Subway tunnel segment staggering analysis method for laser scanning rail transit vehicle

Abstract

According to the laser scanning rail transit vehicle subway tunnel segment staggering analysis method, a three-dimensional laser scanner and a rail transit vehicle are combined, and a shield tunnel is scanned in a spiral line mode; in the scanning process, the scanner can be operated more flexibly, and the scanning process is prevented from being influenced; according to the method, point cloud data obtained through scanning is used for carrying out mileage correction to obtain the time relation between the mileage and the position of a rail car, a high-definition orthographic image graph of the shield tunnel lining surface is generated, and then based on data dynamic scheduling of a quadtree micro-hierarchy model, rapid interactive browsing of shield tunnel lining surface images is achieved; the method is advantaged in that segment boundary extraction is carried out on the orthographic image of the tunnel lining surface, fitting denoising is carried out on section point clouds, then projection to the same plane is carried out, comparison is carried out, size and quantity of slab staggering are acquired, detection precision is high, the method is real and reliable, and slab staggering detection precision requirements are satisfied. Meanwhile, the operation process is simple, rapid and efficient, and labor intensity is reduced.

Description

Technical Field

[0001] The present application relates to a laser rail vehicle tunnel analysis method, and in particular to a laser scanning rail transit vehicle subway tunnel segment misalignment analysis method, belonging to the field of laser scanning tunnel analysis technology. Background Art

[0002] As a new mode of transportation, rail transit systems have significantly alleviated travel stress and are increasingly popular. With the development of subway systems, monitoring the real-time and effective health of subway rail transit has become essential for subway construction. Detecting tunnel segment misalignment is a key component of subway inspections. Traditionally, regular and individual inspections using tape measures and set squares are inefficient and labor-intensive.

[0003] Despite the widespread use of the shield method, problems still arise in actual use in subway tunnels. Segments often misalign and break due to various reasons, such as improper shield machine posture, uneven grouting, and deviations in manufacturing dimensions.

[0004] Segment misalignment refers to the dimensional deviation between adjacent segments in the same ring or between segments in adjacent rings after prefabricated segments are spliced. These deviations are referred to as circumferential and longitudinal misalignment. Segment misalignment has long been overlooked, yet it is a common problem encountered in subway shield tunneling, directly impacting the quality of subway tunnel projects. Segment misalignment can directly generate shear stress in the segments, leading to damage and cracking. This not only seriously affects the aesthetic appearance of the tunnel lining, but can also damage the tunnel structure, reduce the waterproofing capacity between the segments, and cause tunnel leakage, seriously impacting the quality and safety of the subway.

[0005] Segment misalignment can lead to tunnel damage, water seepage, and other problems, posing safety risks to engineering quality and reducing the tunnel's service life and effectiveness. To avoid these safety hazards, timely monitoring of subway tunnel segment misalignment data is beneficial to subway construction and operation. It facilitates the assessment of potential risks such as segment leakage, cracking, and damage, and improves the safety of subway tunnel structures.

[0006] Segment misalignment has always been a common problem in subway construction. The resulting waterproofing and other issues pose significant risks to tunnels, making segment misalignment a key component of subway project inspections. Currently, segment misalignment and cracking are typically detected through individual inspections by inspectors, typically using tape measures and set squares. Inspectors regularly patrol the subway tunnels, detect segment misalignment through observation, and record the amount of misalignment. This method is inefficient, slow, labor-intensive, and has low accuracy.

[0007] The problems that need to be solved in the existing subway tunnel segment misalignment analysis and the key technical difficulties of this application include:

[0008] (1) Pipe segment misalignment is a common problem encountered in subway construction. Waterproofing and other problems caused by misalignment also bring great hidden dangers to the tunnel. However, in subway project inspection, the current inspection of pipe segment misalignment and cracking is generally carried out by inspectors one by one, generally using tape measures and triangles for measurement. Inspectors regularly patrol the inside of the subway tunnel, discover pipe segment misalignment through observation, and record the amount of misalignment after measurement. This method is inefficient, slow, labor-intensive, and has low accuracy. Using a three-dimensional laser rail transit vehicle to scan subway data has the advantages of high scanning efficiency, high speed, and low labor intensity compared to traditional station-based static scanning. However, the existing technology lacks independently developed subway three-dimensional laser scanning rail transit vehicles, lacks the combination of rail transit vehicles with scanners, and lacks sensors such as odometers and high-precision inertial navigation units. It cannot well obtain the posture and position of point cloud data. At the same time, it also lacks a set of efficient laser scanning rail transit vehicle subway tunnel pipe segment misalignment analysis methods.

