A tunnel joint dislocation detection method, system and platform based on laser radar image

Abstract

The application discloses a tunnel joint error detection method, system and platform based on laser radar images. The method generates and acquires first data corresponding to the inner wall of the tunnel based on the laser radar, denoises and pre-processes the first data, and generates corresponding second data. The second data is calibrated and processed, and third data corresponding to the second data is generated. An improved U-Net model is constructed, and fourth data corresponding to the third data is generated based on the improved U-Net model. Based on the tunnel image data, a pixel and point cloud coordinate mapping relationship is created, and corresponding fifth data is extracted and generated. According to the RANSAC algorithm, the cross-sectional curves on both sides of the joint are fitted and processed, and sixth data corresponding to the tunnel joint is calculated and generated. The sixth data is the error amount data, and the system and platform corresponding to the method can effectively improve the precision and efficiency of joint detection and provide technical support for tunnel structure health monitoring.

Description

Technical Field

[0001] This invention belongs to the field of tunnel splice joint misalignment detection and processing technology, specifically involving a method, system and platform for detecting tunnel splice joint misalignment based on lidar imagery. Background Technology

[0002] Tunnels are a crucial component of urban infrastructure construction. With the rapid development of urban rail transit networks, the safety and stability of tunnels have become paramount. Structural health monitoring of tunnels, particularly tunnel deformation, segment joints, and changes in these joints, has become critical for tunnel maintenance and repair. Joint misalignment, as a significant form of tunnel deformation, can substantially impact tunnel stability, service life, and safety. Therefore, accurately monitoring segment joint misalignment and promptly implementing corrective measures is essential for ensuring the safe operation of tunnels.

[0003] Furthermore, in recent years, with the advancements in LiDAR (Light Detection and Ranging) technology and computer vision technology, tunnel deformation detection methods based on point cloud data have become a research hotspot. LiDAR can efficiently acquire three-dimensional data of the tunnel's inner wall, possessing high spatial resolution and accuracy, and can precisely describe the tunnel's geometry. By processing the point cloud data and combining it with machine learning and deep learning algorithms, rapid detection and quantitative analysis of tunnel deformation can be achieved.

[0004] In the inspection of tunnel segment joints, joint misalignment refers to the displacement of the cross-sectional positions on both sides of the joint. This phenomenon is usually caused by internal tunnel stress, soil settlement, or improper construction. The occurrence of misalignment may lead to uneven settlement of the tunnel structure, thus affecting its long-term stability. Therefore, accurate identification and calculation of joint misalignment provides an important basis for tunnel maintenance and reinforcement.

[0005] Therefore, in view of the above-mentioned technical problems and defects, there is an urgent need to design and develop a method, system and platform for detecting misalignment of tunnel splice joints based on lidar imagery. Summary of the Invention

[0006] To overcome the shortcomings and difficulties of the existing technologies, this invention provides a method, system, and platform for detecting misalignment at tunnel splice joints based on lidar imagery. The method identifies joint areas in tunnel images using an improved U-Net model and combines this with the RANSAC algorithm to fit the cross-sectional curves on both sides of the joint, calculating the misalignment. This invention effectively improves the accuracy and efficiency of joint detection, providing technical support for tunnel structural health monitoring.

[0007] The first objective of this invention is to provide a method for detecting misalignment in tunnel splice joints based on lidar imagery; the second objective of this invention is to provide a system for detecting misalignment in tunnel splice joints based on lidar imagery; and the third objective of this invention is to provide a platform for detecting misalignment in tunnel splice joints based on lidar imagery.

[0008] The first objective of this invention is achieved as follows: the method comprises the following steps:

[0009] Based on lidar, first data corresponding to the tunnel inner wall is generated and acquired, and the first data is pre-processed for denoising, and corresponding second data is generated; wherein, the first data is three-dimensional point cloud data; and the second data is three-dimensional point cloud data after denoising preprocessing.

[0010] The second data is calibrated, and a third data corresponding to the second data is generated; wherein, the calibration process includes longitudinal calibration and circumferential calibration; the third data is distortion-free orthophoto grayscale image data;

[0011] An improved U-Net model is constructed, and a fourth data corresponding to the third data is generated based on the improved U-Net model; wherein, the fourth data is the segment splicing seam area data;

[0012] Based on tunnel image data, a pixel-to-point cloud coordinate mapping relationship is created, and the corresponding fifth data is extracted and generated; and the cross-sectional curves on both sides of the joint are fitted and processed according to the RANSAC algorithm to calculate and generate the sixth data corresponding to the tunnel splice joint; wherein, the fifth data is the cross-sectional point cloud data on both sides of the joint; and the sixth data is the misalignment amount data.

[0013] Furthermore, the process of generating and acquiring first data corresponding to the tunnel inner wall based on lidar, performing noise reduction preprocessing on the first data, and generating corresponding second data further includes:

[0014] Two-dimensional spiral path scanning based on lidar generates and acquires the first data corresponding to the tunnel cross-section;

[0015] The K-means algorithm is used to remove outlier and erroneous data points from the first data, and the distance threshold method is used to filter out noisy data points from the first data.

[0016] Furthermore, the calibration process for the second data and the generation of third data corresponding to the second data further includes:

[0017] Generate and acquire the seventh data corresponding to the lidar, and calculate and generate the corresponding eighth data based on the seventh data; wherein, the seventh data is lidar status data, including lidar rotation speed data, carrier speed data, and sleeper spacing data; the eighth data is scan line count data;

[0018] Based on the eighth data, the real-time scan line count data is adjusted by combining linear interpolation or section removal.

[0019] Furthermore, the calibration process for the second data and the generation of third data corresponding to the second data further includes:

[0020] Based on a piecewise linear fitting algorithm, a ninth set of data corresponding to the tunnel cross-section is generated; wherein, the ninth set of data is fitted polygon data.

[0021] The ninth data is processed to eliminate distortion, and then circumferentially corrected using a fixed-interval division method to generate the corresponding third data.

[0022] Furthermore, the construction of the improved U-Net model, and the identification and generation of fourth data corresponding to the third data based on the improved U-Net model, further includes:

[0023] A hybrid convolution module is constructed, and based on the hybrid convolution module, tenth data corresponding to the tunnel image is extracted and generated; wherein, the tenth data is image texture feature data, including: image texture feature data in the horizontal direction, vertical direction and diagonal direction;

[0024] The encoder feature data is fused through upsampling and skip connections, and the hybrid convolution module is used to enhance the detailed data information.

[0025] Based on the Sigmoid activation function, an eleventh data point is generated that corresponds to an image pixel and belongs to the stitching seam region; wherein, the eleventh data point is probability data.

[0026] Furthermore, the process of creating a pixel-to-point cloud coordinate mapping relationship based on tunnel image data and extracting and generating corresponding fifth data; and calculating and generating sixth data corresponding to the tunnel splicing joint by fitting and processing the cross-sectional curves on both sides of the joint using the RANSAC algorithm, also includes:

[0027] The point cloud coordinates corresponding to each pixel are averaged, and the original point cloud data is located from the image pixels using reverse indexing.

[0028] Point cloud data from both sides of the seam are randomly sampled, an arc model is fitted, and internal points are selected. The twelfth data corresponding to both sides of the seam is generated through iterative optimization. The twelfth data is the optimal fitted curve data.

