterize the longitudinal direction of the tunnel, the tunnel point cloud coordinate dataset is divided into multiple local region point clouds according to the longitudinal position and circumferential angle of the tunnel.

[0013] Local plane fitting is performed on the point cloud of each local region to obtain the local fitting plane. The distance from each point in the local point cloud to the local fitting plane is calculated, and points with a distance less than or equal to a preset distance threshold are used as the tunnel lining point cloud.

[0014] The point cloud of the tunnel lining is subjected to coarse clustering based on spatial proximity to obtain coarse classification labels for the point cloud of the tunnel lining.

[0015] Based on the coarse classification labels, a tunnel point cloud scalar field is constructed. A Gaussian mixture model is used to fit the probability distribution of the tunnel point cloud scalar field to determine the boundary interval between adjacent tunnel rings and obtain the tunnel ring segmentation result.

[0016] Based on the arrangement pattern of the tunnel segments within a single ring, the single ring point cloud in the tunnel inter-ring segmentation results is segmented into segments within the ring to obtain the segmentation results. These results are used to determine the boundary positions of adjacent segments within the same single ring, providing a basis for boundary identification of adjacent segments and calculation of misalignment deformation.

[0017] Based on the tunnel inter-ring segmentation results, single-ring point clouds are extracted, the single-ring point clouds are fitted with a center plane, and the single-ring cross-sectional point clouds are extracted based on the fitted plane.

[0018] Based on the geometric parameters of the point cloud of the single ring section, the tunnel convergence deformation parameters are calculated; based on the segmentation results of the tunnel segments within the ring, the boundary positions of adjacent tunnel segments within the same single ring are determined, and the misalignment deformation parameters are calculated based on the point cloud geometric features at the boundaries of adjacent tunnel segments.

[0019] Furthermore, the three-dimensional point cloud data is acquired by a mobile three-dimensional laser scanning device, and the three-dimensional point cloud data includes the three-dimensional coordinate information of each point on the inner surface of the tunnel;

[0020] The tunnel point cloud coordinate dataset is obtained from the 3D point cloud data after coordinate unification, format conversion, preliminary outlier removal, point cloud smoothing, and point cloud downsampling.

[0021] Furthermore, based on a reference line that can characterize the longitudinal orientation of the tunnel, the tunnel point cloud coordinate dataset is divided into multiple local region point clouds according to the tunnel's longitudinal position and circumferential angle, including:

[0022] Using the reference line as the vertical reference, the tunnel point cloud coordinate dataset is vertically partitioned according to a preset vertical distance;

[0023] The tunnel point cloud coordinate dataset is partitioned circumferentially according to a preset circumferential angle;

[0024] Point clouds that simultaneously satisfy both vertical and circumferential partitioning ranges are assigned the same region label to form corresponding local region point clouds.

[0025] The reference line is either the tunnel coarse centerline, the tunnel fitting axis, or the tunnel operation line determined based on the tunnel design axis or the longitudinal continuous distribution characteristics of the tunnel point cloud.

[0026] Further, the distance from each point in the local region point cloud to the local fitting plane is calculated, and points with distances less than or equal to a preset distance threshold are used as the tunnel lining point cloud, including:

[0027] Calculate the vertical distance from each point in the local point cloud to the local fitting plane;

[0028] Calculate the local normal vector of each point in the local point cloud and calculate the angle between the local normal vector and the normal vector of the local fitting plane;

[0029] Point clouds with a vertical distance greater than a preset distance threshold or an included angle greater than a preset angle threshold are identified as interfering point clouds and removed.

[0030] Among them, the interference point clouds that were removed include tunnel ancillary facility point clouds, hole feature point clouds, joint feature point clouds, and discrete noise point clouds;

[0031] The width of the seam is calculated based on the eliminated seam feature points.

[0032] When the seam width is greater than the minimum neighborhood parameter required for subsequent coarse clustering, stop local plane fitting and interference point cloud removal.

[0033] When the seam width is not greater than the minimum neighborhood parameter, adjust the fitting parameters, distance threshold, and normal vector threshold in the local plane fitting, and re-execute the interference point cloud removal.

[0034] Furthermore, a coarse clustering classification based on spatial proximity is performed on the tunnel lining point cloud, including:

[0035] The tunnel lining point cloud is clustered according to a preset clustering neighborhood distance parameter and a preset minimum point cloud quantity parameter; wherein, the clustering neighborhood distance parameter and the minimum point cloud quantity parameter are determined by experimental calibration based on the target tunnel point cloud data;

[0036] Point clouds that satisfy the conditions of spatial proximity and point cloud quantity are divided into the same cluster;

[0037] Assign a coarse classification label to each cluster;

[0038] Point clouds that are not classified into clusters are labeled as unclassified point clouds or point clouds to be classified.

[0039] Furthermore, a Gaussian mixture model is used to fit the probability distribution of the tunnel point cloud scalar field to determine the boundary interval between adjacent tunnel rings. This process includes:

[0040] The coarse classification labels are mapped to scalar field data distributed along the longitudinal or circumferential direction of the tunnel;

[0041] The number of Gaussian components in the Gaussian mixture model is set based on the number of tunnel rings in the target segment, the number of peak values ​​in the coarse classification label distribution, or the periodic characteristics of the scalar field.

[0042] The weights, mean, and variance of each Gaussian component are solved iteratively using the expectation-maximization algorithm.

[0043] The inter-ring boundary interval is determined based on the intersection of the probability distributions of adjacent Gaussian components;

[0044] Extract the corresponding single-ring point cloud based on the inter-ring boundary interval;

[0045] Extract the point cloud sets corresponding to two adjacent tunnel rings from the inter-ring segmentation results in sequence; select the point cloud set with more point clouds from the point cloud sets corresponding to the two adjacent tunnel rings and perform plane fitting to obtain the initial fitting plane;

[0046] Calculate the distance from the point cloud to be classified to the initial fitting plane;

[0047] When the distance is less than the preset classification threshold, the point cloud to be classified is classified into the tunnel loop corresponding to the initial fitting plane;

[0048] When the distance is greater than or equal to the preset classification threshold, plane fitting is performed on the point cloud sets corresponding to the two adjacent tunnel rings respectively, and the distance from the centroid of the point cloud to be classified to the fitting plane of the two adjacent tunnel rings is calculated respectively; the point cloud to be classified is assigned to the tunnel ring corresponding to the fitting plane with the smaller centroid distance.

[0049] Furthermore, based on the arrangement pattern of the tunnel segments within the single ring, the single-ring point cloud in the tunnel inter-ring segmentation result is segmented into tunnel segments within the ring, resulting in the following segmentation results:

[0050] Convert the single-ring point cloud into circumferential angle distribution data with the tunnel ring center as a reference;

[0051] The angle range corresponding to different types of tunnel segments is determined based on the design and arrangement angle of the tunnel segments.

[0052] Count the number of point clouds corresponding to different coarse classification labels within each angle interval;

[0053] The coarse classification labels that meet the preset conditions for the proportion of point cloud quantity in each angle interval are determined as the segment block labels for the corresponding angle interval.

[0054] The segmentation result of the inner ring segment is obtained based on the segment block label.

