A tunnel lining surface flatness detection method based on three-dimensional laser scanning

Abstract

This invention relates to the technical field of tunnel construction quality inspection, and discloses a method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning. The method involves acquiring three-dimensional point cloud data of the tunnel; extracting the tunnel's central axis using a bilinear projection method combined with the RANSAC algorithm; transforming the three-dimensional point cloud data from the world coordinate system XYZ to a local coordinate system X´Y´Z along the tunnel's direction; performing equidistant slicing along the Z-axis of the local coordinate system, then performing left-right separation operations, and dividing the data into equal-length segments along the tunnel axis to generate a series of independent point cloud segments; for each point cloud segment, traversing all point pairs within that segment to obtain the final stable posture after rotation and translation transformation; calculating the vertical distance from each point within each segment to the y´=0 reference plane under the final stable posture, thereby characterizing the smoothness of the corresponding point cloud segment.

Description

Technical Field

[0001] This invention belongs to the technical field of tunnel engineering construction quality inspection, specifically relating to a method for detecting the surface flatness of tunnel lining based on three-dimensional laser scanning. Background Technology

[0002] In tunnel construction, the flatness of the lining surface is a key quality indicator affecting structural safety and waterproofing performance. According to the "Code for Acceptance of Construction Quality of Concrete Structures" (GB50204), the flatness test should use a combination measurement method of "2-meter straightedge + wedge gauge". The physical principle is to measure the maximum gap between the straightedge and the surface after the rigid straightedge comes into contact with the convex point of the surface being measured.

[0003] With the development of measurement technology, flatness detection methods based on 3D laser scanning have gradually become more widespread. However, existing technical solutions mostly use the least squares method to fit the tunnel surface and calculate the vertical distance from the point cloud to the fitted surface to characterize flatness. This method essentially only reflects the overall fluctuation of the surface. It not only fails to reproduce the physical mechanism of the straightedge with "convex point support contact" as specified in the national standard, but also introduces false tilt components into the calculated gap value because it does not fully consider the influence of local tunnel tilt on the straightedge's posture. Ultimately, this results in a systematic deviation of ±8% to ±15% between the detection results and the manually measured values, making it difficult to meet the requirements of high-precision acceptance. There is an urgent need for a digital detection method that strictly follows the principle of "convex point support and gravity direction measurement" to bridge the gap between 3D scanning technology and manual measurement standards. Summary of the Invention

[0004] This invention provides a method for detecting the surface flatness of tunnel lining based on three-dimensional laser scanning, which solves the problems of existing methods being unable to reproduce the physical mechanism of the straightedge with "convex point support contact" as specified in the national standard, resulting in low measurement accuracy and difficulty in meeting practical needs.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning includes the following steps:

[0007] Step 1: Obtain the 3D point cloud data of the tunnel;

[0008] Step 2: Extract the tunnel centerline based on the bilinear projection method combined with the RANSAC algorithm, and transform the three-dimensional point cloud data from the world coordinate system XYZ to the local coordinate system X´Y´Z along the tunnel direction;

[0009] Step 3: Perform equidistant slicing along the Z-axis of the local coordinate system, then perform left and right separation operations, and divide the data into equal length segments along the tunnel axis to generate a series of independent point cloud fragments;

[0010] Step 4: For each point cloud segment, iterate through all point pairs of any two points in the point cloud segment to obtain the final stable attitude after rotation and translation transformation.

[0011] Step 5: Calculate the vertical distance from each point in each point cloud segment to the y´=0 reference plane under the final stable attitude, so as to characterize the flatness of the point cloud segment.

[0012] Furthermore, in step four, the point cloud fragment is projected onto the X´OY´ plane, and all point pair combinations are traversed. The rotation angle required to rotate the line connecting the two points of each point pair to a horizontal state is calculated first. The point cloud segment is then rotated and translated so that the two points fall behind the y´=0 reference plane. The stability score at this pose is then calculated, and the pose with the highest stability score is selected as the final stable pose.

[0013] Furthermore, when performing rotation and translation transformation, firstly, the point cloud fragment is rotated horizontally by an angle θ so that the line connecting the point pairs is horizontal. Then, the point cloud fragment is translated along the Y´ axis so that the two points fall on the y´=0 reference plane, that is, their ordinate is zeroed.