[0029] Based on the twelfth data, the sixth data corresponding to the tunnel splice joint is calculated and generated; the calculation formula is as follows:

[0030]

[0031] Where: x right It is the x-coordinate of the right-side arc and the ray from the origin to the seam position; left It is the x-coordinate of the left arc and the ray from the origin to the seam position; f left (x seam ) represents the ordinate of the intersection point of the fitted left arc with the ray from the origin to the seam position; f right (x seam ) represents the ordinate of the intersection point of the fitted right-side arc with the ray from the origin to the seam position.

[0032] The second objective of this invention is achieved as follows: the system is used to implement the method for detecting misalignment at tunnel splice joints based on lidar imagery, the system comprising:

[0033] The first data generation unit is used to generate and acquire first data corresponding to the tunnel inner wall based on lidar, and to perform noise reduction preprocessing on the first data and generate corresponding second data; wherein, the first data is three-dimensional point cloud data; and the second data is three-dimensional point cloud data after noise reduction preprocessing.

[0034] A first data processing unit is used to calibrate the second data and generate third data corresponding to the second data; wherein the calibration process includes longitudinal calibration and circumferential calibration; and the third data is distortion-free orthophoto grayscale image data.

[0035] The second data processing unit is used to construct an improved U-Net model and, based on the improved U-Net model, identify and generate fourth data corresponding to the third data; wherein, the fourth data is segment splicing seam area data;

[0036] The second data generation unit is used to create a pixel-to-point cloud coordinate mapping relationship based on tunnel image data and extract and generate the corresponding fifth data; and to calculate and generate the sixth data corresponding to the tunnel splicing joint by fitting the cross-sectional curves on both sides of the joint according to the RANSAC algorithm; wherein, the fifth data is the cross-sectional point cloud data on both sides of the joint; and the sixth data is the misalignment amount data.

[0037] Furthermore, the first data generation unit further includes:

[0038] The first generation module is used for two-dimensional spiral path scanning based on lidar to generate and acquire first data corresponding to the tunnel cross-section;

[0039] The first processing module is used to combine the K-means algorithm to remove outlier and erroneous data in the first data, and to filter out noise data in the first data using the distance threshold method.

[0040] And / or, the first data processing unit further includes:

[0041] The second generation module is used to generate and acquire the seventh data corresponding to the lidar, and calculate and generate the corresponding eighth data based on the seventh data; wherein, the seventh data is lidar status data, including lidar rotation speed data, carrier speed data and sleeper spacing data; the eighth data is scan line count data;

[0042] The second processing module is used to adjust the real-time scan line count data based on the eighth data, combined with linear interpolation or section removal.

[0043] And / or, the second data processing unit further includes:

[0044] A generation module is constructed to build a hybrid convolution module and, based on the hybrid convolution module, extract and generate tenth data corresponding to the tunnel image; wherein, the tenth data is image texture feature data, including: image texture feature data in the horizontal direction, vertical direction and diagonal direction;

[0045] The third processing module is used to fuse encoder feature data through upsampling and skip connections, and to enhance detailed data information using the hybrid convolution module.

[0046] The third generation module is used to generate eleventh data corresponding to image pixels and belonging to the stitching seam region based on the Sigmoid activation function; wherein, the eleventh data is probability data;

[0047] And / or, the second data generation unit further includes:

[0048] The fourth processing module is used to average the point cloud coordinates corresponding to each pixel and locate the original point cloud data from the image pixels through reverse indexing.

[0049] The fitting generation module is used to randomly sample point cloud data on both sides of the seam, fit an arc model and filter the inner points, and generate a twelfth data corresponding to both sides of the seam through iterative optimization; wherein, the twelfth data is the optimal fitted curve data.

[0050] The fourth generation module is used to calculate and generate the sixth data corresponding to the tunnel splice joint based on the twelfth data; wherein the calculation formula is:

[0051]

[0052] Where: x right It is the x-coordinate of the right-side arc and the ray from the origin to the seam position; left It is the x-coordinate of the left arc and the ray from the origin to the seam position; f left (x seam ) represents the ordinate of the intersection point of the fitted left arc with the ray from the origin to the seam position; f right (x seam ) represents the ordinate of the intersection point of the fitted right-side arc with the ray from the origin to the seam position.

[0053] Furthermore, the first data processing unit further includes:

[0054] The fifth generation module is used to generate the ninth data corresponding to the tunnel cross-section based on a piecewise linear fitting algorithm; wherein, the ninth data is fitted polygon data;

[0055] The fifth processing module is used to perform distortion elimination processing on the ninth data and to perform circumferential correction processing on the ninth data using the fixed-distance equal division method to generate the corresponding third data.

[0056] The third objective of this invention is achieved as follows: the platform includes a processor, a memory, and a control program for a tunnel splice joint misalignment detection platform based on lidar imagery; wherein the control program for the tunnel splice joint misalignment detection platform based on lidar imagery is executed on the processor, and the control program for the tunnel splice joint misalignment detection platform based on lidar imagery is stored in the memory; the control program for the tunnel splice joint misalignment detection platform based on lidar imagery implements the method for detecting tunnel splice joint misalignment based on lidar imagery.

[0057] This invention utilizes a method based on lidar to generate and acquire first data corresponding to the tunnel inner wall, preprocesses the first data with noise reduction, and generates corresponding second data. The first data is 3D point cloud data; the second data is 3D point cloud data after noise reduction preprocessing. The second data is then calibrated to generate corresponding third data, including longitudinal and circumferential calibration. The third data is distortion-free orthophoto grayscale image data. An improved U-Net model is constructed, and a fourth data corresponding to the third data is generated based on the improved U-Net model. The fourth data is segment splicing joint area data. Based on the tunnel image data, a pixel-to-point cloud coordinate mapping relationship is created, and corresponding fifth data is extracted and generated. Finally, the cross-sectional curves on both sides of the joint are fitted using the RANSAC algorithm to calculate and generate sixth data corresponding to the tunnel splicing joint. The fifth data is cross-sectional point cloud data on both sides of the joint; the sixth data is misalignment data. The invention also includes a corresponding system and platform, which effectively improves the accuracy and efficiency of joint detection, providing technical support for tunnel structural health monitoring.

[0058] In other words, the present invention accurately identifies the joint locations in tunnel images by utilizing the improved U-Net's efficient target detection capabilities, and calculates the misalignment of the joint by fitting the cross-sectional curves on both sides of the joint using the RANSAC algorithm. The method of the present invention exhibits high accuracy and robustness in segment splicing joint detection and misalignment calculation. Especially in complex tunnel cross-sections and irregularly shaped joints, the RANSAC fitting algorithm can effectively handle noise points and anomalous data, providing accurate misalignment calculation results. In summary, the novel tunnel deformation monitoring method provided by this invention can provide strong support for tunnel structural assessment, maintenance decisions, and safety inspection. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a schematic diagram of the process steps of a method for detecting misalignment in tunnel splice joints based on lidar images according to the present invention.

[0061] Figure 2 This is a schematic diagram of the fitting process in an embodiment of the method for detecting misalignment of tunnel splice joints based on lidar images according to the present invention.

[0062] Figure 3This is a schematic diagram illustrating the encoding and decoding of an embodiment of the method for detecting misalignment in tunnel splice joints based on lidar images according to the present invention.

[0063] Figure 4 This is a schematic diagram illustrating the structure for calculating misalignment in an embodiment of a method for detecting misalignment at tunnel splice joints based on lidar imagery, according to the present invention.

[0064] Figure 5 This is a schematic diagram of the tunnel splice joint misalignment detection system based on lidar imagery according to the present invention.