[0055] Further, a center plane fitting is performed on the single-ring point cloud, and the single-ring cross-sectional point cloud is extracted based on the fitted plane, including:

[0056] Calculate the centroid of a single-ring point cloud;

[0057] Subtract the centroid from the coordinates of each point in the single-ring point cloud to obtain the centered point cloud;

[0058] Construct a covariance matrix based on the centralized point cloud;

[0059] Singular value decomposition is performed on the covariance matrix to obtain the eigenvectors corresponding to the minimum singular values;

[0060] The feature vector is used as the normal vector of the single-loop fitting plane, and the single-loop fitting plane is determined based on the centroid.

[0061] Calculate the distance from each point in the single-ring point cloud to the single-ring fitting plane;

[0062] The cross-sectional point cloud located at the center of the single ring is extracted based on the distance.

[0063] Further, based on the geometric parameters of the single-ring cross-section point cloud, the tunnel convergence deformation parameters and misalignment deformation parameters are calculated, including:

[0064] The point cloud of the single-ring cross-section is fitted with a circle, an ellipse, or a B-spline curve. When a circle or ellipse is fitted, the cross-section radius, horizontal diameter, vertical diameter, major axis length, and / or minor axis length are obtained. When a B-spline curve is fitted, the cross-section profile curve is obtained, and the circumferential radius distribution is calculated based on the cross-section profile curve.

[0065] The distribution of the cross-sectional radius, horizontal diameter, vertical diameter, major axis length, minor axis length and / or circumferential radius is compared with the design value, standard test value or historical test value;

[0066] The tunnel convergence deformation was obtained based on the comparison results.

[0067] Based on the segmentation results of the inner ring tube segments, the boundary point cloud between adjacent tube segments is extracted;

[0068] Calculate the distance from the boundary point cloud to the center of the complete loop fitting or the center of the overall fitting circle;

[0069] The distances are sequentially numbered along the tunnel circumference at preset angular intervals to form a circumferential distance sequence;

[0070] The location and amount of misalignment are determined based on the abrupt change in the circumferential distance sequence at the connection point of adjacent segments.

[0071] The present invention also provides a tunnel lining segmentation and cross-sectional deformation detection system, comprising:

[0072] The point cloud acquisition module is used to acquire three-dimensional point cloud data of the target tunnel segment and establish a tunnel point cloud coordinate dataset based on the point cloud coordinate information in the three-dimensional point cloud data.

[0073] The region division module is used to determine a reference line representing the longitudinal direction of the tunnel, and based on the reference line, divide the tunnel point cloud coordinate dataset into multiple local region point clouds according to the longitudinal position and circumferential angle of the tunnel.

[0074] The point cloud denoising module is used to perform local plane fitting on the point cloud of each local region to obtain the local fitting plane, calculate the distance from each point in the local region point cloud to the local fitting plane, and take the points with a distance less than a preset threshold as the tunnel lining point cloud.

[0075] The inter-ring segmentation module is used to perform coarse clustering classification based on spatial proximity of the tunnel lining point cloud to obtain coarse classification labels for the tunnel lining point cloud. Based on the coarse classification labels, a scalar field of the tunnel point cloud is constructed. A Gaussian mixture model is used to fit the probability distribution of the scalar field of the tunnel point cloud. The inter-ring boundary interval between adjacent tunnel rings is determined according to the boundary point between adjacent Gaussian components to obtain the inter-ring segmentation result of the tunnel.

[0076] The inner ring segmentation module is used to segment the single-ring point cloud in the tunnel inter-ring segmentation result into inner ring segmentation blocks according to the design arrangement rules of the inner ring segmentation blocks, so as to obtain the inner ring segmentation result.

[0077] The cross-section extraction module is used to extract single-ring point clouds based on the tunnel inter-ring segmentation results, perform center plane fitting on the single-ring point clouds, and extract single-ring cross-section point clouds based on the fitted plane.

[0078] The deformation analysis module is used to calculate tunnel convergence deformation parameters based on the geometric parameters of the point cloud of the single ring section; determine the boundary position of adjacent segments within the same single ring based on the segmentation results of the segments within the ring; and calculate the misalignment deformation parameters based on the geometric features of the point cloud at the boundary of adjacent segments.

[0079] Compared with the prior art, the present invention has at least the following beneficial effects:

[0080] 1. This invention does not require the tunnel two-dimensional image segmentation result as a necessary input. It can directly realize the segmentation of tunnel rings and the segmentation of tunnel segments within rings based on three-dimensional point cloud coordinate information, thereby reducing the impact of image quality, lighting conditions, intensity image unfolding and image point cloud registration errors on the lining segmentation result.

[0081] 2. This invention divides the tunnel point cloud into longitudinal and circumferential local regions based on reference lines, and performs local plane fitting in each local region. It can identify interfering point clouds based on the distance from the point cloud to the fitting plane and the difference in normal vectors, which is beneficial for removing pipeline, track, support, hole, joint and discrete noise point clouds, and improving the quality of lining point cloud.

[0082] 3. This invention determines whether the removal of interfering point clouds meets the requirements of subsequent clustering segmentation by the relationship between the seam width and the minimum neighborhood parameter of clustering. This helps to form identifiable spatial intervals between adjacent pipe segments and improves the stability of coarse clustering.

[0083] 4. This invention combines coarse clustering classification with scalar field segmentation using Gaussian mixture models, and can utilize the periodic distribution of tunnel rings along the longitudinal or circumferential direction to achieve automated segmentation between tunnel rings.

[0084] 5. This invention divides the ring into segments based on the design and arrangement rules of the segments within the single ring, enabling precise identification of different segments and providing a segment-level boundary point cloud basis for subsequent misalignment deformation analysis.

[0085] 6. This invention uses the center plane fitting result of the single-ring point cloud itself to extract the cross section, which reduces the cross section position deviation caused by the center axis error, the setting of the truncation step size, or the change of the direction of the curved tunnel in the traditional center axis normal slicing.

[0086] 7. This invention can calculate convergence deformation based on the distribution of cross-sectional radius, horizontal diameter, vertical diameter, major and minor axes, or circumferential radius, and identify the location and amount of misalignment based on the abrupt change in the circumferential distance sequence of the boundary point cloud of adjacent tunnel segments, which is beneficial for realizing the quantitative analysis of tunnel structural defects. Attached Figure Description

[0087] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments are briefly described below. Obviously, the following drawings are only some embodiments of the present invention; those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0088] Figure 1 A flowchart of a method for tunnel lining segmentation and cross-sectional deformation detection based on three-dimensional point cloud coordinates provided in this embodiment of the invention;

[0089] Figure 2 This is a schematic diagram of local region segmentation and local plane fitting denoising provided in an embodiment of the present invention;

[0090] Figure 3 This is a schematic diagram of the tunnel inter-ring and intra-ring segmentation provided in an embodiment of the present invention;

[0091] Figure 4 This is a schematic diagram of single-ring centering plane fitting and cross-section extraction provided in an embodiment of the present invention;

[0092] Figure 5 This is a schematic diagram of convergence deformation and misalignment deformation calculation provided in an embodiment of the present invention;

[0093] Figure 6 This is a schematic diagram of a tunnel lining segmentation and cross-sectional deformation detection system based on three-dimensional point cloud coordinates, provided as an embodiment of the present invention. Detailed Implementation

[0094] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and embodiments. It should be understood that the described embodiments are only for illustrating the present invention and are not intended to limit the scope of protection of the present invention.

[0095] In this specification, the terms "first," "second," etc., are used only to distinguish different objects and do not indicate a specific order. The term "comprising" and its variations indicate non-exclusive inclusion. For those skilled in the art, various modifications or substitutions can be made to the embodiments of the present invention without departing from the concept of the present invention, and all such modifications or substitutions should fall within the protection scope of the present invention.