[0009] (2) The existing technology lacks the development of line scanning data acquisition software. There is no line scanning data acquisition software. The device's built-in WiFi and IP address are not used to connect to the scanner, remotely set the scanner's parameters and modes, create new scanning projects and scan files, or control the scanner's start and stop. Scanning data is not recorded, and scanning data files cannot be saved or exported. Information such as the scanning time, two-dimensional coordinates, and reflection intensity of the scanning point cannot be obtained. The existing technology lacks the structure of a three-dimensional laser scanning subway rail transit vehicle system and lacks a method for performing spiral, full-section, high-density scanning of shield tunnels.

[0010] (3) The existing technology lacks a method for generating stretched images of shield tunnel point cloud data, and does not generate orthophoto images from the scanned point cloud data, making it inconvenient to observe the surface of the shield tunnel. The scanned point cloud data of the tunnel lining surface is not stretched from two-dimensional coordinates to obtain the three-dimensional absolute coordinates of all points, making it impossible to obtain image information of the tunnel lining surface. Since the subway project is long and narrow, the tunnel is long and the cross-section is large, and the image generated by the point cloud data is a rectangular image with a length much greater than the width. At the same time, in order to obtain accurate information, the generated image has high pixels and a large amount of data, resulting in a slow loading process, and it is also inconvenient to drag and observe the enlarged image. The existing technology lacks design software to dynamically schedule images and visualize them, and lacks a method to dynamically schedule tunnel images. The memory usage efficiency is low, and it is inconvenient to read the information on the image.

[0011] (4) The existing technology lacks a segment misalignment analysis method that integrates stretched images and point cloud data. It lacks a method to stretch the acquired point cloud data into the same coordinate system based on the relative position of the rail car and the scanner, as well as the attitude data in the odometer and the tunnel design file, to obtain a point cloud map of the entire tunnel. It lacks a method to obtain the corresponding mileage value through the generated image map of the tunnel surface. The segment misalignment analysis system has not been written to output the misalignment analysis map in DXF format, and it is impossible to calculate the misalignment amount and generate a misalignment analysis table. The characteristics of 3D laser scanning technology such as high efficiency, high precision, large area, no need to contact the target, and no requirements for light are not fully utilized. It is impossible to accurately obtain complete information on the tunnel lining surface, and the measurement efficiency and accuracy are still low. Summary of the Invention

[0012] The traditional method of using tape measures and triangles for regular inspections and one-by-one inspections is inefficient and labor-intensive. This application uses a three-dimensional laser scanning rail transit vehicle, combines a three-dimensional laser scanner with a rail transit vehicle, and uses mobile laser scanning to quickly scan the tunnel, obtain complete information on the tunnel lining surface, greatly improves the efficiency of field measurement, and analyzes the tunnel misalignment based on point cloud data. The three-dimensional laser scanner scans the shield tunnel in a spiral manner; during the scanning process, the scanner can be operated more flexibly to avoid affecting the scanning process; the point cloud data obtained by scanning is used to perform mileage correction to obtain the time relationship between mileage and rail vehicle position, and generate a high-definition orthophoto image of the shield tunnel lining surface. Then, based on the dynamic data scheduling of the quadtree micro-level model, fast interactive browsing of the shield tunnel lining surface image is achieved; the segment boundary line is extracted from the orthophoto image of the tunnel lining surface, the cross-section point cloud is fitted and denoised, and then projected onto the same plane for comparison to obtain the size and number of misalignments; this technology has high detection accuracy, is true and reliable, and can well meet the accuracy requirements of misalignment detection. At the same time, the system's operating process is simple, fast and efficient, which greatly reduces the labor intensity of staff and has great application prospects.