[0065] Figure 6 This is a schematic diagram of the tunnel splice joint misalignment detection platform based on lidar imagery according to the present invention. Detailed Implementation

[0066] To facilitate a clearer understanding of the objectives, technical solutions, and advantages of this invention, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art can easily understand other advantages and effects of this invention from the content disclosed in this specification.

[0067] This invention can also be implemented or applied through other different specific examples, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of this invention.

[0068] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0069] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Secondly, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0070] Preferably, the tunnel splice joint misalignment detection method based on lidar imagery of the present invention is applied in one or more terminals or servers. The terminal is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0071] The terminal can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal can interact with the customer via a keyboard, mouse, remote control, touchpad, or voice control device.

[0072] This invention provides a method, system, and platform for detecting misalignment in tunnel splice joints based on lidar imagery.

[0073] like Figure 1 The diagram shown is a flowchart of a method for detecting misalignment of tunnel splice joints based on lidar images provided in an embodiment of the present invention.

[0074] In this embodiment, the tunnel splice joint misalignment detection method based on lidar imagery can be applied to a terminal or fixed terminal with display function. The terminal is not limited to personal computers, smartphones, tablets, desktop computers or all-in-one computers with cameras, etc.

[0075] The tunnel splice joint misalignment detection method based on lidar imagery can also be applied to a hardware environment consisting of a terminal and a server connected to the terminal via a network. The network includes, but is not limited to, a wide area network (WAN), a metropolitan area network (MAN), or a local area network (LAN). The tunnel splice joint misalignment detection method based on lidar imagery in this embodiment can be executed by the server, by the terminal, or by both the server and the terminal.

[0076] For example, for a terminal requiring tunnel splice joint misalignment detection based on LiDAR imagery, the tunnel splice joint misalignment detection function based on LiDAR imagery provided by the method of this invention can be directly integrated into the terminal, or a client for implementing the method of this invention can be installed. Alternatively, the method provided by this invention can also run on servers or other devices in the form of a Software Development Kit (SDK), providing an interface for the tunnel splice joint misalignment detection function based on LiDAR imagery. Terminals or other devices can then implement the tunnel splice joint misalignment detection function based on LiDAR imagery through the provided interface. The invention will be further described below with reference to the accompanying drawings.

[0077] like Figures 1-4 As shown, this invention provides a method for detecting misalignment at tunnel splice joints based on lidar imagery. The method includes the following steps:

[0078] S01. Based on lidar, generate and acquire first data corresponding to the tunnel inner wall, and preprocess the first data to remove noise, and generate corresponding second data; wherein, the first data is three-dimensional point cloud data; the second data is three-dimensional point cloud data after noise removal and preprocessing;

[0079] S02. The second data is calibrated and a third data corresponding to the second data is generated; wherein, the calibration process includes longitudinal calibration and circumferential calibration; the third data is distortion-free orthophoto grayscale image data;

[0080] S03. Construct an improved U-Net model, and based on the improved U-Net model, identify and generate fourth data corresponding to the third data; wherein, the fourth data is segment splicing seam area data;

[0081] S04. Based on tunnel image data, create a pixel-to-point cloud coordinate mapping relationship and extract and generate the corresponding fifth data; and use the RANSAC algorithm to fit and process the cross-sectional curves on both sides of the joint to calculate and generate the sixth data corresponding to the tunnel splice joint; wherein, the fifth data is the cross-sectional point cloud data on both sides of the joint; and the sixth data is the misalignment data.

[0082] The process of generating and acquiring first data corresponding to the tunnel inner wall based on lidar, performing noise reduction preprocessing on the first data, and generating corresponding second data further includes:

[0083] S011. Two-dimensional spiral path scanning based on lidar to generate and acquire the first data corresponding to the tunnel cross-section;

[0084] S012. Combine the K-means algorithm to remove outlier and erroneous data in the first data, and filter out noise data in the first data using the distance threshold method.

[0085] The calibration process for the second data, and the generation of third data corresponding to the second data, further includes:

[0086] S021. Generate and acquire the seventh data corresponding to the lidar, and calculate and generate the corresponding eighth data based on the seventh data; wherein, the seventh data is lidar status data, including lidar rotation speed data, carrier speed data, and sleeper spacing data; the eighth data is scan line count data;

[0087] S022. Based on the eighth data, adjust the real-time scan line count data by combining linear interpolation or section removal.

[0088] The calibration process for the second data, and the generation of third data corresponding to the second data, further includes:

[0089] S023. Based on the piecewise linear fitting algorithm, generate the ninth data corresponding to the tunnel cross-section; wherein, the ninth data is fitted polygon data;

[0090] S024. The ninth data is processed by distortion elimination, and the ninth data is circumferentially corrected by the fixed-distance equal division method to generate the corresponding third data.

[0091] The process of constructing an improved U-Net model and generating fourth data corresponding to the third data based on the improved U-Net model further includes:

[0092] S031. Construct a hybrid convolution module, and based on the hybrid convolution module, extract and generate tenth data corresponding to the tunnel image; wherein, the tenth data is image texture feature data, including: image texture feature data in the horizontal direction, vertical direction and diagonal direction;

[0093] S032. Encoder feature data is fused through upsampling and skip connections, and the hybrid convolution module is used to enhance detailed data information;

[0094] S033. Based on the Sigmoid activation function, generate eleventh data corresponding to the image pixels and belonging to the stitching seam region; wherein, the eleventh data is probability data.

[0095] The process of creating a pixel-to-point cloud coordinate mapping relationship based on tunnel image data and extracting and generating corresponding fifth data; and calculating and generating sixth data corresponding to the tunnel splicing joint by fitting the cross-sectional curves on both sides of the joint using the RANSAC algorithm, also includes:

[0096] S041. Average the point cloud coordinates corresponding to each pixel, and locate the original point cloud data from the image pixels using reverse indexing.

[0097] S042. Randomly sample point cloud data from both sides of the seam, fit an arc model and filter the inner points, and generate the twelfth data corresponding to both sides of the seam through iterative optimization; wherein, the twelfth data is the optimal fitted curve data;

[0098] S043. Based on the twelfth data, calculate and generate the sixth data corresponding to the tunnel splice joint; wherein the calculation formula is:

[0099]

[0100] Where: x right It is the x-coordinate of the right-side arc and the ray from the origin to the seam position; left It is the x-coordinate of the left arc and the ray from the origin to the seam position; f left (x seam ) represents the ordinate of the intersection point of the fitted left arc with the ray from the origin to the seam position; f right (x seam ) represents the ordinate of the intersection point of the fitted right-side arc with the ray from the origin to the seam position.

[0101] Specifically, in this embodiment of the invention, laser point cloud data acquisition and preprocessing are involved. Data acquisition: Point cloud data acquisition is the foundation for tunnel deformation and defect measurement. The present invention employs a track-mounted inspection vehicle equipped with a Faro lidar as the primary measuring device. During the measurement process, as the inspection vehicle moves forward, the lidar scans along a two-dimensional spiral path to acquire tunnel cross-sectional point cloud data. The acquired data is relative measurement data in the scanner coordinate system, eliminating the need for absolute coordinate transformation. Since the point cloud acquired by the TLSD system inherently contains intensity information, tunnel cross-sectional images can be generated based on this information.

[0102] Data Preprocessing: After acquiring the raw point cloud, preprocessing is necessary to ensure data accuracy and usability. This invention employs the K-means algorithm and distance thresholding for point cloud denoising. K-means improves the accuracy and recognizability of orthophotos by removing outliers and erroneous points; the distance thresholding method removes outliers exceeding a threshold by analyzing the distance from cross-section points to the origin, and also removes outliers that are too close to the tunnel surface.