[0014] Furthermore, let the two points of any given point be respectively... and The rotation angle is calculated using the following formula. ,

[0015]

[0016] The stability score is calculated using the following formula. ,

[0017]

[0018] in, This represents the distance from the geometric center point of the point cloud segment to the support line segment. The difference between the x-coordinates (X') of the midpoints; Indicates support line segment Length; Indicates numerical tolerance.

[0019] Furthermore, the local coordinate system X´Y´Z is set with the origin of the world coordinate system as the origin, the direction of the tunnel's central axis as the X´ axis, and the vertically upward direction (the opposite direction of gravity) as the Z-axis. The Y´ axis is determined by the X´ axis and the Z-axis according to the right-hand rule.

[0020] In step three, the three-dimensional point cloud in the local coordinate system X´Y´Z is sliced ​​along the Z-axis using slices of preset thickness. Then, a left-right separation operation is performed with the Y´ axis as the boundary. Next, the point cloud is divided into equal length segments along the tunnel axis at a standard length of L=2m to generate a series of independent point cloud segments. Finally, point cloud segments with fewer than a set threshold are removed, and the remaining point cloud segments are optimized for data density using a voxel downsampling algorithm to obtain the final point cloud segments.

[0021] Furthermore, in step two, firstly, the three-dimensional point cloud data of the tunnel is orthogonally projected onto the XOZ plane and the YOZ plane respectively. Then, the centerlines of the contour boundaries on both sides of the tunnel are extracted in the XOZ plane as horizontal candidate lines, and the centerlines of the upper and lower contour boundaries of the tunnel are extracted in the YOZ plane as vertical candidate lines. Then, the three-dimensional spatial discrete points (X,Y,Z) corresponding to the tunnel central axis are reconstructed based on the horizontal and vertical candidate lines. Finally, the RANSAC algorithm is used to perform spatial line fitting on these discrete points to obtain the straight line where the central axis is located.

[0022] Furthermore, firstly, any point on the central axis is selected as the intercept point, and its direction vector is the extension direction of the tunnel. Then, the three-dimensional point cloud data in the world coordinate system is transformed to the local coordinate system through a rigid transformation operation. That is, the intercept point B of the three-dimensional point cloud is moved to the origin of the local coordinate system through a translation operation. Then, the translated three-dimensional point cloud is rigidly rotated through a rotation operation so that the direction vector of the central axis of the three-dimensional point cloud is completely coincident with the X' axis of the local coordinate system.

[0023] Furthermore, in step one, the number of neighborhood points is set to 20, and the standard deviation multiple is 1.5 to effectively filter out dust and splash noise. During the data acquisition process, a multi-segment step-by-step scanning method is first adopted, requiring an overlap area of ​​≥30% to obtain multi-segment three-dimensional point cloud data of the tunnel. Then, point cloud denoising is performed based on the statistical outlier removal algorithm (SOR). Finally, the multi-segment three-dimensional point cloud data is stitched together based on the ICP algorithm to obtain complete three-dimensional point cloud data of the tunnel.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] 1. Overcoming the limitations of statistical fitting, it realistically reproduces the national standard measurement process from the perspective of physical mechanisms:

[0026] This invention abandons the traditional approach of relying on statistical principles such as least squares to calculate the "average distance" from point clouds to ideal surfaces, as is common in existing technologies. Instead, it innovatively constructs a 2-meter-long virtual ruler rigid body model, i.e., a 2-meter point cloud segment (rigid point cloud), and a y´=0 reference plane (virtual ruler), and introduces strict geometric anti-penetration constraints at the algorithm level. By employing mechanically stable contact constraints, this invention accurately simulates the process of a rigid body freely falling in a gravitational field and finding the optimal contact point pair in computer space. This method successfully replicates the real physical mechanism of the "2-meter straightedge + wedge gauge" measurement method specified in the national standard (GB50204), where the straightedge provides "convex point support for contact," fundamentally solving the problem of essential deviations between traditional automated testing results and standardized acceptance logic.