[0013] To achieve the above technical effects, the technical solutions adopted in this application are as follows:

[0014] A laser scanning rail transit vehicle is used to analyze the misalignment of subway tunnel segments. The laser scanning rail transit vehicle combines a 3D laser scanner with a rail transit vehicle and is equipped with an odometer, a built-in computer, and a power supply. It uses a mobile laser to quickly scan the tunnel, obtain complete information on the tunnel lining surface, and analyze the misalignment of the tunnel based on the point cloud data.

[0015] The 3D laser scanner scans the shield tunnel in a spiral pattern. Programs are compiled for the connection, control, and output of the 3D laser scanner, allowing for flexible operation of the scanner during the scanning process. Using the scanned point cloud data, mileage correction is performed to determine the temporal relationship between mileage and railcar position. Projection transformation is performed based on coordinates and reflection intensity values to generate a high-definition orthophoto of the shield tunnel lining surface. Dynamic data scheduling based on the quadtree micro-level model enables rapid interactive browsing of the shield tunnel lining surface image. Segment boundary lines are extracted from the orthophoto of the tunnel lining surface. Cross-section point clouds on both sides are extracted based on the mileage values of the boundary lines. The cross-section point clouds are fitted and denoised, then projected onto the same plane for comparison to determine the size and amount of misalignment.

[0016] a. Generating and processing stretched images of the shield tunnel lining surface: First, a shield tunnel lining surface image is generated. Using the mileage data recorded in the odometer for the corresponding time period, the scanned point cloud data of the tunnel lining surface is stretched from two-dimensional coordinates to three-dimensional absolute coordinates. A geometric transformation is then performed to obtain a projection plane. This is then used to generate a shield tunnel lining surface image through gridding. Using a dynamic data scheduling method based on a quadtree micro-level model, the micro-level model is divided into different subdivision levels. By establishing a quadtree index, images of the appropriate subdivision level are dynamically scheduled based on the distance from the viewpoint to the image, achieving stretched image processing.

[0017] b. Analysis of shield tunnel segment misalignment: First, based on line transformation and image edge detection, the segment boundary line in the image is extracted. Then, the segment cross-section information on both sides of the boundary line is extracted based on the mileage value, and fitting and denoising are performed. Finally, the cross-section points are projected onto a plane for comparison to obtain the segment misalignment. The size and amount of misalignment are calculated to generate a misalignment analysis table to obtain the shield tunnel misalignment situation.

[0018] Preferably, a stretched image is generated: the scanning starting mileage is used as the origin, a three-dimensional coordinate system is established, and the three-dimensional coordinates of the point cloud data relative to the scanning center are calculated. The high-precision and high-density three-dimensional coordinate information of the tunnel surface is obtained, and the contour of the real object is restored by building a model, thereby analyzing the actual situation of the measured tunnel surface;

[0019] 1) Calculate the coordinates of the scanning point: Use the mileage value of the scanning point as the Y coordinate of the scanning point, and restore the 2D point cloud to a 3D point cloud. The 3D point cloud at this time is cylindrical. The scanning center at the start of the scan is used as the origin, and a coordinate system is established with the radial direction as the X axis, the forward direction of the railcar as the Y axis, and the direction perpendicular to the road surface as the Z axis.

[0020] The three-dimensional coordinates of any point P on the tunnel surface are:

[0021] (X P ,Y P,Z P )=(S·SINβ,y P -y0,S·COSβ) Formula 1

[0022] Where S is the scanning distance, β is the scanning altitude angle, and y P is the mileage value of the scanning point, y O is the mileage value of the coordinate origin;

[0023] 2) Correct coordinates based on design files: Design files refer to files generated based on the drawings used during design and construction, and are divided into horizontal curves and vertical curves. The horizontal curve design file contains information such as the line type of each mileage segment, the starting mileage, ending mileage, starting point coordinates, ending point coordinates, starting azimuth, and ending azimuth of the corresponding line type. The vertical curve design file contains the elevation of the slope change point, the slope of the front and back slopes of the slope change point, the mileage of the slope change point, and the radius of the circular curve near the slope change point. Based on the design file, obtain the tunnel design line type at any mileage, find the corresponding line type based on the mileage given by the odometer, and correct the coordinates of the projection point.