[0103] Point cloud calibration algorithm: To ensure data accuracy, the coordinate system of the tunnel point cloud must first be determined. Each lap of the LiDAR scan generates a cross-section, with the forward direction as the Y-axis, and the X and Z axes forming a plane. The Y value represents the mileage. The Y value of the point cloud is assigned by the carrier's velocity, but due to track irregularities and carrier swaying, cumulative errors occur, affecting the accuracy of the grayscale image. Therefore, longitudinal calibration is necessary. Circumferential calibration addresses the issue of irregular cross-section shapes, ensuring the mapping accuracy between pixels and true distances.

[0104] Longitudinal point cloud calibration: For mining tunnels, the sleeper width is used as the quantization benchmark for longitudinal calibration. Longitudinal calibration corrects longitudinal errors in the image based on the relationship between the lidar rotation speed, carrier speed, and the actual distance between sleepers, ensuring accurate mileage information for each pixel. The calibration steps are as follows:

[0105] Standard scan line count calculation: Given a preset lidar rotation speed f and known speed v and sleeper spacing w, calculate the standard scan line count n.

[0106] n=(w / v)*f(1)

[0107] The number of scan lines is standardized by uniformly determining the number of scan lines between odometer markers based on the standard scan line count n. If the actual number of scan lines is less than the standard value, linear interpolation is performed using the first five sections to make up the difference; if it is more than the standard value, the extra sections are removed to ensure a consistent number of scan lines. This process ultimately generates uniformly distributed scan lines, thus completing the longitudinal calibration.

[0108] In longitudinal calibration, the sleeper spacing w is obtained through on-site measurement or design drawings, with a standard of 1667 sleepers per kilometer, a spacing of 600mm, and an allowable error of ±20mm. Annotation software is used to mark every 5 sleepers, ultimately generating a grayscale matrix of 3m / frame for calibration.

[0109] Tunnel circumferential point cloud calibration: By fitting polygons to approximate the true contour of the tunnel cross section, accurate distance information is obtained, and the mapping between pixels and the true distance is realized through projection center transformation and fixed distance equal division method, finally generating a distortion-free orthophoto grayscale image.

[0110] Polygon Fitting: To fit the shape of the tunnel cross-section and control the number of sides, the present invention employs a piecewise linear fitting algorithm. This algorithm connects points in the cross-section into multiple straight lines (linear sides), and adjusts the number of sides of the polygon by controlling the number of fitted segments, ensuring that the polygon can both approximate the cross-section shape and meet the accuracy requirements.

[0111] Algorithm steps:

[0112] 1. Set the number of sides: Determine the number of sides of the polygon based on the required polygon accuracy or image resolution. Fewer sides simplify the fitting process but may lead to larger shape approximation errors, while more sides improve fitting accuracy.

[0113] 2. Segmented Fitting: The cross-sectional point cloud is divided into several segments, and each segment is linearly fitted using the least squares method. The start and end points of each line segment are determined by the order of the point cloud and the number of segments set.

[0114] 3. Connecting points: Connect the endpoints of each fitted segment to form the boundary of a polygon. This polygon will approximate the contour of the tunnel cross-section.

[0115] By adjusting the number of segments (edges) used in the fitting process, the complexity and accuracy of the fitted polygon can be controlled. A higher number of segments will generate more edges, enabling a more accurate fitting of complex shapes; a lower number of segments simplifies the fitting process and is suitable for relatively regular cross-sections.

[0116] Advantages: This method allows for flexible control of the number of sides of the polygon according to user needs, adapting to tunnel cross-sections of different shapes. Through simple piecewise fitting, it ensures both high computational efficiency and high fitting accuracy.

[0117] Projection Center Conversion: Since the scanning center of the LiDAR is not located at the center of the tunnel cross-section, the image will suffer from projection distortion, affecting the grayscale image quality and subsequent area calculation. Therefore, it is necessary to convert the projection center from the LiDAR center to the cross-section center. The centroid of the fitted polygon is used as the cross-section center, and projection correction is performed.

[0118] The formula for calculating the geometric center is:

[0119]

[0120] Where N is the number of vertices in the polygon, and is the coordinate of each vertex. This method accurately corrects the projection center, making the generated tunnel image closer to the orthographic projection, further improving image quality, and providing a reference for subsequent interval division processing. i ,y i ) accurate benchmark.

[0121] Correction of Cross-Section Point Clouds Using Fixed-Distance Equal Division: Circumferential projection correction of shield tunnels typically employs the central angle of a fitted circle for cross-section correction. However, to make the algorithm applicable to various cross-section shapes, the traditional circular fitting method for shield tunnels is not suitable. This invention proposes a fixed-distance equal division algorithm, which involves dividing the fitted polygon into equal parts at fixed intervals. Its main principle is to calculate the centroidal angle of the fitted polygon corresponding to the equal division distance, and then further correct the cross-section point cloud.

[0122] Equal division distance of the fitted polygon: First, the total length S of the fitted polygon is obtained. Then, according to the actual project requirements, the actual distance represented by each pixel is set. In this invention, this parameter is denoted as d (d should be less than the total number of cross-sectional points). Therefore, the height of the corrected image can be obtained. (That is, the number of rows in the image matrix). After determining the image height, the fitted polygon is divided into equal parts at fixed intervals. As shown in the figure below, S1 and S2 are two sides of the fitted polygon, and S1 is the starting side for calculating the equal parts at fixed intervals. The length of S1 can be obtained from (X0, Y0) and (X1, Y1), and the length of S2 can be obtained from (X1, Y1) and (X2, Y2). Once S1 is known, the division of side S1 into equal parts can be calculated. The remaining length Δd1 of edge S1 at point S1 is calculated as shown in equation (5):

[0123]

[0124] Therefore, when calculating the first equally spaced point of side S2, Δd1 must be taken into account, and Δd1 + Δd2 = Δd must always be guaranteed. Similarly, all sides of the fitted polygon are equally spaced to obtain line segments or polylines of equal length.

[0125] The coordinates and centroidal angles of the points at fixed intervals can be calculated using trigonometric formulas, given that Δd, Δd1, Δd2, (X0, Y0), (X1, Y1) are all known. n y n The specific formula for calculation will not be elaborated here. Taking (x0, y0) and (x1, y1) in the figure as an example, we calculate the distances a and b from (x0, y0) and (x1, y1) to the centroid of the fitted polygon. Then, we calculate the centroid angle using the cosine theorem formula. Similarly, we can obtain The calculation formulas are as follows: (6), (7), (8):

[0126]

[0127] Through the above calculation steps, the fitted polygons are divided into equal parts at specified distances, and the centroid angles are calculated.

[0128] Pixel Correction: Next, the pixels are corrected. In order to improve the calculation efficiency of the algorithm in the code, the centroid angle and the original cross-section point of the lidar are normalized before calculation. P is defined as the sum of all centroid angles, and it is assumed that the total number of pixels in the circumferential direction after correction is n (i.e., image height H). The calculation method of P is as shown in equations (9) and (10).