[0096] Example 1: Overall Method Flow

[0097] like Figure 1 As shown in the figure, this embodiment provides a method for tunnel lining segmentation and cross-sectional deformation detection, including the following steps:

[0098] S101. Steps for acquiring 3D point cloud data: Acquire the 3D point cloud data of the target tunnel segment and establish a tunnel point cloud coordinate dataset.

[0099] S102. Reference line determination and local region division steps: Based on the reference line that can characterize the longitudinal direction of the tunnel, the tunnel point cloud coordinate dataset is divided into multiple local region point clouds according to the longitudinal position and circumferential angle of the tunnel.

[0100] S103. Local plane fitting and interference point cloud removal steps: Perform local plane fitting on the point cloud of each local region to obtain the local fitting plane, calculate the distance from each point in the local region point cloud to the local fitting plane, and take the points with a distance less than or equal to a preset threshold as the tunnel lining point cloud.

[0101] S104. Clustering coarse classification step: Perform coarse clustering based on spatial proximity on the tunnel lining point cloud to obtain coarse classification labels for the tunnel lining point cloud.

[0102] S105.GMM Inter-ring Segmentation Steps: Construct a scalar field of tunnel point cloud based on the coarse classification labels, and use a Gaussian mixture model to fit the probability distribution of the tunnel point cloud scalar field to determine the boundary interval between adjacent tunnel rings, thereby obtaining the inter-ring segmentation results.

[0103] S106. Inner Ring Segment Segmentation Steps: Based on the arrangement pattern of the inner ring segments, the single ring point cloud in the tunnel inter-ring segmentation results is segmented into inner ring segments to obtain the inner ring segmentation results.

[0104] S107. Single-ring cross-section extraction steps: Extract single-ring point clouds based on the tunnel inter-ring segmentation results, perform center plane fitting on the single-ring point clouds, and extract single-ring cross-section point clouds based on the fitted plane.

[0105] S108. Deformation analysis steps: Calculate the tunnel convergence deformation parameters and misalignment deformation parameters based on the geometric parameters of the single-ring cross-section point cloud.

[0106] The tunnel inter-ring segmentation results and the intra-ring segmentation results are calculated from the point cloud coordinate information in the tunnel point cloud coordinate dataset. The tunnel two-dimensional image segmentation results are not used as necessary inputs. The lining segmentation and cross-sectional deformation detection are completed based on the point cloud coordinate information.

[0107] Example 2: Acquisition and Preprocessing of 3D Point Cloud Data

[0108] In this embodiment, the three-dimensional point cloud data can be acquired by a mobile three-dimensional laser scanning device. The three-dimensional point cloud data includes the three-dimensional coordinate information of each point on the inner surface of the tunnel; that is, each point can be represented as:

[0109] Pi = (xi, yi, zi);

[0110] Where Pi represents the i-th point cloud point, and xi, yi, and zi represent the coordinate values ​​of the point in the three-dimensional coordinate system.

[0111] In practical implementation, the 3D point cloud data may also include additional information such as reflection intensity, scanning time, color, and labels. However, the tunnel inter-ring segmentation and intra-ring segmentation of this invention are mainly calculated based on point cloud coordinate information, and do not use the tunnel 2D image segmentation results as a necessary input.

[0112] Preferably, to improve the stability of subsequent processing, the original 3D point cloud data can be preprocessed to obtain a tunnel point cloud coordinate dataset. The preprocessing includes:

[0113] 1. Point cloud format conversion, for example: converting the original scan format to txt, las, ply or other point cloud formats that are easy to calculate.

[0114] 2. Coordinate unification: unify point clouds from different acquisition segments or different equipment coordinate systems into the same tunnel coordinate system.

[0115] 3. Outlier removal is used to delete points that are significantly deviated from the main spatial range of the tunnel.

[0116] 4. Point cloud smoothing is used to reduce local measurement noise.

[0117] 5. Point cloud downsampling is used to reduce data size and improve computational efficiency.

[0118] 6. The bottom track area is initially removed to reduce the interference of track point cloud on the lining point cloud processing.

[0119] The above preprocessing steps can be implemented in one or more combinations depending on the quality of the engineering data, and it is not required that all of them be executed simultaneously.

[0120] Example 3: Determination of Reference Line and Tunnel Working Line

[0121] In this embodiment, the reference line is used to characterize the longitudinal orientation of the tunnel and serves as a spatial reference for dividing the local area of ​​the tunnel point cloud. The reference line can be the tunnel's coarse central axis, the tunnel's fitted axis, or a tunnel operation line determined based on the tunnel's design axis or the longitudinal continuous distribution characteristics of the tunnel point cloud.

[0122] The tunnel operation line is not required to pass strictly through the tunnel's geometric center, nor is it required to completely coincide with the design axis. Its function is to provide a spatial reference for the longitudinal sorting, local partitioning, and subsequent local fitting of the tunnel point cloud. As long as the line can maintain consistency with the longitudinal extension direction of the tunnel and enable the point cloud to be sorted and segmented according to the distance along the line, it can be used as the tunnel operation line in this invention.

[0123] In a preferred embodiment, the reference line is the coarse central axis of the tunnel extracted from the apex cloud. The extraction process is as follows:

[0124] First, an initial coordinate system is established in the tunnel point cloud coordinate dataset. The initial coordinate system allows one coordinate axis to roughly correspond to the longitudinal direction of the tunnel, and the other coordinate axis to correspond to the vertical or elevation direction.

[0125] Secondly, the tunnel point cloud is sampled along the longitudinal direction of the tunnel at a preset interval m. The preset interval m can be determined based on the tunnel point cloud density, tunnel ring width, data length, or calculation accuracy requirements.

[0126] Next, within each sampling interval, the arch apex cloud or representative arch point is extracted based on the point cloud elevation values. For example, several points with larger elevation values ​​within the sampling interval can be selected as the arch apex cloud, or the centroid of the arch apex cloud can be calculated as the representative arch point.

[0127] Then, multiple representative points at the arch are smoothed and curve-fitted to obtain a coarse central axis representing the tunnel's orientation. Curve fitting can be performed using least-squares curve fitting, polynomial fitting, spline curve fitting, or other curve fitting methods.

[0128] In another implementation, the reference line is the tunnel fitting axis. The point cloud can be divided into several segments along the longitudinal direction of the tunnel. Within each segment, the geometric center, cross-sectional center, local center, or representative point of the lining point cloud can be extracted. Curve fitting is then performed on the geometric center, cross-sectional center, local center, or representative point to obtain the tunnel fitting axis.

[0129] In another implementation, the reference line can be determined based on the tunnel design axis. For example, when the tunnel design axis data is known, the design axis can be imported and registered into the point cloud coordinate system as the tunnel operation line for the longitudinal partition of the point cloud.

[0130] The aforementioned reference line methods are all used to sort and partition the point cloud along the tunnel longitudinal direction, and they do not change the technical essence of this invention, which is to directly perform lining segmentation based on three-dimensional point cloud coordinates.

[0131] Example 4: Local Region Division Based on Reference Lines

[0132] like Figure 2 As shown, after determining the reference line, the tunnel point cloud coordinate dataset is divided into multiple local area point clouds based on the reference line.

[0133] Specifically, using the reference line as the vertical reference, the tunnel point cloud coordinate dataset is vertically partitioned according to a preset vertical distance; simultaneously, it is circumferentially partitioned according to a preset circumferential angle. Point clouds that simultaneously satisfy a certain vertical partition range and a certain circumferential partition range are assigned the same region label, forming corresponding local region point clouds.