[0024] Preferably, 3) establishing a projection plane: generating an orthophoto image of the shield tunnel lining surface, and projecting the cylindrical three-dimensional point cloud data onto a two-dimensional plane; first establishing the projection plane, calculating the coordinates of all projection points under the projection plane, and using the angle values of the three-dimensional laser scanning points, through equiangular tangent projection, converting the tunnel from the original three-dimensional coordinate system to two-dimensional plane coordinates based on geometric relationships;

[0025] A two-dimensional coordinate system is established with the arch position at the start of the tunnel scan as the origin, the railcar forward direction as the m-axis, and the vertical forward direction as the n-axis. The coordinates of the cylindrical point cloud data are converted into a rectangular projection plane, with the arch position n = 0;

[0026] The coordinates of any point P on the plane are:

[0027] (m P , n P )=(Y P ,βπr / 180) Equation 2

[0028] Where Y P is the coordinate of point P in the three-dimensional coordinate system, β is the scanning height angle, and r is the radius of the shield tunnel;

[0029] The point cloud data stretched into a two-dimensional rectangular plane is the basis for generating orthophoto images. A cross section obtained by a 3D laser scanner scanning a circle is stretched into a column of points with the same m coordinates. The length in the m direction is the distance traveled in the scan.

[0030] The rectangular projection plane is the minimum enclosing rectangle of all projection points. Through the conversion formula, the position of any point on the tunnel lining surface can be obtained from its mileage value and the scanning point angle value, providing coordinate conversion for the generation of the image map.

[0031] 4) Establish the image coordinate system: Use the projection plane to establish the image plane coordinate system xoy. The coordinate axis direction is consistent with the projection plane coordinate axis. The lower left corner of the projection rectangle is the coordinate origin. The coordinates of any point P on the plane are:

[0032] (x P ,y P )=(m P , n P -n min ) Formula 3

[0033] Where m P 、n P is the coordinate of the scanning point P in the projection plane, n min is the point with the smallest n coordinate in the projection plane;

[0034] After establishing the image plane coordinate system, all scanning points are converted into two-dimensional coordinate points on the image plane. The image plane is used as the base map for generating the reflection intensity image. The content of the image depends on the scanning point information on each coordinate. The scanning sampling interval of the point cloud data is used as the pixel size (step size) of the image to be generated. The image plane is divided into a grid according to the step size, and each grid point is a pixel point. The width M and height N of the image are calculated as shown in Formula 4:

[0035] M=ceil(m max / delta)+1

[0036] N=ceil((n max -n min ) / delta)+1 Formula 4

[0037] In the formula, ceil() means rounding; delta means the sampling interval of the point cloud;

[0038] The image plane coordinate system is divided into grid points according to pixel size, and the projection plane is also divided into corresponding grids. All scanning points fall into the corresponding grids according to their positions. The scanning parameters set by the scanner determine the size of the grid.

[0039] Preferably, 5) generating a reflection intensity map: generating a reflection intensity image based on the image plane divided into grid points, wherein the image content is determined by the size and pixel value of the grid, and the pixel value of the grid point depends on the number of scanning points falling within the grid and the reflection intensity value of each scanning point;

[0040] The reflection intensity value of the laser point cloud data is the value obtained by the laser pulse emitted by the 3D laser scanner, reflected by the target, and then received by the scanner's receiving component. The magnitude of the reflection intensity depends on the reflective characteristics, color, material, scanning distance, and scanning angle of the target. Different scanning targets will obtain different reflection intensity values, which serve as the basis for generating the image.

[0041] The reflection intensity values of the scanning points recorded by the 3D laser scanner are in the range of [0, 1]. This range is converted to [0, 255] and used as the grayscale value of the corresponding pixel on the image to generate the required reflection intensity image.

[0042] Image generation is achieved based on OPENCV. The CvMat function is used to create an image matrix. The height of the matrix is the number of rows of the image, and the width of the matrix is the number of columns of the image. The number of rows of the image depends on the number of points scanned in the vertical direction, that is, the scanning accuracy of the 3D laser scanner. The number of columns of the image depends on the number of sections scanned in the forward direction of the rail vehicle, that is, the scanning frequency of the 3D laser scanner and the forward speed of the rail vehicle.