[0027] 2. Precise coordinate system normalization and attitude envelope solution eliminate systematic biases caused by local tilt:

[0028] This invention accurately extracts the tunnel's centerline based on the biline projection method combined with the RANSAC algorithm, achieving coordinate system normalization of the 3D point cloud data. Building upon this, it solves for the minimum envelope attitude of the tunnel's local boundaries in the direction of gravity through rotation and translation transformations, simulating the process of a physical straightedge actively conforming to the tunnel wall. This dimensionality-reduced attitude search strategy effectively eliminates pseudo-tilt components caused by the overall tunnel structure's inclination or local settlement, completely eliminating the systematic bias of ±8% to ±15% commonly found in existing least-squares surface fitting techniques, and significantly improving the absolute accuracy of smoothness gap value calculation.

[0029] 3. Continuous full-section inspection and precise capture of "long-tail" defects improve the automation level and reliability of engineering acceptance:

[0030] Compared to the traditional manual "2-meter straightedge" sampling inspection method, which suffers from low efficiency, insufficient coverage, and safety hazards associated with working at heights, this system achieves continuous, high-density flatness assessment of the entire tunnel cross-section. By automatically outputting the maximum gap value and generating a flatness distribution heatmap and statistical histogram, this invention can not only efficiently verify the overall construction quality of large-scale components but also accurately capture localized defects exceeding standards in the "long tail" region of skewed distributions (such as extreme values ​​caused by localized shotcrete spalling or rebound). This provides comprehensive, objective, and traceable data support for subsequent refined repairs and acceptance, significantly improving the reliability of quality control for initial tunnel support and secondary lining. Attached Figure Description

[0031] Figure 1 This is a schematic diagram illustrating the design concept of the present invention.

[0032] Figure 2 This is a schematic diagram of the overall process of the present invention;

[0033] Figure 3 This is a specific example of each step in the detection method of the present invention, wherein, 'a' represents the acquired three-dimensional point cloud data, 'b' represents a partial calculation process of the central axis, 'c' represents the result after slicing, and 'd' represents the result of equal-length segmentation.

[0034] Figure 4 This is a histogram of flatness deviation distribution generated in a specific embodiment of the present invention. Detailed Implementation

[0035] To make the technical means, creative features, objectives and effects of the present invention easier to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the method for detecting the surface flatness of tunnel lining based on three-dimensional laser scanning. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0036] To address the problems of existing technologies, such as the inability of least squares fitting to reproduce the physical measurement process and the large systematic deviations caused by local tilting, this invention proposes a method for detecting the surface flatness of tunnel lining based on three-dimensional laser scanning. This method is a detection scheme based on rigid body contact constraints, treating a 2-meter point cloud segment divided into equal lengths as an indeformable "rigid body object," and then solving for the surface flatness of the tunnel lining surface. The reference plane is abstracted as an absolutely flat "virtual straightedge". A rotation and translation transformation is performed on the rigid body of the point cloud in the local coordinate system, and this transformation is forcibly applied in the algorithm. The geometric anti-penetration constraint, in a mathematical model, is essentially equivalent to: simulating the process of the point cloud rigid body falling freely downwards under the action of gravity until it collides with the virtual ruler below and reaches a mechanically stable state supported by at least two convex points, such as... Figure 1 As shown, the physical mapping model constructed at the algorithm level through the digital restoration of the 2-meter straightedge measurement method in the national standard (GB50204) using the above-mentioned physical mapping mechanism breaks through the technical bottleneck of point cloud floating on the average surface caused by traditional statistical surface fitting (such as the least squares method), and truly achieves strict alignment between digital detection and manual straightedge measurement in terms of physical mechanism.

[0037] like Figure 2 As shown, firstly, three-dimensional point cloud data of the tunnel is acquired. Then, the tunnel's central axis is extracted, and a local coordinate system is established to eliminate pseudo-tilt components. Subsequently, point cloud slicing and segmentation operations are performed to generate a series of independent point cloud segments. Finally, through rotation and translation transformations, the final stable posture of each point cloud segment is obtained. This allows the determination of the minimum vertical gap envelope posture of the virtual straightedge in the direction of gravity. The vertical distance from each point to the virtual straightedge under this posture is then calculated as the detection result. This invention's detection method realistically replicates the physical measurement mechanism of a "2-meter straightedge + wedge gauge," effectively eliminating systematic deviations of ±8% to ±15%, and significantly improving the accuracy and automation level of tunnel initial support and secondary lining flatness detection.