[0134] In one implementation, the longitudinal partition width can be determined based on the tunnel ring width, point cloud density, tunnel curvature, and local fitting accuracy requirements. For curved tunnel sections, the longitudinal partition width can be appropriately reduced to lessen the impact of tunnel alignment changes on the local fitting results.

[0135] In one implementation, the circumferential partitioning angle can be determined based on point cloud density, segment size, fitting accuracy requirements, and computational efficiency. For example, the entire ring of point cloud can be circumferentially divided according to a preset angle interval.

[0136] Since the overall lining surface of a tunnel is curved, directly performing uniform planar fitting on long-distance or large-scale point clouds can easily lead to significant fitting errors. This invention divides the point cloud into multiple local regions by using longitudinal distance and circumferential angle, allowing each local region's point cloud to approximate a plane within a smaller spatial range. This improves the stability of subsequent local planar fitting and interference point cloud removal.

[0137] Example 5: Local Plane Fitting and Removal of Interference Point Clouds

[0138] like Figure 2 As shown, in this embodiment, local plane fitting is performed on the point cloud of each local region to obtain a local fitting plane, and interfering point clouds that affect the lining segmentation and cross-section extraction are removed according to the distance of the point cloud to the local fitting plane.

[0139] In one embodiment, the plane equation of the local fitting plane can be expressed as ax + by + cz + d = 0, where a, b, and c are the normal vector parameters of the local fitting plane, and d is the plane offset parameter. For any point in the local region point cloud, its perpendicular distance D to the local fitting plane can be expressed as:

[0140] ;

[0141] When the vertical distance D is greater than the preset distance threshold λ1, it can be determined that the point has a large geometric deviation relative to the local lining surface and is regarded as an interference point cloud; when the vertical distance D is less than or equal to the preset distance threshold λ1, it is regarded as a lining point cloud.

[0142] In a preferred embodiment, local plane fitting can be achieved using a sampling consistency fitting method.

[0143] Specifically, a set of points is selected from the local point cloud for candidate plane fitting, the distance from each point to the candidate fitting plane is calculated, and the candidate plane that meets the conditions is selected as the local fitting plane based on the number of interior points or the fitting error.

[0144] In other embodiments, local normal vector constraints, neighborhood point count thresholds, or known spatial ranges of ancillary facilities can be further combined to assist in the determination of interfering point clouds. For example, when the angle between the local normal vector of a point cloud and the normal vector of the local fitted plane exceeds a preset angle threshold λ2, the point can be regarded as an interfering point cloud; for discrete noise point clouds, neighborhood point count thresholds can also be used for auxiliary identification.

[0145] The interference point cloud includes point clouds of tunnel ancillary facilities, point clouds of hole features, point clouds of joint features, and discrete noise point clouds. Among them, the point cloud of tunnel ancillary facilities includes point clouds of pipelines, tracks, supports, etc.; the point cloud of hole features includes point clouds of grouting holes, connecting holes, etc.; and the point cloud of joint features includes point clouds of pipe joints, circumferential joints, etc.

[0146] The joint feature point cloud is not the point cloud of tunnel ancillary facilities, but rather the feature point cloud of boundary abrupt changes between adjacent tunnel segments or adjacent tunnel rings in the lining structure. By removing the joint feature point cloud, identifiable spatial intervals can be formed between adjacent tunnel segments or adjacent tunnel rings, thereby improving the accuracy of subsequent clustering coarse classification and segment segmentation.

[0147] The distance threshold λ1 and angle threshold λ2 can be determined based on the point cloud measurement accuracy, local point cloud density, lining surface roughness, and joint width. In actual implementation, the distance threshold λ1, angle threshold λ2, or local plane fitting parameters can also be adjusted based on whether the joint width H meets the requirements of subsequent coarse clustering.

[0148] Example 6: Relationship between seam width and minimum neighborhood parameter in clustering

[0149] In this embodiment, after removing the joint feature point cloud, the joint width H can be calculated based on the spatial interval formed between adjacent lining point clouds or adjacent pipe segment point clouds. The joint width H is used to determine whether the result of removing interfering point clouds meets the separation requirements of subsequent coarse clustering classification.

[0150] In one embodiment, the joint width H can be calculated along the longitudinal direction of the tunnel to characterize the circumferential joint width between adjacent tunnel rings; it can also be calculated along a local circumferential direction or a local normal direction to characterize the joint width between adjacent tunnel segments.

[0151] In a preferred embodiment, for two adjacent sets of lining point clouds A and B, the joint width H can be defined as the minimum Euclidean distance between the boundary points of the two sets:

[0152] H=min||pa-pb||, pa∈A, pb∈B;

[0153] Where pa represents the boundary point in point cloud set A, and pb represents the boundary point in point cloud set B.

[0154] In other embodiments, the seam width H can also be the average distance between the two set boundary points, or the projected distance between the two set boundary points along the local normal direction.

[0155] In the subsequent coarse clustering, a minimum neighborhood parameter ε can be set. When the seam width H is greater than the minimum neighborhood parameter ε, adjacent pipe segments are less likely to be mistakenly clustered into the same cluster, and the local plane fitting and interference point cloud removal iterations can be stopped. When the seam width H is not greater than the minimum neighborhood parameter ε, the local plane fitting parameters, distance threshold λ1, and angle threshold λ2 can be adjusted, and the interference point cloud removal step can be re-executed until the seam width meets the requirements for subsequent clustering and segmentation.

[0156] Example 7: Coarse Cluster Classification

[0157] In this embodiment, after obtaining the tunnel lining point cloud, the tunnel lining point cloud is subjected to coarse clustering based on spatial proximity to obtain coarse classification labels.

[0158] Specifically, the tunnel lining point cloud is clustered based on the joint distance parameter between tunnel segments and the minimum number of point clouds required to form the segments. Point clouds that meet the spatial proximity and point cloud quantity conditions are grouped into the same cluster, and each cluster is assigned a coarse classification label; point clouds not included in a cluster are marked as unclassified point clouds or point clouds to be classified.

[0159] In one implementation, the minimum neighborhood parameter ε for clustering can be determined based on the joint width H, the average spacing of the point cloud, and the designed width of the segment joint. Preferably, ε is less than or equal to the joint width H to avoid adjacent segment blocks being mistakenly merged during the clustering process. The minimum number of point clouds required to form a segment block can be estimated based on the point cloud density, the area of ​​the local region, and the minimum area of ​​the segment block.

[0160] The coarse classification label can be an integer label, for example, 0 represents an unclassified point cloud, and other integers represent different clusters. The coarse classification label is used to construct the subsequent tunnel point cloud scalar field.

[0161] Example 8: Construction of Scalar Field of Tunnel Point Cloud and Inter-ring Segmentation of GMM

[0162] like Figure 3 As shown, in this embodiment, a scalar field of tunnel point cloud is constructed based on coarse classification labels, and a Gaussian mixture model is used to fit the probability distribution of the tunnel point cloud scalar field in order to determine the inter-ring boundary interval between adjacent tunnel rings.

[0163] The tunnel point cloud scalar field refers to associating the coarse classification labels of the point cloud with the spatial location information of the point cloud to form a scalar sequence or scalar set that can characterize the distribution characteristics of the tunnel lining point cloud along the longitudinal direction of the tunnel.