[0043] Calculate the width M and height N of the image plane, use the CvMat function of OPENCV to create a matrix with a width of M and a height of N, and then assign values to the MXN size one by one. Starting from the origin of the image plane coordinate system, with the pixel size as the step size, search for the scanning point closest to the grid point row by row and column by column, and use the reflection intensity value of the point as the grayscale value of the grid point, and assign the grayscale value to the matrix member of the corresponding row and column;

[0044] This application uses a weighted balance filtering method to solve the grid pixel assignment problem. A weighted balance filtering template is constructed, and the original pixel value is replaced by the average value in the template. If there are multiple scanning points in the grid, the average value of the reflection intensity values of all scanning points is taken as the grayscale value of the pixel. If there are no scanning points in the grid, the average value of several adjacent pixels within a certain range is taken as the grayscale value of the pixel. The image grayscale value at the pixel point (x, y) is:

[0045] g(x, y)=∑(f(x, y)*p) / N Equation 5

[0046] Where N is the total number of scanning points or adjacent pixels;

[0047] Finally, after calculating the pixel value of each grid point on the image plane, traverse and assign the value to the Mat matrix member at the corresponding position, then write the image data into the image file and compress the image size as needed;

[0048] The image matrix that has been assigned is used to directly generate an image, obtain the image information of the shield tunnel lining surface, and mark the tunnel mileage value above and below the image.

[0049] Preferably, the processing platform for stretched images: performs dynamic data scheduling based on a quadtree micro-level model to achieve stretched image processing, adjusts the subdivision level according to needs, reduces the amount of data required for rendering, and first saves pre-generated content of different levels of subdivision and pre-establishes indexes;

[0050] 1) Establishing a micro-level model: By combining micro-level data control strategies with internal and external memory scheduling, real-time interactive browsing of shield tunnel lining surface images is achieved. This allows the computer to selectively ignore some subdivisions during image visualization, depending on the image's distance from the viewpoint. The core of the micro-level model is the establishment of a micro-level model. This is achieved by simplifying the original high-resolution image at different subdivision levels and arranging them at different levels according to the resolution, forming a top-down model.

[0051] Generate an image of the shield tunnel lining surface extension at its original resolution. Due to the long and narrow shape of the original image, the image is subdivided into 256 parts by length. This is used as the bottommost and most subdivided level in the micro-level model. Then, every four images are merged, resampled, or compressed to obtain an image of a lower resolution. This is used as the next subdivision level. This step is repeated to generate images at different depth levels, thus obtaining different subdivision levels of the micro-level model.

[0052] 2) Indexing: The entire image is divided into images of different subdivision levels. The subdivision levels of low-level images are coarser, and the subdivision levels of high-level images are richer. The number of images in each layer is 4 to the power of n. Then, the image is processed using the aforementioned quadtree encoding. The entire tunnel image is used as the root node. The root node is continuously divided from top to bottom. Each non-leaf node has four child nodes. In other words, the image of the previous layer is divided into four equal parts to form the image of the next layer. The image area of the parent node is the same as the image area of the four child nodes.

[0053] The image structure index is established using a quadtree. id is the node code, which serves as the index number that uniquely identifies the node. The structure also records the parent node and four child nodes of the node. Length is the number of layers the node is located at.

[0054] 3) Fast rendering of stretched images: Implementing dynamic micro-level image scheduling based on OSG. First, define the image reading method. Based on the image file format, construct osgImage images of the same size and traverse and assign values to them according to the pixel storage format. Then, based on the image size and subdivision level, define the coordinates of its outer bounding rectangle and create a texture for it.

[0055] After completing the above steps, use OSG to create a micro-level model. Based on the quadtree index, traverse all leaf nodes and load images of different resolutions into different levels of the micro-level model to build a top-down micro-level model. Then, set the center position and display distance range of each image in the micro-level model, and OSG automatically loads the image.

[0056] After the above settings are completed, when the computer performs real-time rendering, the distance is calculated based on the current viewpoint position, and the images of the corresponding subdivision level in the micro-level model are automatically loaded. The high-resolution images are displayed when they are closer to the viewpoint, and the low-resolution images are displayed when they are farther away from the viewpoint. Only the image part within the window is displayed, and the images outside the window are automatically unloaded.

[0057] Micro-level models are used to interactively browse the generated shield tunnel lining surface images. When the distance from the viewpoint is far, the micro-level model automatically uses a coarse image. When the image is zoomed in, a fine image is loaded. At the same time, the micro-level model preloads images near the window to improve rendering efficiency when zooming, and automatically unloads the parts outside the window to reduce unnecessary rendering subdivisions.