[0038] Specifically as follows:

[0039] S1: Acquire the 3D point cloud data of the tunnel and perform preprocessing.

[0040] S11. Use a Trimble X7 terrestrial 3D laser scanner or an equivalent precision device (such as a Leica RTC360) to acquire 3D point cloud data. This device is based on digital pulse EDM technology, with a ranging range of 0.6~80.0m, a measurement accuracy of 2mm, a scanning field of view of 360°×282°, and a scanning resolution set to... To ensure the capture of surface micro-protrusion features, it exhibits strong environmental adaptability, with a single-station operation time of less than 15 minutes. During data acquisition, a multi-segment step-by-step scanning method is employed, with scanning stations set at predetermined intervals (e.g., 10 meters) along the tunnel axis, ensuring sufficient overlap (overlap rate ≥ 30%) between adjacent stations. The acquired raw measured point cloud completely records the spatial geometric information of the tunnel interior wall, such as... Figure 3 As indicated by label (a), this point cloud segment is approximately 28m long, 17.4m wide, and has a clearance of approximately 9.2m from the arch to the road surface.

[0041] S12. Next, data preprocessing is performed, and the process is as follows:

[0042] First, the statistical outlier removal algorithm (SOR) is used to denoise the point cloud data. The number of neighborhood points is set to 20 and the standard deviation factor is 1.5, which effectively filters out dust and splash noise.

[0043] When performing point cloud registration, the first station's scanning coordinate system is used as the global reference. The overlapping geometric features between adjacent stations are utilized, and the local point clouds of each station are rigidly stitched to the global reference coordinate system based on the Iterative Closest Point (ICP) algorithm. Multiple segments of collected data are integrated, and the overall registration error RMS of multiple stations is controlled within 2mm to obtain complete three-dimensional point cloud data of the tunnel.

[0044] S2: Extract the tunnel centerline based on the bilinear projection method combined with the RANSAC algorithm, and convert the three-dimensional point cloud data obtained in step S1 to a local coordinate system along the tunnel direction.

[0045] S21. Project the three-dimensional point cloud of the tunnel orthogonally onto the XOZ plane and YOZ plane respectively;

[0046] S22. Extract the centerlines of the upper and lower boundaries of the tunnel in the YOZ plane as vertical candidate lines, and extract the centerlines of the contour boundaries on both sides of the tunnel in the XOZ plane as horizontal candidate lines, so as to obtain the geometric features of the tunnel orientation in two orthogonal two-dimensional planes.

[0047] S23. To convert two-dimensional projections into three-dimensional spatial lines, the two-dimensional coordinates (X, Z) of each point on the horizontal candidate line of the XOZ plane and the two-dimensional coordinates (Y, Z) of the corresponding points on the vertical candidate line of the YOZ plane are extracted respectively. Using the Z coordinate representing the tunnel's depth as a common matching benchmark, a spatial mapping relationship between the x-coordinate (X) and y-coordinate (Y) is established. For a given Z coordinate benchmark, based on interpolation algorithms or the nearest neighbor matching principle, the X and Y coordinates at the corresponding Z points in the two orthogonal planes are obtained respectively. Then, the spatial coordinates are combined to accurately reconstruct the two two-dimensional plane lines distributed in different orthogonal planes into a discrete set of points (X, Y, Z) of the tunnel's central axis in three-dimensional space, as detailed below:

[0048] First, orthogonal projection and step size settings are performed. Complete 3D point cloud data of the tunnel section is selected and orthogonally projected onto the XOY and YOZ planes respectively. Using the Y-axis, which reflects the overall orientation of the tunnel, as a reference, an appropriate search step size is set. .

[0049] Then, the search for the envelope extrema proceeds along the path. Starting from the minimum value of the point cloud data in the Y direction, the search proceeds with a set step size. The point cloud is divided into n intervals along the Y direction. Within any step size interval... Within, search and extract the maximum value of the projected point cloud along the X-axis. and minimum value and the maximum value in the Z-axis direction. and minimum value .