[0164] In one implementation, the coarse classification labels obtained from clustering are mapped to scalar field data distributed along the longitudinal direction of the tunnel. The coarse classification labels can be integer labels, wherein point clouds not assigned to valid clusters correspond to preset unclassified labels, and different clusters correspond to different coarse classification labels.

[0165] Since shield tunnels are formed by continuously assembling multiple segment rings along the tunnel's longitudinal direction, the scalar field of the tunnel point cloud exhibits a repetitive or periodic distribution characteristic along the tunnel's longitudinal direction. By fitting the probability distribution of the tunnel point cloud scalar field using a Gaussian mixture model, the distribution intervals corresponding to different tunnel rings can be identified.

[0166] A Gaussian mixture model is a probabilistic model formed by a weighted combination of multiple Gaussian components, and its probability density function can be expressed as:

[0167] ;

[0168] Where P(x) represents the probability density of the input variable x under the Gaussian mixture model; x represents the scalar field variable used for fitting the Gaussian mixture model. In this embodiment, x is the cumulative distance of the point cloud points along the reference line; K represents the number of Gaussian components; k represents the index of the Gaussian component. This represents the weight of the k-th Gaussian component; This represents the mean of the k-th Gaussian component; Let represent the variance of the k-th Gaussian component; The mean is variance is The Gaussian probability density function.

[0169] The number of Gaussian components, K, can be set based on the number of tunnel loops in the target segment, the number of peak values ​​in the coarse classification label distribution, or the periodic characteristics of the scalar field. For example, when it is known that the target segment contains M tunnel loops, the number of Gaussian components can be set to M or a value related to M; when the number of tunnel loops in the target segment is unknown, the number of Gaussian components can be estimated based on the number of peak values ​​in the coarse classification label distribution or the periodic characteristics of the scalar field.

[0170] The parameters of the Gaussian mixture model can be solved iteratively using the expectation-maximization algorithm. The parameters include the weights, mean, and variance of each Gaussian component.

[0171] After obtaining each Gaussian component, the Gaussian components are sorted according to their mean values, and the boundary points between adjacent Gaussian components are determined based on the intersection of their probability distributions.

[0172] The tunnel point cloud scalar field is divided into multiple continuous distribution intervals based on the defined boundary points, and the point cloud set corresponding to each distribution interval is extracted. Different tunnel ring labels are assigned to the point cloud sets corresponding to different distribution intervals; these labels indicate the tunnel ring to which the point cloud belongs, thus obtaining the tunnel ring segmentation result.

[0173] Example 9: Inter-ring fine classification

[0174] After obtaining the inter-ring boundary intervals based on the Gaussian mixture model, further fine classification of the inter-rings can be performed to handle the first and last rings, locally missing rings, or point clouds to be classified.

[0175] Specifically, the inter-ring fine classification step includes:

[0176] Extract the point cloud labels corresponding to two adjacent tunnel loops in sequence;

[0177] Select the label interval with a large number of point clouds for plane fitting;

[0178] Calculate the distance from the point cloud to be classified to the fitting result of the plane, and determine whether the distance is less than the preset classification threshold;

[0179] For point clouds that cannot be directly classified, calculate the distance from their centroid to the fitting plane of the two adjacent tunnel rings.

[0180] The point cloud to be classified is assigned to the tunnel ring label with smaller distance.

[0181] The point cloud to be classified can be a point cloud that was not assigned to a valid cluster in the coarse clustering, or a point cloud whose affiliation is unclear near the boundary interval of the Gaussian mixture model. The centroid of the point cloud to be classified can be calculated by averaging the coordinates of the point clouds it contains.

[0182] In one embodiment, letting the point cloud set to be classified be Q, whose centroid is Cq, and the fitting planes of two adjacent tunnel rings are Plane1 and Plane2 respectively, then the distances d1 and d2 from Cq to Plane1 and Plane2 are calculated respectively. When d1<d2, the point cloud set Q to be classified is assigned to the tunnel ring label corresponding to Plane1; when d2<d1, the point cloud set Q to be classified is assigned to the tunnel ring label corresponding to Plane2.

[0183] Through the above-mentioned fine inter-ring classification, the continuity and integrity of the tunnel inter-ring segmentation result can be improved, and the present invention is particularly applicable to situations where the single-ring point cloud is not completely collected, the point clouds of the first and last rings are incomplete, or the local boundary is blurred.

[0184] Example 10: Intra-ring segment block segmentation

[0185] as shown in Figure 3 , after obtaining the inter-ring segmentation result of the tunnel, intra-ring segment block segmentation is performed on the single-ring point cloud according to the design arrangement rule of the segment blocks in a single ring. It specifically includes:

[0186] converting the single-ring point cloud into circumferential angle distribution data with the tunnel ring center as a reference;

[0187] determining the angle intervals corresponding to different types of segment blocks according to the design arrangement angles of the shield tunnel segments;

[0188] counting the number of point clouds corresponding to different coarse classification labels in each angle interval;

[0189] determining a coarse classification label whose point cloud quantity proportion in each angle interval meets a preset condition as the segment block label of the corresponding angle interval;

[0190] obtaining an intra-ring segment block segmentation result according to the segment block labels.

[0191] In one embodiment, the tunnel ring center can be determined by the circle fitting center, ellipse fitting center or point cloud centroid of the single-ring point cloud. Taking the tunnel ring center as the origin, the single-ring point cloud is converted into polar coordinates or circumferential angle coordinates, and the circumferential angle αi of each point cloud point is calculated. When calculating the circumferential angle, the vertically upward direction, horizontal radial direction or the designed direction of the center of the closing block can be taken as the zero direction of the angle, and the angle is normalized to the range of 0° to 360°.

[0192] According to the starting and ending angles of each segment block in the design drawing, an angle interval set {A1,A2,…,Am} is established. For each angle interval Aj, the number of points corresponding to each coarse classification label is counted therein, and the label with the largest proportion of the number of points is determined as the segment block label corresponding to the angle interval.

[0193] For example, if a certain angle interval Aj includes multiple coarse classification labels L1, L2, and L3, and the number of point clouds corresponding to label L2 accounts for the largest proportion of the total number of point clouds in that angle interval, and this proportion is greater than a preset proportion threshold, then label L2 is determined as the pipe block label corresponding to that angle interval.

[0194] In this embodiment, different types of tunnel segments can be determined according to the actual engineering design. This invention does not limit the specific names of tunnel segments; as long as different tunnel segments have an angle range within a single ring that can be represented by the designed arrangement angle, this embodiment can be used for intra-ring segmentation.

[0195] The segment arrangement angle is used only as a structural prior for segmenting within the ring, determining the theoretical angle range of the segment blocks within a single ring. This segment arrangement angle is not a result of two-dimensional image segmentation, nor does it need to be obtained through two-dimensional tunnel image segmentation.

[0196] Using the above method, the single-ring point cloud can be further divided into multiple inner-ring segment blocks, providing segment block-level boundary point clouds for subsequent misalignment deformation calculations.

[0197] Example 11: Single-loop centering plane fitting and cross-section extraction

[0198] like Figure 4 As shown, in this embodiment, a single-ring point cloud is extracted based on the tunnel ring segmentation results, and the spatial distribution of the single-ring point cloud itself is used to determine the single-ring center fitting plane. Then, the single-ring cross-sectional point cloud is extracted based on the fitting plane.

[0199] Specifically, the first step is to extract the single-ring point cloud set corresponding to a certain tunnel ring. The single-ring point cloud set includes the three-dimensional coordinate information of multiple point cloud points within the tunnel ring.