[0058] Preferably, the segment boundary line is extracted: the location of the segment boundary line is obtained, the segment point clouds on both sides of the boundary line are compared, a mapping relationship is established between the two coordinate spaces, the detected features are peaked at a point in the other coordinate system, and the problem of detecting the shape is converted into a problem of calculating the peak value; the three curves intersect at one point on the polar coordinate plane, representing a straight line passing through the three points (x0, y0), (x1, y1), and (x2, y2) at the same time. The θ-ρ polar coordinate parameter value of the point is the polar diameter and polar angle of the straight line. For all points on a straight line, the curves drawn on the θ-ρ plane of the polar coordinate system intersect at one point;

[0059] The straight line is detected by counting the number of curves that intersect at a point on the plane θ-ρ. By converting each point on the image into polar coordinates and tracing the intersection points between the corresponding curves, the number of curves that intersect at a point is obtained. The greater the number, the more points there are on the straight line. At the same time, a critical value is pre-set to define the detection of a straight line. Only when the number of curves that intersect at a point exceeds the critical value, a straight line is detected. The parameter value of this intersection (θ, ρ θ ) are the polar diameter and polar angle of the line.

[0060] Preferably, edge image high-frequency filtering: before extracting the characteristic shape of the image, the image is first subjected to edge high-frequency filtering: first, using Gaussian filtering to smooth the image, second, finding the intensity gradient of the image, third, non-maximum suppression, fourth, using double critical value for detection, and fifth, hysteresis boundary tracking;

[0061] The pixels of the confirmed strong edge are regarded as the true edge points, while the weak edge points still need to be judged and tracked. If there is a strong edge point in the 8-connected area of the weak edge point, the weak edge point is determined to be the true edge.

[0062] Preferably, segment boundary line detection: all pixel points on the image are converted into a polar coordinate system, and a straight line passing through each point is drawn as a sine curve on the plane θ-ρ. The number of curves intersecting at a point is counted, and points with a number exceeding a critical value are considered to be a straight line. Finally, all detected straight lines are drawn on the original image;

[0063] Write a line transformation function to detect straight lines on the image. The input value of the function is the edge binary image. The distance accuracy of the line search is determined. The default setting is 1 pixel, that is, all pixels on the image are calculated. Then the angle accuracy of the line search is determined. The default setting is 1°, that is, each point is calculated 360 times during the line detection. When the line detection is not accurate enough, the angle accuracy is improved to solve it. Finally, the pixel coordinates of each point are used to calculate the corresponding ρ value. For each pair of parameter values (θ, ρ θ ) to perform calculations;

[0064] The critical value parameters are pre-set in the online transformation function, and only the parameter values (θ, ρ θ ) is greater than the critical value, it can pass the detection and be judged as a straight line. At the same time, the minimum length of the detected line segment is also set. If the detected line segment length is lower than this value, it will not be displayed.

[0065] The function is modified to only extract lines with polar angles around 0° to reduce horizontal interference. The arch part of the tunnel is intercepted for line extraction. Two lines that are very close to each other will be considered as the same line. Finally, the position of each line is output and converted into the corresponding mileage value.

[0066] Preferably, extract cross-sectional information:

[0067] 1) Section point extraction: By using the obtained segment boundary line mileage, the section information at the corresponding location is extracted. The core of the section point extraction process is to directly extract a series of points with specific mileage values from the scan file;

[0068] The steps for extracting cross-section point clouds based on a specific mileage are as follows:

[0069] Step 1: Based on the given mileage value S, obtain the time t when the laser scanning railcar arrives at the mileage S;

[0070] Step 2: Extract a series of points whose scanning time is closest to t in the scan file, which are the cross-section points to be extracted;

[0071] In actual operation, the segment sections on both sides of the extracted dividing line are compared. A 3cm thick cross-section point cloud is extracted from both ends of a given mileage value and then projected onto the same plane for comparison.