[0050] Then, calculate the coordinates of the cross-section center. For each interval... Calculate the average coordinates of the points in the X and Z directions respectively, and use these averages as the coordinate components of the midline point of the section. The calculation formula is as follows:

[0051]

[0052]

[0053] Finally, the central axis in three-dimensional space is reconstructed. The calculated central axis is then... and Its corresponding vertical interval coordinates By combining spatial coordinates, the discrete set of points representing the tunnel's central axis can be accurately reconstructed in three-dimensional space. A spatial line fitting algorithm can then be used on this discrete point set to obtain the equation of the continuous central axis.

[0054] S24. The horizontal and vertical candidate lines are merged to generate a three-dimensional discrete point set. The RANSAC algorithm is then used to fit these discrete points to a spatial straight line to obtain the tunnel's central axis. This process accurately solves the spatial straight line equation representing the tunnel's central axis and obtains the direction vector of the tunnel's orientation relative to the central axis. Take any point on the central axis as its intercept point. The direction vector obtained by calculation and fitting intercept point 6.72306347, -4.34534454, 1633.18969727 ,like Figure 3 The identifier (b) is shown in the image.

[0055] S25. Using rigid body transformation, the complete 3D point cloud data obtained in step S1 is transformed into a local coordinate system.

[0056] First, the three reference axes of the local coordinate system used for flatness analysis are defined as follows: the origin of the world coordinate system is taken as the origin, the direction of the tunnel centerline is taken as the X' axis, the vertical upward direction (i.e., the opposite direction of gravity) is taken as the Z axis, and the Y' axis is determined by the X' axis and the Z axis according to the right-hand rule, with its origin being the origin of the world coordinate system.

[0057] Subsequently, the rotation matrix R for implementing the rigid body transformation operation is calculated using the following formula to determine the angle and direction of rotation in the point cloud space.

[0058] The rotation matrix R is constructed based on Rodriguez's formula, and its calculation formula is as follows:

[0059]

[0060] in, For rotation matrix, Let be the antisymmetric matrix of the rotational axis. For rotation angle, It is an identity matrix.

[0061] The specific transformation process is as follows: First, the 3D point cloud is translated as a whole through a translation operation, so that the intercept point B of the fitted central axis is moved to the origin of the world coordinate system, and this point is used as the origin of the local coordinate system; then, the rotation matrix R is applied to the translated 3D point cloud to perform a rigid body rotation, so that the direction vector A of the tunnel central axis is completely coincident with the X' axis of the local coordinate system. In this way, the point cloud coordinates are normalized through this transformation, that is, the overall transformation of the 3D point cloud from the original global scanner coordinate system to the local flatness analysis coordinate system, i.e., the local coordinate system, is completed.

[0062] S3: Slice and segment the normalized tunnel point cloud.

[0063] First, slice the point cloud after coordinate normalization, along the Z-axis of the local coordinate system. To achieve this, the tunnel point cloud was sliced ​​at equal intervals, resulting in multiple point cloud slices. To ensure data validity and avoid vertical overlap of point cloud data, the slice thickness was strictly set to a specific value. .

[0064] Then, a left-right separation operation is performed on the point cloud slices. Utilizing the spatial characteristics after coordinate normalization, the left and right boundary lines of the point cloud are automatically defined based on the sign of the Y-coordinate. In the established local coordinate system, observing along the positive X' axis (tunnel excavation direction), according to the right-hand rule, the positive Y' axis points to the left, and the negative Y' axis points to the right. Specifically, point cloud regions with Y' < 0 are identified as the right boundary line, and regions with Y' > 0 are identified as the left boundary line, thereby achieving automated extraction and classification of boundary lines.

[0065] Subsequently, the point cloud boundary line from the previous step is segmented into equal-length segments, and the extracted boundary line is then segmented along the X-axis, i.e., along the tunnel axis. Divide the data into standard length segments to generate a series of independent boundary segments. To ensure data validity, a data integrity threshold is set during this process. The fragment quality is controlled by automatically removing invalid fragments with insufficient points. For the remaining point cloud fragments, the data density is optimized based on the voxel downsampling algorithm, and the grid size is dynamically adjusted to optimize the data density while preserving curvature features.