[0200] Then, the centroid of the single-ring point cloud set is calculated, and the single-ring point cloud is centered based on the centroid. A covariance matrix is ​​constructed based on the centered point cloud coordinates, and singular value decomposition is performed on the covariance matrix to obtain the eigenvectors corresponding to the minimum singular values. These eigenvectors are used to characterize the normal vector of the best-fit plane of the single-ring point cloud.

[0201] Furthermore, a single-ring centering fitting plane is determined based on the normal vector and the centroid of the single-ring point cloud set. This single-ring centering fitting plane is used to characterize the centering face position of the tunnel ring within its longitudinal width.

[0202] Subsequently, the distance from each point in the single-ring point cloud to the single-ring centering fitting plane is calculated. A preset cross-sectional thickness threshold δ is set. When the distance from a point in the point cloud to the single-ring centering fitting plane is less than or equal to δ, that point in the point cloud is considered a single-ring cross-sectional point cloud. The cross-sectional thickness threshold δ can be set according to the point cloud density, single-ring width, scanning error, and cross-sectional extraction accuracy requirements; preferably, δ is less than half the single-ring width, so that the extracted cross-sectional point cloud is located in the center region of the single ring.

[0203] Therefore, this embodiment does not require generating a cutting plane along the tunnel's central axis with a fixed step size. Instead, it determines the centering fitting plane based on the single-ring point cloud itself and extracts the cross-sectional point cloud. Thus, compared with the traditional central axis normal section extraction method, this embodiment can reduce the influence of central axis extraction error, cutting step size setting error, and changes in the curved tunnel's orientation on the cross-sectional position.

[0204] Example 12: Convergence Deformation Calculation

[0205] like Figure 5 As shown, in this embodiment, the tunnel convergence deformation parameters are calculated based on the geometric parameters of the point cloud of the single-ring cross-section.

[0206] Specifically, the point cloud of a single-ring cross-section can be fitted with circles, ellipses, or contour curves to obtain geometric parameters such as the fitting center, radius, horizontal diameter, vertical diameter, major and minor axes, or circumferential radius distribution of the measured cross-section. The contour curve fitting can be achieved using B-spline curve fitting or other curve fitting methods.

[0207] In one implementation, using the fitting center of the measured cross-section as a reference center, the distance from the measured cross-section profile to the reference center is calculated according to a preset circumferential angle interval, resulting in the measured circumferential radius distribution. By comparing the measured circumferential radius distribution with the radii at corresponding angles of the design cross-section, historical benchmark cross-section, or benchmark detection cross-section, the circumferential convergence deformation can be obtained.

[0208] Let R(θ) be the radius of the measured cross-section at the circumferential angle θ, and R0(θ) be the radius of the reference cross-section at the corresponding circumferential angle θ. Then the circumferential convergence deformation can be expressed as:

[0209] ΔR(θ) = R(θ) - R0(θ);

[0210] When ΔR(θ) is negative, it indicates that the circumferential position has contracted inward relative to the reference section; when ΔR(θ) is positive, it indicates that the circumferential position has expanded outward relative to the reference section. In practical applications, |ΔR(θ)| can also be taken as the absolute value of the deformation at the circumferential position.

[0211] In another embodiment, the measured horizontal diameter, vertical diameter, major axis length, or minor axis length of the cross section can be compared with the corresponding design value, historical benchmark value, or benchmark detection value to obtain the horizontal convergence amount, vertical deformation amount, or elliptic deformation parameters.

[0212] The above method can be used to quantitatively calculate the convergence deformation of a single-ring section of a tunnel.

[0213] Example 13: Calculation of Deformation Due to Misalignment

[0214] like Figure 5 As shown, in this embodiment, the misalignment deformation parameters are calculated based on the segmentation results of the tunnel segments within the ring. The misalignment deformation refers to the local radial displacement that occurs between adjacent tunnel segments within the same tunnel ring.

[0215] Specifically, based on the segmentation results of the pipe segments within the ring, the boundary point clouds within a preset neighborhood range on both sides of the boundary of adjacent pipe segments within the same single ring are extracted; the whole ring fitting center or the cross-section fitting center of the single ring cross-section point cloud is used as the reference center, and the distance from the point clouds on both sides of the boundary to the reference center is calculated respectively; the misalignment position and misalignment amount are determined according to the difference in distance on both sides of the boundary.

[0216] In the calculation of misalignment deformation, the point cloud on both sides of the boundary of adjacent pipe segments refers to the retained lining point cloud located within the preset neighborhood range on both sides of the boundary of adjacent pipe segments, based on the segmentation result of the pipe segments within the ring, rather than the joint feature point cloud that was previously removed as an interference point cloud.

[0217] In one implementation, several boundary points can be selected on both sides of the boundary of adjacent segments, and the average distance from the boundary points on both sides to the reference center can be calculated. Let the average radii on both sides of the segment connection position be rL and rR, respectively, then the misalignment can be expressed as:

[0218] Δs = |rL - rR|;

[0219] When Δs is greater than the preset misalignment threshold, it can be determined that there is misalignment deformation at the connection position of the segment. The above-mentioned method for calculating misalignment based on the difference in average radii on both sides of the boundary can reduce the influence of local noise on the calculation results of misalignment.

[0220] In other embodiments, a distance sequence from the boundary point cloud to the reference center can be constructed along the tunnel circumference at preset angular intervals, and the misalignment position can be identified based on abrupt changes in the distance sequence near the boundaries of adjacent tunnel segments.

[0221] Example 14: Engineering Application Case

[0222] This embodiment provides an engineering application example to illustrate the specific implementation process of the method of the present invention. This example is not intended to limit the scope of protection of the present invention.

[0223] In a target section of a subway shield tunnel, a mobile 3D laser scanning device was used to acquire 3D point cloud data of the tunnel's inner surface. After unifying the coordinates, converting the format, and initially removing outliers from the original point cloud data, a tunnel point cloud coordinate dataset was obtained.

[0224] First, the coarse central axis of the tunnel is extracted based on the point cloud at the arch apex, and this coarse central axis is used as a reference line. The point cloud is then divided into zones along the reference line at preset longitudinal distances, and simultaneously divided into circumferential zones according to preset circumferential angles, resulting in multiple local area point clouds.

[0225] Secondly, local plane fitting is performed on the point clouds of each local region. The distance from the point cloud to the local fitting plane is calculated, and interfering point clouds are removed by combining distance thresholds and normal vector thresholds, resulting in the lining point cloud mainly containing the tunnel lining surface. After removing the joint feature point cloud, the joint width H between adjacent lining point clouds is calculated, and it is determined whether it is greater than the minimum neighborhood parameter ε required for coarse clustering. If it does not meet the requirement, the threshold is adjusted, and the interfering point cloud removal is performed again.

[0226] Next, the tunnel lining point cloud is clustered and coarsely classified to obtain coarse classification labels. The cumulative distance si along the reference line and its corresponding coarse classification label Li are used to construct a scalar field of the tunnel point cloud. A Gaussian mixture model is used to fit the probability distribution of the cumulative distance si to determine the inter-ring boundary interval between adjacent tunnel rings, thus obtaining the inter-ring segmentation results of the tunnel.

[0227] Subsequently, based on the circumferential angle distribution of the single-ring point cloud and the arrangement angle of the pipe segments, the number of point clouds with different coarse classification labels in each angle interval was counted to obtain the segmentation results of the pipe segments within the ring.