[0072] 2) Cross-section point fitting denoising: Remove noise points and debris points in the cross-section information:

[0073] Step 1: The extracted cross-section information N contains normal points, noise points, and debris points, and the model M of the normal points is obtained in advance. The cross-section model in the shield tunnel is a standard circle;

[0074] Step 2: Randomly extract the minimum subset n that can generate the model M (circular section) from the cross-section point cloud, and fit a circular section m with the subset n;

[0075] Step 3: Set a critical value e, and determine whether all points in N can meet the critical value e for the circular cross section m. If the point can meet the critical value e, then add it to n;

[0076] Step 4: When all points in N have completed the third step, the points in n can be identified as normal points describing the model M, and the remaining points are noise points and debris points;

[0077] Randomly select several points in the cross-section point cloud to fit a circle, then calculate the distance from each scanning point on the cross section to the fitted circle, retain the points with a distance less than 0.05 meters as normal points on the cross section, repeat the above steps, if the number of normal points accounts for 60% of the total, then the fitted circle is considered reasonable.

[0078] Optimally, segment misalignment data and analysis:

[0079] After fitting and denoising the cross-sectional point clouds on both sides of the segment dividing line, the next step is to calculate the misalignment of the cross-sectional points. Since the intercepted cross-sectional information is of a certain thickness on both sides of the dividing line, the two parts of the point cloud must first be projected onto the same plane for comparison. The misalignment value is the distance between the two cross-sectional points. The point cloud information is projected onto the same vertical plane of the tunnel. The scanned data of the point cloud is extended with the forward direction as the third-dimensional coordinate. The coordinates of the two parts of the point cloud in this direction are set to 0, and then projected onto the same vertical plane of the tunnel.

[0080] After projecting the two cross-section point clouds onto the same plane, they are represented by different colors. After visualization, the places with obvious differences are the places where the misalignment occurs. One of the cross sections is used as the reference surface and the other as the comparison surface. The center position of the reference surface is calculated. Then, the angle of all cross-section points relative to the center of the circle is obtained by calculating the inverse tangent value. The points with the same angle on the two cross sections are extracted, and the distance between them is the misalignment value.

[0081] The misalignment value on the cross-section circumference is calculated with a step size of only 1°. Several segment rings scanned in the interval are selected, and the misalignment values between the segment rings are calculated. The size, number and proportion of the misalignment values are calculated to generate an analysis table of the misalignment situation of the tunnel segments.

[0082] Compared with the existing technology, the innovation and advantages of this application are:

[0083] (1) The traditional method of using tape measures and triangles for regular inspections and one-by-one inspections is inefficient and labor-intensive. This application utilizes the characteristics of 3D laser scanning technology, such as high efficiency, high precision, large area, no need to contact the target, and no requirements for light, and applies it to the detection of shield tunnel segment misalignment. A 3D laser scanning rail transit vehicle is used, which combines a 3D laser scanner with a rail transit vehicle and is equipped with an odometer, a built-in computer, and a power supply component. A mobile laser scanner is used to quickly scan the tunnel and obtain complete information on the tunnel lining surface, greatly improving the efficiency of field measurement and analyzing the tunnel misalignment based on point cloud data. The 3D laser scanner scans the shield tunnel in a spiral pattern. Programs are compiled for the connection, control, and output of the 3D laser scanner, enabling more flexible scanner operation during the scanning process to avoid interruptions. Using the scanned point cloud data, mileage correction is performed to determine the temporal relationship between mileage and railcar position. Projection transformation is performed based on coordinates and reflection intensity values to generate a high-definition orthophoto of the shield tunnel lining surface. Dynamic data scheduling based on a quadtree micro-level model enables fast interactive browsing of the shield tunnel lining surface image. The orthophoto of the tunnel lining surface is then used to extract the segment boundary line. Cross-sectional point clouds on both sides are extracted based on the mileage value of the boundary line. These cross-sectional point clouds are fitted and denoised, then projected onto the same plane for comparison to determine the size and number of misalignments. This technology offers high detection accuracy and reliability, meeting the precision requirements for misalignment detection. The system also features a simple, fast, and efficient workflow, significantly reducing labor intensity and possessing great application prospects.