[0066] Voxel downsampling is a data simplification and filtering algorithm for 3D point clouds. Its technical solution is to divide the original 3D point cloud space into countless regular tiny 3D cubes (i.e., "voxel mesh") according to a set size. Then, the algorithm will traverse each voxel, calculate the geometric centroid of all discrete points falling inside the voxel, and retain only the centroid point to replace the original multiple discrete points.

[0067] In this embodiment, 92 sets of point clouds were successfully extracted along the longitudinal direction of the tunnel, such as... Figure 3 As indicated by marker (c), it is divided into 1267 groups of equal-length segments of 2m each, as shown in the figure. Figure 3 As indicated by label (d) in the diagram, this lays the data foundation for subsequent high-precision analysis.

[0068] S4. For each point cloud segment, iterate through all point pairs of any two points in the point cloud segment to obtain the final stable attitude after rotation and translation transformation.

[0069] S41. Dimensionality Reduction: By projecting the point cloud fragments obtained in step S3 onto the X´OY´ plane of the local coordinate system, this system essentially transforms the three-dimensional flatness detection problem of the tunnel surface into a series of two-dimensional planes, namely the X´-Y´ plane of the local coordinate system, to detect the flatness of local contour lines. This dimensionality reduction mechanism is based on two core considerations: First, at the physical mechanism level, the "2-meter straightedge + wedge gauge" measurement method specified in the national standard (GB50204) essentially involves attaching a physical straightedge to a local contour line of a specific orientation on the wall for measurement, which is a typical two-dimensional linear contact physical model. The dimensionality reduction extraction of this invention strictly restores the physical detection dimension of this specification. Second, at the algorithm performance level, after transforming the massive three-dimensional surface point cloud into an independent set of two-dimensional boundary fragments, the subsequent virtual straightedge attitude solution only needs to be performed in the two-dimensional attitude space (i.e., translation and single-axis rotation in the two-dimensional plane), which not only avoids the systematic deviation caused by traditional three-dimensional surface fitting, but also significantly reduces the computational power consumption and computational complexity of envelope attitude search.

[0070] S42. Select the final stable posture.

[0071] First, set geometric constraints, requiring that the ordinates of all points in the boundary segment after rigid body transformation must satisfy... This constraint aims to ensure that the virtual straightedge model is always above the reference plane, preventing the non-physical phenomenon of a rigid body penetrating the reference plane, thus conforming to the real physical logic of the straightedge fitting against the wall.

[0072] Subsequently, the rotation angle is calculated, and any two points in the boundary fragment point cloud are traversed. and All point-pair combinations, total There are several scenarios. For each pair of points, calculate the rotation angle required to rotate the line connecting these two points to a horizontal position. The calculation formula is as follows:

[0073]

[0074] Then, a rotation and translation transformation is performed to forcefully "level" and "fit" the selected two support protrusions using a mathematical model. Specifically: First, the point cloud fragment is rotated horizontally using an angle θ to eliminate the inclination of the line connecting the two points, making their ordinates equal, i.e., in an absolutely horizontal state; then, a translation along the Y' axis is performed to precisely place these two points on the reference plane of y=0, making the ordinates of these two points equal and thus zero. This process of rotation followed by translation perfectly simulates the physical action of placing a physical ruler stably on the two protrusions, thereby transforming the bottom surface of the ruler into an absolute coordinate system zero reference. This lays the computational foundation for subsequent "anti-penetration" checks (i.e., determining whether the remaining points satisfy y≥0) and directly reading the y-coordinate as the vertical gap value.

[0075] After performing rotation and translation transformations, the point and Precisely landed After the reference plane, calculate the stability score of the point cloud segment in that pose. The calculation formula is as follows:

[0076]

[0077] in, For stability score; x-coordinate of the projection of the geometric center point To support line segment x-coordinate of the midpoint The difference between them; For supporting line segments Length; This represents the numerical tolerance.

[0078] The aforementioned geometric center point refers to the spatial geometric centroid of all point clouds within the current 2m-long point cloud segment. If there are n valid point clouds within this segment, the coordinates of any point are... Then the x-coordinate of the geometric center y-axis and vertical coordinates All are obtained by calculating the arithmetic mean of the coordinates of all discrete points. For example, the x-coordinate is calculated as follows: .