[0228] Finally, a centered plane is fitted to the single-ring point cloud, and the single-ring cross-sectional point cloud is extracted based on the fitted plane. Circular, elliptical, or B-spline curve fitting is performed on the single-ring cross-sectional point cloud to calculate the horizontal diameter, vertical diameter, major axis, minor axis, or circumferential radius distribution. Based on the segmentation results of the inner ring segments, the retained lining point clouds on both sides of the boundary of adjacent segments are extracted, a circumferential distance sequence is constructed, and the misalignment position and amount are determined based on the abrupt change values ​​of the distance sequence.

[0229] Through the above process, the lining segmentation, cross-section extraction, convergence deformation calculation, and misalignment deformation analysis of the target section of the shield tunnel can be realized.

[0230] Example 15: System Implementation

[0231] like Figure 6 As shown in the figure, this embodiment provides a tunnel lining segmentation and cross-sectional deformation detection system, including a point cloud acquisition module, a region division module, a point cloud denoising module, an inter-ring segmentation module, an intra-ring segmentation module, a cross-sectional extraction module, and a deformation analysis module.

[0232] The point cloud acquisition module is used to acquire three-dimensional point cloud data of the target tunnel segment and establish a tunnel point cloud coordinate dataset based on the point cloud coordinate information in the three-dimensional point cloud data.

[0233] The region division module is used to determine a reference line representing the longitudinal direction of the tunnel, and based on the reference line, divides the tunnel point cloud coordinate dataset into multiple local region point clouds according to the longitudinal position and circumferential angle of the tunnel.

[0234] The point cloud denoising module is used to perform local plane fitting on the point cloud of each local region to obtain the local fitting plane, calculate the distance from each point in the local point cloud to the local fitting plane, and remove the interfering point cloud that affects the lining segmentation and cross-section extraction based on the distance to obtain the tunnel lining point cloud.

[0235] The inter-ring segmentation module is used to perform coarse clustering classification of the tunnel lining point cloud based on spatial proximity to obtain coarse classification labels. Based on the coarse classification labels, a scalar field of the tunnel point cloud is constructed. A Gaussian mixture model is used to fit the probability distribution of the tunnel point cloud scalar field. The inter-ring boundary interval between adjacent tunnel rings is determined according to the boundary point between adjacent Gaussian components, and the inter-ring segmentation result of the tunnel is obtained.

[0236] The inner ring segmentation module is used to segment the single-ring point cloud in the tunnel inter-ring segmentation result into inner ring segments based on the design arrangement rules of the inner ring segments, and obtain the inner ring segmentation result.

[0237] The cross-section extraction module is used to extract single-ring point clouds based on the tunnel inter-ring segmentation results, perform center plane fitting on the single-ring point clouds, and extract single-ring cross-section point clouds based on the fitted plane.

[0238] The deformation analysis module is used to calculate tunnel convergence deformation parameters based on the geometric parameters of the point cloud of the single ring section; determine the boundary position of adjacent segments within the same single ring based on the segmentation results of the segments within the ring; and calculate the misalignment deformation parameters based on the geometric features of the point cloud at the boundary of adjacent segments.

[0239] The data flow between the above modules is as follows:

[0240] The point cloud acquisition module outputs the tunnel point cloud coordinate dataset to the region partitioning module;

[0241] The region segmentation module outputs a local region point cloud, which is then transferred to the point cloud denoising module.

[0242] The point cloud denoising module outputs the tunnel lining point cloud to the inter-ring segmentation module;

[0243] The inter-ring segmentation module outputs the tunnel inter-ring segmentation results and the single-ring point cloud to the intra-ring segmentation module and the cross-section extraction module;

[0244] The inner ring segmentation module outputs the inner ring segmentation results to the deformation analysis module;

[0245] The cross-section extraction module outputs a single-ring cross-section point cloud to the deformation analysis module;

[0246] The deformation analysis module outputs tunnel convergence deformation parameters and / or misalignment deformation parameters.

[0247] The modules mentioned above can be software functional modules, hardware modules, or a combination of software and hardware modules. These modules can be integrated into the same computing device or distributed across multiple computing devices to operate collaboratively.

[0248] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, various modifications, combinations, or substitutions can be made to the above embodiments without departing from the concept of the present invention. Any modifications, equivalent substitutions, or improvements 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 segmenting and detecting cross-sectional deformation of tunnel lining, characterized in that, Includes the following steps: Acquire 3D point cloud data of the target tunnel segment and establish a tunnel point cloud coordinate dataset; Based on a reference line that can characterize the longitudinal direction of the tunnel, the tunnel point cloud coordinate dataset is divided into multiple local region point clouds according to the longitudinal position and circumferential angle of the tunnel. Local plane fitting is performed on the point cloud of each local region to obtain the local fitting plane. The distance from each point in the local point cloud to the local fitting plane is calculated, and points with a distance less than or equal to a preset distance threshold are used as the tunnel lining point cloud. The point cloud of the tunnel lining is subjected to coarse clustering based on spatial proximity to obtain coarse classification labels for the point cloud of the tunnel lining. Based on the coarse classification labels, a tunnel point cloud scalar field is constructed. A Gaussian mixture model is used to fit the probability distribution of the tunnel point cloud scalar field to determine the boundary interval between adjacent tunnel rings and obtain the tunnel ring segmentation result. Based on the arrangement pattern of the inner ring segments, the single ring point cloud in the tunnel inter-ring segmentation result is segmented into inner ring segments to obtain the inner ring segmentation result. Based on the tunnel inter-ring segmentation results, single-ring point clouds are extracted, the single-ring point clouds are fitted with a center plane, and the single-ring cross-sectional point clouds are extracted based on the fitted plane. Based on the geometric parameters of the point cloud of the single ring section, the tunnel convergence deformation parameters are calculated; based on the segmentation results of the tunnel segments within the ring, the boundary positions of adjacent tunnel segments within the same single ring are determined, and the misalignment deformation parameters are calculated based on the point cloud geometric features at the boundaries of adjacent tunnel segments.

2. The method according to claim 1, characterized in that, The three-dimensional point cloud data is acquired by a mobile three-dimensional laser scanning device, and the three-dimensional point cloud data includes the three-dimensional coordinate information of each point on the inner surface of the tunnel; The tunnel point cloud coordinate dataset is obtained from the 3D point cloud data after coordinate unification, format conversion, preliminary outlier removal, point cloud smoothing, and point cloud downsampling.

3. The method according to claim 1, characterized in that, Based on a reference line that can characterize the longitudinal orientation of the tunnel, the tunnel point cloud coordinate dataset is divided into multiple local region point clouds according to the tunnel's longitudinal position and circumferential angle, including: Using the reference line as the vertical reference, the tunnel point cloud coordinate dataset is vertically partitioned according to a preset vertical distance; The tunnel point cloud coordinate dataset is partitioned circumferentially according to a preset circumferential angle; Point clouds that simultaneously satisfy both vertical and circumferential partitioning ranges are assigned the same region label to form corresponding local region point clouds. The reference line is either the tunnel coarse centerline, the tunnel fitting axis, or a tunnel operation line determined based on the tunnel design axis or the longitudinal continuous distribution characteristics of the tunnel point cloud.