[0129]

[0130] Therefore, we have:

[0131]

[0132] Suppose the total number of points in a certain cross-section of the original point cloud of the lidar is S, and the points in the cross-section are denoted as Q. Similarly, the normalization method for Q is as shown in Equation (11):

[0133]

[0134] Traverse Q1 to Qs between P1 and P2. If Q1 ≥ P1 and Q1 < P2, store the gray value of Q1. If not satisfied, continue to judge Q2 until Qs is traversed; next, traverse Q1 to Qs in the interval from P2 to P3. If Q1 ≥ P2 and Q1 < P3, store the gray value of Ql. If not satisfied, continue to judge Q2 until Qs is traversed, and so on. After calculation by the algorithm of the present invention, two situations will occur. The first situation is that only the gray value of one point in the interval is stored, and the gray value of this point is used as the corrected pixel gray value; the second situation is that the gray values of multiple points in the interval are stored, then calculate the average gray value of these points as the corrected pixel gray value. Traverse Q1 to QS until the interval from P(n - 1) to Pn, and the retrieval ends. Calculate each cross-section, and finally obtain a corrected tunnel point cloud gray map with a fixed height.

[0135] Establish the mapping relationship between pixels and real point cloud coordinates: During the generation of tunnel images, each image pixel point corresponds to a real point cloud coordinate. Therefore, each pixel in the image not only carries a gray value (usually representing the intensity information of the point cloud), but also is directly mapped to the actual three-dimensional coordinates. To reverse-index the point cloud data, it is necessary to establish the mapping relationship between pixel coordinates and point cloud coordinates to ensure that the corresponding three-dimensional point cloud coordinates can be accurately deduced from the image.

[0136] The mapping of pixels and point cloud coordinates. During the acquisition of point cloud data, the three-dimensional coordinates (x, y, z) of each point cloud data correspond to a pixel in the image. The gray value of each pixel point is usually generated by the intensity information or reflectivity of the point cloud. Therefore, each pixel in the image already embeds the corresponding spatial coordinate information. Therefore, reverse-indexing the point cloud data from the pixels in the image is actually to establish the mapping relationship through these pixel coordinates and the corresponding point cloud coordinates. Suppose a certain pixel (u, v) in the image corresponds to a point (x, y, z) in the point cloud data, then the coordinates of this pixel and the point cloud data are in one-to-one correspondence. The relationship between each point cloud coordinate (x, y, z) and the image pixel (u, v) is directly mapped through the image generated by the intensity information of the point cloud.

[0137] Multi-point interval mapping processing is necessary because some pixel regions in an image may correspond to multiple point cloud data points. Therefore, in addition to averaging the pixel grayscale values ​​within these intervals, it is also necessary to average the point cloud coordinates within these regions to ensure that the coordinate information of each pixel region can accurately reflect the spatial location of the corresponding point cloud data.

[0138] Suppose there are N point cloud data points within a certain interval, each corresponding to N point cloud coordinates (x, y, y). i ,y i ,z i ) and the corresponding grayscale value I i Then the average point cloud coordinates and average gray value within the interval are respectively:

[0139]

[0140] Among them, (x i ,y i ,z i ) is the data coordinate I of the i-th point cloud. i It is the grayscale value of the i-th point cloud. It is the average coordinate within this interval. It is the average gray value within that range. Using this method, the coordinates and gray values ​​within a pixel range can be accurately calculated, avoiding errors caused by multiple point clouds corresponding to a single pixel.

[0141] Reverse indexing of point cloud data: Once the average coordinates and average grayscale value of each pixel region are calculated, a reverse index between pixels and point cloud data can be achieved. For each pixel, the grayscale value I or... and the corresponding coordinates (x) i ,y i ,z i )or It can accurately index the 3D point cloud coordinates corresponding to the pixel.

[0142] Improved U-Net Network Structure Design: An improved U-Net network structure based on a Hybrid Convolution Module (HCM) is presented. To enhance the detection accuracy of tunnel segment seams in tunnel images, the traditional U-Net network was optimized, particularly in feature extraction and image segmentation capabilities. The improved network, by introducing a Hybrid Convolution Module (HCM), effectively enhances its ability to capture features from different directions, thereby improving the detection accuracy of tunnel segment seam regions. The entire network structure includes an encoder, a Hybrid Convolution Module, a decoder, and an output layer. These components are carefully designed to work collaboratively, enabling the model to perform tunnel segment seam detection more efficiently and accurately.

[0143] Encoder Section: In traditional U-Net networks, the encoder section is primarily used to extract features from the input image layer by layer. This process typically involves multiple convolutional and pooling layers. Convolutional operations at each layer extract features, while pooling operations progressively abstract image information by reducing spatial resolution. However, the limitation of traditional convolutional operations is that they can usually only capture texture information from a fixed direction. Especially in specific tasks, local texture and directional features in the image are often overlooked. For example, in the task of detecting pipe segment seams, the seam region often has obvious directionality, and local texture features are crucial.

[0144] To address the limitations of traditional convolutional algorithms in handling directional and detailed information, a Hybrid Convolution Module (HCM) is proposed and introduced into each layer of the encoder. This module combines horizontal, vertical, and diagonal convolutions to capture image features from multiple directions. The advantage of this approach is that it simultaneously considers texture information in the horizontal, vertical, and diagonal directions, rather than being limited to the single direction that standard convolutional kernels can handle. Specifically, HCM obtains three distinct feature maps by performing convolution operations in the horizontal, vertical, and diagonal directions. Each convolution operation extracts detailed features in that direction, thus comprehensively capturing multi-angle local information in the image, especially significantly improving the network's detection capability for areas with significant directional seams, such as pipe segment splicing seams.

[0145] For a given input feature map Where H, W, and C represent the height, width, and number of channels of the image, respectively. The operation of the hybrid convolution module can be represented as:

[0146] X H =Conv H (X),X V =Conv V (X),X D =Conv D (X) (13)

[0147] Among them, X H The feature map after performing a convolution operation in the horizontal direction, X V The feature map after convolution in the vertical direction, X D Conv is the feature map after convolution operation in the diagonal direction. H and Conv V and Conv DThese represent convolution operations performed in the horizontal, vertical, and diagonal directions, respectively. The output feature maps of the three operations are merged through a concatenation operation and then fused through a 1×1 convolutional layer to generate the final feature map.

[0148] X fused =Conv 1×1 (X H ||X V ||X D (14)

[0149] X fused The introduction of this fused feature map structure enables the network to extract useful features in multiple directions, particularly in the detection of pipe segment splicing seams. It better captures the directional features of the seams, improving the network's robustness and accuracy. The innovation of this method lies in introducing multi-directional convolutional operations, overcoming the limitation of traditional convolutional methods that rely on only a single direction. This allows the network to extract richer features from multiple directions. This multi-angle feature capture is especially important for detection tasks, particularly in images with strong directional and textural features, such as pipe segment splicing seams. Compared to traditional convolutional networks, HCM-based networks demonstrate better performance and accuracy when handling similar problems.

[0150] Decoder Section: The decoder section of the U-Net network is primarily responsible for restoring the spatial resolution of the image and fusing the feature maps generated by the encoder with the decoder output through skip connections. Traditional decoder structures typically restore the spatial resolution of the image through layer-by-layer upsampling operations and help recover image details by stitching them together with feature maps from the encoder. However, with increasing network depth, especially when processing complex tunnel images, detailed information may be lost during the decoding process. Particularly in segment seam detection tasks, subtle features such as seam edges are crucial; the loss of any detail can affect the accuracy of the detection results.

[0151] To enhance the decoder's ability to recover image details, a Hybrid Convolutional Module (HCM) is introduced in each decoder layer. Similar to the encoder, the HCM in the decoder further enhances the network's ability to capture complex textures and details, especially in tasks such as pipe segment seam detection, where detailed features often have a decisive impact on the final detection accuracy. Through this design, the network can not only recover the spatial resolution of the image but also ensure that fine-grained information in the image is preserved as much as possible during the decoding process.