[0079] Finally, the stability score is selected. The highest pose is taken as the final stable pose, and it is required that all points within the point cloud segment corresponding to the final stable pose satisfy the coordinates. Furthermore, the geometric constraint of the point pairs with coordinates y´= 0, which serves as the support line segment, is applied, and points that do not meet the geometric constraint are directly eliminated.

[0080] S5: Under the final stable attitude, calculate the vertical distance from each point in the point cloud segment to the y=0 reference plane to characterize the flatness of the point cloud segment.

[0081] The calculation and analysis results show that in the tunnel arch region, at Z=4.21m and X´=8~10m, [the following is a partial sentence and requires more context:] ... The maximum gap exists at this location. ,exist The gap is The result exceeds the allowable value specified in the standard (e.g., The system automatically determined that there was a risk of shotcrete falling off in the area and recommended re-spraying.

[0082] A second aspect of the present invention provides a tunnel lining surface flatness detection system based on three-dimensional laser scanning, comprising:

[0083] Point cloud preprocessing module: used to denoise and register the raw point cloud, integrating SOR denoising and ICP registration functions, and supporting real-time stream processing of point cloud data;

[0084] The centerline extraction module is used to perform bilinear projection and RANSAC fitting operations to extract the tunnel centerline and establish a local coordinate system. It has a built-in threshold adaptive algorithm to adapt to different tunnel scenarios.

[0085] Rigid body attitude solving module: It is used to slice and segment point cloud data, and perform virtual ruler rotation and translation transformation search for each point cloud segment to determine the minimum vertical gap envelope attitude, which is the final stable attitude. The core is an attitude traversal engine that supports a phased search strategy.

[0086] Gap Calculation Module: Used to output gap values ​​as a smoothness index and generate a heat map of gap distribution for tunnel surface smoothness. After determining the final stable attitude, it calculates and outputs the vertical distance between the virtual straightedge (i.e., the y´=0 reference plane) and the point cloud segment on the tunnel surface, using this distance as a smoothness index for that local segment, and generates a heat map or bar chart of gap distribution for tunnel surface smoothness.

[0087] The surface smoothness analysis of the Fenghuangshan Tunnel project on the G30 Expressway was conducted using the detection method of this invention. The tunnel is located in Jinya Town, Yuzhong County, Lanzhou City, Gansu Province. It is a twin-tube long tunnel with separate left and right lanes. The right lane is from YK26+530 to YK30+475, with a length of 3945m and a maximum burial depth of 185m. The left lane is from ZK26+542 to ZK30+487, also with a length of 3945m and a maximum burial depth of 180m. Three-dimensional laser scanning data was collected at ZK26+540, approximately 28m long and 17.4m wide, with a clearance of approximately 9.2m from the arch to the road surface. Figure 3As shown in label (a), after slicing and segmenting according to the above method, a total of 1267 point cloud segments were obtained. After calculation, a histogram of flatness deviation distribution was generated, as shown in... Figure 4 As shown, the overall pass rate reached 87.2%, and the average flatness deviation was only 2.74 mm. Therefore, the detection method of this invention can not only verify the overall quality of large-scale components, but also accurately capture local defects exceeding the standard in the "long tail" region of the skewed distribution, providing a reliable basis for refined repair.

[0088] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples. Various changes or modifications can be made to these embodiments without departing from the principles and essence of the present invention. Therefore, the scope of protection of the present invention is defined by the appended claims.

[0089] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Various modifications or variations that can be made by those skilled in the art without creative effort within the scope of the appended claims are still within the scope of protection of the present invention.