4. The method according to claim 1, characterized in that, Calculate the distance from each point in the local point cloud to the local fitting plane, and use points with distances less than or equal to a preset distance threshold as the tunnel lining point cloud, including: Calculate the vertical distance from each point in the local point cloud to the local fitting plane; Calculate the local normal vector of each point in the local point cloud and calculate the angle between the local normal vector and the normal vector of the local fitting plane; Point clouds with a vertical distance greater than a preset distance threshold or an included angle greater than a preset angle threshold are identified as interfering point clouds and removed. Among them, the interference point clouds that were removed include the point clouds of tunnel ancillary facilities, the point clouds of hole features, the point clouds of joint features, and the discrete noise point clouds; The width of the seam is calculated based on the eliminated seam feature points. When the seam width is greater than the minimum neighborhood parameter required for subsequent coarse clustering, stop local plane fitting and interference point cloud removal. When the seam width is not greater than the minimum neighborhood parameter, adjust the fitting parameters, distance threshold, and normal vector threshold in the local plane fitting, and re-execute the interference point cloud removal.

5. The method according to claim 4, characterized in that, The point cloud of the tunnel lining is subjected to coarse clustering based on spatial proximity, including: The tunnel lining point cloud is clustered according to the preset clustering neighborhood distance parameter and the preset minimum point cloud quantity parameter; Point clouds that meet the conditions of spatial proximity and point cloud quantity are divided into the same cluster; Assign a coarse classification label to each cluster; Point clouds that are not classified into clusters are labeled as unclassified point clouds or point clouds to be classified.

6. The method according to claim 5, characterized in that, The probability distribution of the tunnel point cloud scalar field is fitted using a Gaussian mixture model to determine the boundary interval between adjacent tunnel rings. This includes: The coarse classification labels are mapped to scalar field data distributed along the longitudinal or circumferential direction of the tunnel; The number of Gaussian components in the Gaussian mixture model is set based on the number of tunnel loops in the target segment, the number of peak values ​​in the coarse classification label distribution, or the periodic characteristics of the scalar field. The weights, mean, and variance of each Gaussian component are solved iteratively using the expectation-maximization algorithm. The inter-ring boundary interval is determined based on the intersection of the probability distributions of adjacent Gaussian components; Extract the corresponding single-ring point cloud based on the inter-ring boundary interval; Extract the point cloud sets corresponding to two adjacent tunnel rings from the inter-ring segmentation results in sequence; select the point cloud set with more point clouds from the point cloud sets corresponding to the two adjacent tunnel rings and perform plane fitting to obtain the initial fitting plane; Calculate the distance from the point cloud to be classified to the initial fitting plane; When the distance is less than the preset classification threshold, the point cloud to be classified is classified into the tunnel loop corresponding to the initial fitting plane; When the distance is greater than or equal to the preset classification threshold, plane fitting is performed on the point cloud sets corresponding to the two adjacent tunnel rings respectively, and the distance from the centroid of the point cloud to be classified to the fitting plane of the two adjacent tunnel rings is calculated respectively; the point cloud to be classified is assigned to the tunnel ring corresponding to the fitting plane with the smaller centroid distance.

7. The method according to claim 1, characterized in that, Based on the arrangement pattern of the tunnel segments within a single ring, the single-ring point cloud in the inter-ring segmentation results is segmented into tunnel segments within the ring, resulting in the following segmentation results: Convert the single-ring point cloud into circumferential angle distribution data with the tunnel ring center as a reference; The angle range corresponding to different types of tunnel segments is determined based on the design and arrangement angle of the tunnel segments. Count the number of point clouds corresponding to different coarse classification labels within each angle interval; The coarse classification labels that meet the preset conditions for the proportion of point cloud quantity in each angle interval are determined as the segment block labels for the corresponding angle interval. The segmentation result of the inner ring segment is obtained based on the segment block label.

8. The method according to claim 1, characterized in that, The single-ring point cloud is fitted with a center plane, and the single-ring cross-sectional point cloud is extracted based on the fitted plane, including: Calculate the centroid of a single-ring point cloud; Subtract the centroid from the coordinates of each point in the single-ring point cloud to obtain the centered point cloud; Construct a covariance matrix based on the centralized point cloud; Singular value decomposition is performed on the covariance matrix to obtain the eigenvectors corresponding to the minimum singular values; The feature vector is used as the normal vector of the single-loop fitting plane, and the single-loop fitting plane is determined based on the centroid. Calculate the distance from each point in the single-ring point cloud to the single-ring fitting plane; The cross-sectional point cloud located at the center of the single ring is extracted based on the distance.

9. The method according to claim 1, characterized in that, Based on the geometric parameters of the single-ring cross-section point cloud, the tunnel convergence deformation parameters and misalignment deformation parameters are calculated, including: The point cloud of the single-ring cross-section is fitted with a circle, an ellipse, or a B-spline curve. When a circle or ellipse is fitted, the cross-section radius, horizontal diameter, vertical diameter, major axis length, and / or minor axis length are obtained. When a B-spline curve is fitted, the cross-section profile curve is obtained, and the circumferential radius distribution is calculated based on the cross-section profile curve. The distribution of the cross-sectional radius, horizontal diameter, vertical diameter, major axis length, minor axis length and / or circumferential radius is compared with the design value, standard test value or historical test value; The tunnel convergence deformation was obtained based on the comparison results. Based on the segmentation results of the inner ring tube segments, the boundary point cloud between adjacent tube segments is extracted; Calculate the distance from the boundary point cloud to the center of the complete loop fitting or the center of the overall fitting circle; The distances are sequentially numbered along the tunnel circumference at preset angular intervals to form a circumferential distance sequence; The location and amount of misalignment are determined based on the abrupt change in the circumferential distance sequence at the connection point of adjacent segments.

10. A tunnel lining segmentation and cross-sectional deformation detection system, characterized in that, include: The point cloud acquisition module is used to acquire three-dimensional point cloud data of the target tunnel segment and establish a tunnel point cloud coordinate dataset based on the point cloud coordinate information in the three-dimensional point cloud data. The region division module is used to determine a reference line representing the longitudinal direction of the tunnel, and based on the reference line, divide the tunnel point cloud coordinate dataset into multiple local region point clouds according to the longitudinal position and circumferential angle of the tunnel. The point cloud denoising module is used to perform local plane fitting on the point cloud of each local region to obtain the local fitting plane, calculate the distance from each point in the local region point cloud to the local fitting plane, and take the points with a distance less than a preset threshold as the tunnel lining point cloud. The inter-ring segmentation module is used to perform coarse clustering classification based on spatial proximity of the tunnel lining point cloud to obtain coarse classification labels for the tunnel lining point cloud. Based on the coarse classification labels, a scalar field of the tunnel point cloud is constructed. A Gaussian mixture model is used to fit the probability distribution of the scalar field of the tunnel point cloud. The inter-ring boundary interval between adjacent tunnel rings is determined according to the boundary point between adjacent Gaussian components to obtain the inter-ring segmentation result of the tunnel. The inner ring segmentation module is used to segment the single-ring point cloud in the tunnel inter-ring segmentation result into inner ring segmentation blocks according to the design arrangement rules of the inner ring segmentation blocks, so as to obtain the inner ring segmentation result. The cross-section extraction module is used to extract single-ring point clouds based on the tunnel inter-ring segmentation results, perform center plane fitting on the single-ring point clouds, and extract single-ring cross-section point clouds based on the fitted plane. The deformation analysis module is used to calculate tunnel convergence deformation parameters based on the geometric parameters of the point cloud of the single ring section; determine the boundary position of adjacent segments within the same single ring based on the segmentation results of the segments within the ring; and calculate the misalignment deformation parameters based on the geometric features of the point cloud at the boundary of adjacent segments.

Images