[0152] During the decoder's operation, the spatial resolution of the feature map is first restored through upsampling. Next, the decoder concatenates the upsampled feature map with the skip connection feature map from the encoder and further processes it through convolution. This process ensures that the decoder can fully utilize the rich features from the encoder while recovering higher-precision image details. To enhance feature fusion in this process, an HCM module is introduced at each layer of the decoder. After upsampling, the hybrid convolution module extracts detailed information from the feature map from the encoder again and combines it with the currently upsampled feature map. Figure 1 The images are then fused together. This method allows the network to better preserve fine-grained information in the images, especially key details such as the joints between tunnel segments, thus improving detection accuracy.

[0153] X up =Upsample(X),X decode =Conv 3×3 (X up ||X skip (15)

[0154] Among them, X up X is the upsampled feature map. skip X is the encoder feature map passed from the skip connection. decode This is the feature map after the convolution operation in the decoder, where Conv3×3 represents the convolution operation. This design allows the network to better preserve fine-grained information in the image, especially in key detail areas such as the edges of the pipe segment seams.

[0155] Output Layer: The output layer is the final part of the network, and its task is to generate the probability of whether each pixel belongs to the pipe segment splice seam region. In the traditional U-Net, the output layer usually compresses the number of channels into a binary classification output through a 1×1 convolution. In the scheme of this invention, the output layer still uses a 1×1 convolution operation, but on this basis, a sigmoid activation function is used to map the output value to the [0,1] interval, representing the probability of each pixel belonging to the pipe segment splice seam region. The operation of the output layer can be expressed as:

[0156] Y = Sigmoid(Conv) 1×1 (X decode ))(16)

[0157] Y is the output, representing the probability that each pixel belongs to the segment seam region. This design effectively converts the network output into an actual binary classification result, thus providing accurate probability values ​​for the detection of segment seam regions. This decoder design aims to maximize image information restoration through multi-level feature fusion and detail recovery, and enhances the network's sensitivity to directional texture features through hybrid convolutional modules. Especially when processing complex scenes such as tunnel images, this structure effectively improves network performance, ensuring that subtle features such as segment seams are fully preserved, thereby improving detection accuracy and robustness.

[0158] Loss Function and Optimization: This chapter introduces the model's loss function and optimization methods. In the task of detecting pipe segment joints, the class imbalance between the pipe segment joint region and the background region is a key factor affecting detection accuracy. Common loss functions such as mean squared error (MSE) and cross-entropy loss are insufficient for the task of pipe segment joint detection. To improve the model's performance under this imbalance condition, the proposed solution combines Dice loss and cross-entropy loss. Dice loss effectively solves the class imbalance problem, while cross-entropy loss helps improve the network's classification accuracy for pipe segment joints and background regions. The hybrid loss function takes the following form:

[0159] L=λ1·L Dice +λ2·L CE (17)

[0160] Among them, L Dice L represents Dice loss. CE Let λ1 and λ2 represent the cross-entropy loss, and λ1 and λ2 are the weighting coefficients of the loss term, which are designed to balance the contributions of the two loss functions.

[0161] To optimize the network training process, this invention employs the Adam optimizer. The update rules of the Adam optimizer are as follows:

[0162]

[0163] Where, θ t Let η represent the parameters at the t-th iteration, η be the learning rate, and m be the number of iterations. t and v t These are the first and second moment estimates of the gradient, respectively, and ò is a small constant used to prevent division by zero errors. The Adam optimizer can automatically adjust the learning rate to ensure stability and convergence speed during training.

[0164] The U-Net network, based on an improved Hybrid Convolutional Module (HCM), significantly enhances its ability to extract complex image features by introducing multi-directional convolutional operations. Compared to the traditional U-Net network, the improved structure can better handle texture information from different directions in the tunnel segment splice seam detection task, and improves detection accuracy by enhancing the decoder's detail recovery capability. By using a hybrid loss function and the Adam optimizer, the network can be effectively trained on imbalanced data, ultimately achieving efficient detection of tunnel segment splice seams in tunnel images.

[0165] For misalignment calculation, after identifying the segment splicing seams using an improved U-Net algorithm, the cross-sectional point clouds on both sides of the seam location are indexed through pixel-to-point cloud mapping. To calculate the misalignment of the seam, this invention proposes using the RANSAC algorithm to fit the cross-sections on both sides of the seam. By fitting two arcs and calculating the distances they are offset from each other on the rays from the origin to the seam location, the misalignment of the seam is obtained. The specific steps are as follows:

[0166] RANSAC algorithm for fitting cross-sections on both sides of a seam: The two sides of a seam usually exhibit a certain curved shape, so the RANSAC (Random Sample Consensus) algorithm can be used to fit the cross-sections on both sides of the seam. RANSAC is a robust fitting method that can effectively fit a model from noisy point clouds. It fits a model by randomly selecting a set of sample data and determines whether other data points are inliers by calculating the distance between them and the model. The advantage of this algorithm is its ability to remove outliers, ensuring the accuracy of the fitting results.

[0167] For the segment splicing seam, the point cloud data of the left and right sides of the seam needs to be extracted first. Assuming the point cloud data of the left side of the seam is Pleft and the point cloud data of the right side is Pright, the RANSAC algorithm will be used to fit these two sets of data respectively, obtaining the fitted arcs for the left and right sides. The RANSAC fitting process is as follows:

[0168] Randomly select sample points: Randomly select the minimum number of points from the point cloud data on each side to fit the arc model.

[0169] Calculate the fitting error: For all other points, calculate their distances to the fitted model and determine which points are interior points.

[0170] Optimize the fitting model: Refit the model based on the coordinates of the interior points, and select the best fitting model through multiple iterations.

[0171] The left and right fitted curves obtained by RANSAC fitting can be represented by parametric equations. The equation for fitting the curves into arcs is:

[0172] (x-x0) 2 +(y-y0)2 =r 2 (19)

[0173] Where (x0, y0) are the coordinates of the center of the circle, and r is the radius of the circle.

[0174] The formula for calculating misalignment is as follows: Misalignment Δy is obtained by calculating the distance between the two points where the arcs on both sides of the joint intersect with the rays from the origin to the joint position. Let the joint position be at x... seam At the point where the fitted left and right arcs intersect with the ray from the origin to the seam, the ordinates of the two points are f, respectively. left (x seam ) and f right (x seam Therefore, the formula for calculating misalignment can be simplified to the horizontal distance between two points, that is:

[0175]

[0176] Where: x right and x left These are the abscissas of the right and left arcs and the rays from the origin to the seam. This formula can be used to calculate the misalignment of the arcs on both sides of the seam at the seam. When calculating the misalignment, deviations may occur due to uneven point cloud density or fitting errors. Therefore, the calculated misalignment needs to be optimized. This error can be reduced by performing multiple fittings and calculating their average.

[0177] To achieve the above objectives, the present invention also provides a tunnel splice joint misalignment detection system based on lidar imagery, such as... Figure 5 As shown, the system is used to implement the tunnel splice joint misalignment detection method based on lidar imagery, and the system specifically includes:

[0178] The first data generation unit is used to generate and acquire first data corresponding to the tunnel inner wall based on lidar, and to perform noise reduction preprocessing on the first data and generate corresponding second data; wherein, the first data is three-dimensional point cloud data; and the second data is three-dimensional point cloud data after noise reduction preprocessing.