Claims

1. A method for detecting the surface flatness of tunnel lining based on three-dimensional laser scanning, characterized in that: Includes the following steps: Step 1: Obtain the 3D point cloud data of the tunnel; Step 2: Extract the tunnel centerline based on the bilinear projection method combined with the RANSAC algorithm, and transform the three-dimensional point cloud data from the world coordinate system XYZ to the local coordinate system X´Y´Z along the tunnel direction; Step 3: Perform equidistant slicing along the Z-axis of the local coordinate system, then perform left and right separation operations, and divide the data into equal length segments along the tunnel axis to generate a series of independent point cloud fragments; Step 4: For each point cloud segment, iterate through all point pairs of any two points in the point cloud segment to obtain the final stable attitude after rotation and translation transformation. Step 5: Calculate the vertical distance from each point in each point cloud segment to the y´=0 reference plane under the final stable attitude, so as to characterize the flatness of the point cloud segment. In step four, the point cloud fragment is projected onto the X´OY´ plane, and all point pair combinations are traversed. First, the rotation angle required to rotate the line connecting the two points of each point pair to a horizontal state is calculated. The point cloud segment is then rotated and translated so that the two points fall behind the y´=0 reference plane. The stability score at this pose is then calculated, and the pose with the highest stability score is selected as the final stable pose.

2. The method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning according to claim 1, characterized in that: When performing rotation and translation transformation, firstly, the point cloud fragment is rotated horizontally by an angle θ so that the line connecting the point pairs is horizontal. Then, the point cloud fragment is translated along the Y´ axis so that the two points fall on the y´=0 reference plane, that is, their ordinate is zeroed.

3. The method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning according to claim 1, characterized in that: Let any point be a pair of points and have the following two points: and The rotation angle is calculated using the following formula. , The stability score is calculated using the following formula. , in, This represents the distance from the geometric center point of the point cloud segment to the support line segment. The difference between the x-coordinates (X') of the midpoints; Indicates support line segment Length; Indicates numerical tolerance.

4. The method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning according to claim 1, characterized in that: The local coordinate system X´Y´Z is set with the origin of the world coordinate system as the origin, the direction of the tunnel's central axis as the X´ axis, and the vertically upward direction (the opposite direction of gravity) as the Z-axis. The Y´ axis is determined by the X´ axis and the Z-axis according to the right-hand rule. In step three, the three-dimensional point cloud in the local coordinate system X´Y´Z is sliced ​​along the Z-axis using slices of preset thickness. Then, a left-right separation operation is performed with the Y´ axis as the boundary. Next, the point cloud is divided into equal length segments along the tunnel axis at a standard length of L=2m to generate a series of independent point cloud segments. Finally, point cloud segments with fewer than a set threshold are removed, and the remaining point cloud segments are optimized for data density using a voxel downsampling algorithm to obtain the final point cloud segments.

5. The method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning according to claim 1, characterized in that: In step two, firstly, the three-dimensional point cloud data of the tunnel is orthogonally projected onto the XOZ plane and the YOZ plane respectively. Then, the center lines of the contour boundaries on both sides of the tunnel are extracted in the XOZ plane as horizontal candidate lines, and the center lines of the upper and lower contour boundaries of the tunnel are extracted in the YOZ plane as vertical candidate lines. Then, the three-dimensional spatial discrete points (X,Y,Z) corresponding to the tunnel central axis are reconstructed based on the horizontal and vertical candidate lines. Finally, the RANSAC algorithm is used to perform spatial line fitting on these discrete points to obtain the straight line where the central axis is located.

6. The method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning according to claim 5, characterized in that: First, select any point on the central axis as the intercept point, and its direction vector is the extension direction of the tunnel. Then, transform the 3D point cloud data in the world coordinate system to the local coordinate system through a rigid transformation operation. That is, move the intercept point B of the 3D point cloud to the origin of the local coordinate system through a translation operation. Then, perform a rigid body rotation on the translated 3D point cloud through a rotation operation so that the direction vector of the central axis of the 3D point cloud completely coincides with the X' axis of the local coordinate system.

7. The method for detecting the surface smoothness of tunnel lining based on three-dimensional laser scanning according to claim 1, characterized in that: In step one, the number of neighborhood points is set to 20, and the standard deviation multiple is 1.5 to effectively filter out dust and splash noise. During the data acquisition process, a multi-segment step-by-step scanning method is first used, requiring an overlap area of ​​≥30% to obtain multi-segment three-dimensional point cloud data of the tunnel. Then, point cloud denoising is performed based on the statistical outlier removal algorithm (SOR). Finally, the multi-segment three-dimensional point cloud data is stitched together based on the ICP algorithm to obtain complete three-dimensional point cloud data of the tunnel.

Images