[0179] A first data processing unit is used to calibrate the second data and generate third data corresponding to the second data; wherein the calibration process includes longitudinal calibration and circumferential calibration; and the third data is distortion-free orthophoto grayscale image data.

[0180] The second data processing unit is used to construct an improved U-Net model and, based on the improved U-Net model, identify and generate fourth data corresponding to the third data; wherein, the fourth data is segment splicing seam area data;

[0181] The second data generation unit is used to create a pixel-to-point cloud coordinate mapping relationship based on tunnel image data and extract and generate the corresponding fifth data; and to calculate and generate the sixth data corresponding to the tunnel splicing joint by fitting the cross-sectional curves on both sides of the joint according to the RANSAC algorithm; wherein, the fifth data is the cross-sectional point cloud data on both sides of the joint; and the sixth data is the misalignment amount data.

[0182] The first data generation unit further includes:

[0183] The first generation module is used for two-dimensional spiral path scanning based on lidar to generate and acquire first data corresponding to the tunnel cross-section;

[0184] The first processing module is used to combine the K-means algorithm to remove outlier and erroneous data in the first data, and to filter out noise data in the first data using the distance threshold method.

[0185] And / or, the first data processing unit further includes:

[0186] The second generation module is used to generate and acquire the seventh data corresponding to the lidar, and calculate and generate the corresponding eighth data based on the seventh data; wherein, the seventh data is lidar status data, including lidar rotation speed data, carrier speed data and sleeper spacing data; the eighth data is scan line count data;

[0187] The second processing module is used to adjust the real-time scan line count data based on the eighth data, combined with linear interpolation or section removal.

[0188] And / or, the second data processing unit further includes:

[0189] A generation module is constructed to build a hybrid convolution module and, based on the hybrid convolution module, extract and generate tenth data corresponding to the tunnel image; wherein, the tenth data is image texture feature data, including: image texture feature data in the horizontal direction, vertical direction and diagonal direction;

[0190] The third processing module is used to fuse encoder feature data through upsampling and skip connections, and to enhance detailed data information using the hybrid convolution module.

[0191] The third generation module is used to generate eleventh data corresponding to image pixels and belonging to the stitching seam region based on the Sigmoid activation function; wherein, the eleventh data is probability data;

[0192] And / or, the second data generation unit further includes:

[0193] The fourth processing module is used to average the point cloud coordinates corresponding to each pixel and locate the original point cloud data from the image pixels through reverse indexing.

[0194] The fitting generation module is used to randomly sample point cloud data on both sides of the seam, fit an arc model and filter the inner points, and generate a twelfth data corresponding to both sides of the seam through iterative optimization; wherein, the twelfth data is the optimal fitted curve data.

[0195] The fourth generation module is used to calculate and generate the sixth data corresponding to the tunnel splice joint based on the twelfth data; wherein the calculation formula is:

[0196]

[0197] Where: x right It is the x-coordinate of the right-side arc and the ray from the origin to the seam position; left It is the x-coordinate of the left arc and the ray from the origin to the seam position; f left (x seam ) represents the ordinate of the intersection point of the fitted left arc with the ray from the origin to the seam position; f right (x seam ) represents the ordinate of the intersection point of the fitted right-side arc with the ray from the origin to the seam position.

[0198] The first data processing unit further includes:

[0199] The fifth generation module is used to generate the ninth data corresponding to the tunnel cross-section based on a piecewise linear fitting algorithm; wherein, the ninth data is fitted polygon data;

[0200] The fifth processing module is used to perform distortion elimination processing on the ninth data and to perform circumferential correction processing on the ninth data using the fixed-distance equal division method to generate the corresponding third data.

[0201] In the system solution embodiment of the present invention, the specific details of the method steps involved in the detection of misalignment of tunnel splice joints based on lidar images have been described above. That is to say, the functional modules in the system are used to implement the steps or sub-steps in the above method embodiment, which will not be repeated here.

[0202] To achieve the above objectives, the present invention also provides a tunnel splice joint misalignment detection platform based on lidar imagery, such as... Figure 6As shown, the system includes a processor, a memory, and a control program for a tunnel splice joint misalignment detection platform based on LiDAR imagery. The processor executes the control program, which is stored in the memory. This control program implements the steps of the LiDAR imagery-based tunnel splice joint misalignment detection method. The specific details of these steps have been described above and will not be repeated here.

[0203] In this embodiment of the invention, the tunnel splice joint misalignment detection platform based on lidar imagery has a built-in processor, which can be composed of integrated circuits. For example, it can be composed of a single packaged integrated circuit, or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor connects to various components using various interfaces and lines, and executes programs or units stored in the memory, as well as calling data stored in the memory, to perform various functions of tunnel splice joint misalignment detection based on lidar imagery and process data. The memory, used to store program code and various data, is installed in the tunnel splice joint misalignment detection platform based on lidar imagery and enables high-speed, automatic access to programs or data during operation. The memory includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0204] This invention utilizes a method based on lidar to generate and acquire first data corresponding to the tunnel inner wall, preprocesses the first data with noise reduction, and generates corresponding second data. The first data is 3D point cloud data; the second data is 3D point cloud data after noise reduction preprocessing. The second data is then calibrated to generate corresponding third data, including longitudinal and circumferential calibration. The third data is distortion-free orthophoto grayscale image data. An improved U-Net model is constructed, and a fourth data corresponding to the third data is generated based on the improved U-Net model. The fourth data is segment splicing joint area data. Based on the tunnel image data, a pixel-to-point cloud coordinate mapping relationship is created, and corresponding fifth data is extracted and generated. Finally, the cross-sectional curves on both sides of the joint are fitted using the RANSAC algorithm to calculate and generate sixth data corresponding to the tunnel splicing joint. The fifth data is cross-sectional point cloud data on both sides of the joint; the sixth data is misalignment data. The invention also includes a corresponding system and platform, which effectively improves the accuracy and efficiency of joint detection, providing technical support for tunnel structural health monitoring.

[0205] In other words, this invention utilizes the improved U-Net's efficient target detection capabilities to accurately identify joint locations in tunnel images. It then combines this with the RANSAC algorithm to fit the cross-sectional curves on both sides of the joint, calculating the misalignment. This invention's method demonstrates high accuracy and robustness in segment splicing joint detection and misalignment calculation. Especially in complex tunnel cross-sections and irregularly shaped joints, the RANSAC fitting algorithm effectively handles noise points and anomalies, providing accurate misalignment calculation results. In summary, this invention provides a novel tunnel deformation monitoring method that can strongly support tunnel structural assessment, maintenance decisions, and safety inspection. In the future, with further development of lidar technology and optimization of deep learning models, the accuracy and applicability of related methods will be further improved, and they are expected to be widely applied in tunnel engineering, infrastructure maintenance, and intelligent transportation systems.

[0206] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for detecting misalignment of a tunnel splicing joint based on a laser radar image, characterized by, The method comprises the steps of: Based on the laser radar, the first data corresponding to the inner wall of the tunnel is generated and acquired, and the first data is denoising preprocessed to generate the corresponding second data; wherein the first data is three-dimensional point cloud data; the second data is denoising preprocessed three-dimensional point cloud data; The second data is calibrated and the third data corresponding to the second data is generated; wherein the calibration process includes longitudinal calibration and ring calibration; the third data is a non-distorted orthographic gray image data; An improved U-Net model is constructed, and the fourth data corresponding to the third data is identified and generated based on the improved U-Net model; wherein the fourth data is a pipe piece joint area data; further comprising: constructing a hybrid convolution module, and based on the hybrid convolution module, the tenth data corresponding to the tunnel image is extracted and generated; the encoder feature data is fused by upsampling and jump connection, and the hybrid convolution module is used to enhance the detail data information; based on the Sigmoid activation function, the eleventh data corresponding to the image pixel and belonging to the joint area is generated; the tenth data is image texture feature data, including: horizontal direction, vertical direction and diagonal direction image texture feature data; the eleventh data is probability data; Based on the tunnel image data, the pixel and point cloud coordinate mapping relationship is created, and the fifth data corresponding thereto is extracted and generated; and according to the RANSAC algorithm, the cross section curve on both sides of the joint is fitted and processed, and the sixth data corresponding to the tunnel joint is calculated and generated; wherein the fifth data is the cross section point cloud data on both sides of the joint; the sixth data is the amount of error data; further comprising: average processing the point cloud coordinates corresponding to each pixel, and locating the original point cloud data from the image pixel through reverse indexing; randomly sampling the point cloud data on both sides of the joint, fitting the arc line model and screening the inner points, and generating the twelfth data corresponding to both sides of the joint through iterative optimization; the twelfth data is the optimal fitting curve data; according to the twelfth data, the sixth data corresponding to the tunnel joint is calculated and generated; wherein the calculation formula is: (20), wherein: is the x-coordinate of the right side arc and the ray from the origin to the seam location; is the x-coordinate of the left side arc and the ray from the origin to the seam location; is the y-coordinate of the intersection of the fitted left side arc and the ray from the origin to the seam location; is the y-coordinate of the intersection of the fitted right side arc and the ray from the origin to the seam location. 2.The tunnel joint misalignment detection method based on laser radar image according to claim 1, characterized in that, The first data corresponding to the inner wall of the tunnel is generated and acquired based on the laser radar, and the first data is denoising preprocessed to generate the corresponding second data, further comprising: Based on the two-dimensional spiral path scanning of the laser radar, the first data corresponding to the tunnel cross section is generated and acquired; The outlier data and error point data in the first data are removed by combining the K-means algorithm, and the noise point data in the first data is filtered by distance threshold method. 3.The tunnel joint misalignment detection method based on laser radar image according to claim 1, characterized in that, The second data is calibrated and the third data corresponding to the second data is generated, further comprising: The seventh data corresponding to the laser radar is generated and acquired, and the eighth data is calculated and generated based on the seventh data; wherein the seventh data is laser radar state data, including laser radar rotating speed data, carrier speed data and sleeper spacing data; the eighth data is scanning line number data; Based on the eighth data, the real-time scanning line number data is adjusted by combining linear interpolation or cross section removal.

4. The tunnel joint misalignment detection method based on laser radar image according to claim 1 or 3, characterized in that, The calibration processing the second data and generating third data corresponding to the second data further comprises: generating ninth data corresponding to the tunnel section based on a piecewise linear fitting algorithm; wherein the ninth data is fitting polygon data; the distortion elimination processing the ninth data, and generating corresponding third data through equidistant ring correction processing the ninth data. 5.A system for detecting misalignment of a tunnel splicing joint based on a laser radar image, characterized by The system is used to implement the tunnel joint misalignment detection method based on laser radar image as claimed in any one of claims 1 to 4, and the system comprises: a first data generation unit configured to generate and acquire first data corresponding to the tunnel inner wall based on a laser radar, and to denoise and preprocess the first data and generate corresponding second data; wherein the first data is three-dimensional point cloud data; and the second data is denoised and preprocessed three-dimensional point cloud data; a first data processing unit configured to calibrate the second data and generate third data corresponding to the second data; wherein the calibration processing comprises longitudinal calibration processing and ring calibration processing; and the third data is distortion-free orthographic gray image data; a second data processing unit configured to build an improved U-Net model and identify and generate fourth data corresponding to the third data based on the improved U-Net model; wherein the fourth data is pipe piece joint area data; a second data generation unit configured to create a pixel and point cloud coordinate mapping relationship based on tunnel image data, and to extract and generate corresponding fifth data; and to fit and process the section curves on both sides of the joint according to a RANSAC algorithm, and to calculate and generate sixth data corresponding to the tunnel joint; wherein the fifth data is section point cloud data on both sides of the joint; and the sixth data is misalignment amount data.

6. The system for detecting misalignment of a tunnel splice joint based on a laser radar image according to claim 5, wherein The first data generation unit further comprises: a first generation module configured to generate and acquire first data corresponding to the tunnel section based on two-dimensional spiral path scanning of the laser radar; a first processing module configured to remove outlier data and error point data in the first data in combination with a K-means algorithm, and to filter and process noise point data in the first data through a distance threshold method; and / or, the first data processing unit further comprises: a second generation module configured to generate and acquire seventh data corresponding to the laser radar, and to calculate and generate corresponding eighth data based on the seventh data; wherein the seventh data is laser radar state data, including laser radar rotation speed data, carrier speed data and sleeper spacing data; and the eighth data is scanning line number data; a second processing module configured to adjust and process real-time scanning line number data in combination with linear interpolation or section removal based on the eighth data; and / or, the second data processing unit further comprises: a construction generation module configured to build a hybrid convolution module, and to extract and generate tenth data corresponding to the tunnel image based on the hybrid convolution module; wherein the tenth data is image texture feature data, including image texture feature data in horizontal, vertical and diagonal directions; The third processing module is configured to fuse the encoder feature data through upsampling and skip connection processing, and enhance the detail data information by using the hybrid convolution module; The third generation module is configured to generate, based on a Sigmoid activation function, eleventh data corresponding to image pixels and belonging to a splicing seam region, wherein the eleventh data is probability data. And / or, the second data generation unit further includes: The fourth processing module is configured to average process the point cloud coordinates corresponding to each pixel, and locate the original point cloud data from the image pixels through reverse indexing; The fitting generation module is configured to randomly sample the point cloud data on both sides of the joint, fit an arc line model, filter inliers, and generate twelfth data corresponding to both sides of the joint through iterative optimization, wherein the twelfth data is optimal fitting curve data. The fourth generation module is configured to calculate and generate, according to the twelfth data, sixth data corresponding to the tunnel splicing seam, wherein the calculation formula is: (20), wherein: is the x-coordinate of the right side arc and the ray from the origin to the seam location; is the x-coordinate of the left side arc and the ray from the origin to the seam location; is the y-coordinate of the intersection of the fitted left side arc and the ray from the origin to the seam location; is the y-coordinate of the intersection of the fitted right side arc and the ray from the origin to the seam location.

7. The system according to claim 5 or 6, wherein the system is characterized by, The first data processing unit further includes: The fifth generation module is configured to generate, based on a piecewise linear fitting algorithm, ninth data corresponding to a tunnel section, wherein the ninth data is fitting polygon data. The fifth processing module is configured to perform distortion elimination processing on the ninth data, and perform ring correction processing on the ninth data by distance equalization, to generate corresponding third data. 8.A platform for detecting misalignment of a tunnel splicing joint based on a laser radar image, characterized in that, The tunnel splicing seam dislocation detection platform control program based on laser radar image includes a processor, a memory, and a tunnel splicing seam dislocation detection platform control program based on laser radar image; wherein the tunnel splicing seam dislocation detection platform control program based on laser radar image is stored in the memory, and the tunnel splicing seam dislocation detection platform control program based on laser radar image is executed by the processor; the tunnel splicing seam dislocation detection platform control program based on laser radar image realizes the tunnel splicing seam dislocation detection method based on laser radar image as claimed in any one of claims 1 to 4.

Images