[0084] (2) This application realizes the development of line scanning data acquisition software, writes Leica P30 line scanning data acquisition software, uses the WiFi and IP address in the instrument to connect to the scanner, and then remotely sets the scanner's parameters and modes, creates new scanning projects and scanning files, controls the scanner to start and stop, and records the scanning data. Finally, it can save and export the scanning data file, thereby obtaining the scanning time, two-dimensional coordinates, reflection intensity and other information of the scanning point. The composition structure of the three-dimensional laser scanning subway rail transit vehicle system is constructed, which consists of a rail transit vehicle, a built-in computer, a three-dimensional laser scanner, and an odometer sensor. When using this system for inspection, the inspection personnel push the rail transit vehicle to move forward along the center line of the track. The three-dimensional laser scanner transmits and receives high-speed laser pulses through the line scanning mode, and can perform spiral full-section, high-density scanning of the shield tunnel.

[0085] (3) The stretched image generation of shield tunnel point cloud data is realized. The scanned point cloud data is converted into an orthophoto image to facilitate the observation of the shield tunnel surface. By combining the mileage data of the corresponding time period recorded in the odometer, the scanned point cloud data of the tunnel lining surface is stretched from two-dimensional coordinates to obtain the three-dimensional absolute coordinates of all points. Finally, the point cloud is projected onto a two-dimensional plane with the scanner as the center. Using OPENCV software, an image plane coordinate system is established to generate a laser reflection intensity map, and image information of the tunnel lining surface can be obtained. By designing software to dynamically schedule images and visualize them, the tunnel image is first generated into a small unit image, an index structure is established through a quadtree, and then an image with lower pixels is generated upwards. The OSG plug-in is used to dynamically read and write images. The dynamic scheduling method prevents the image portion outside the window from being loaded and only the portion inside the window is displayed, effectively improving memory usage efficiency and facilitating the reading of information on the image.

[0086] (4) A reflection intensity map of the shield tunnel surface was generated. By establishing a point cloud model for projection transformation, the real scene information of the tunnel surface was obtained. The required content was extracted from the reflection intensity map, which is intuitive, rich in information and high in precision, and reflects the real situation of the tunnel. By constructing a quadtree index and using a micro-level model to dynamically and interactively browse the image, the efficiency of image drawing was improved. It can analyze the misalignment of the shield tunnel, complete the extraction of the segment boundary line in the image, extract the cross-section point information according to the mileage and perform fitting and denoising, and finally obtain the misalignment of the shield tunnel after comparison. This application uses mobile three-dimensional laser scanning technology to detect segment misalignment. The experimental data shows that this technology has rich scanning subdivisions and high detection accuracy, which can well meet the requirements of misalignment detection. At the same time, the system has a simple operation process and high work efficiency, which greatly reduces the labor intensity of the staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 This is a schematic diagram of the coordinates of scanning points for shield tunnel calculation.

[0088] Figure 2 It is to calculate the coordinates of the projection point to obtain the isotropic tangent projection.

[0089] Figure 3 It converts the cylindrical point cloud data coordinates into a rectangular projection plane.

[0090] Figure 4 It is an image of the shield tunnel lining surface.

[0091] Figure 5 It is an image structure index map established using a quadtree.

[0092] Figure 6 It is a vision image of a fast rendering image management platform for stretching images.

[0093] Figure 7 It is a close-up view of a fast-rendering image management platform for stretching images.

[0094] Figure 8 It is a high-frequency filtering image of the edge image of the shield tunnel lining surface extension image.

[0095] Figure 9 It is a schematic diagram for extracting cross-sectional information.

[0096] Figure 10 It is a schematic diagram of data denoising by cross-section point fitting.

[0097] Figure 11 The calculation result diagram of the misalignment on the cross-section circumference is obtained with a step size of 1°. DETAILED DESCRIPTION

[0098] Below, in conjunction with the accompanying drawings, the technical solution of the laser scanning rail transit vehicle subway tunnel segment misalignment analysis method provided in this application is further described so that technical personnel in this field can better understand this application and implement it.

[0099] As a new mode of transportation, the rail transit system has greatly alleviated the pressure of travel and is becoming more and more popular among people. With the development of the subway system, mastering the real-time and effective health status of subway rail transit has become a necessary part of subway construction. Tunnel segment misalignment detection is an important part of subway inspection work. Traditionally, tape measures and triangles are used for regular inspections and one-by-one inspections, which are inefficient and labor-intensive. This application utilizes the characteristics of three-dimensional laser scanning technology, such as high efficiency, high precision, large area, no need to contact the target, and no requirements for light, and applies it to the detection of shield tunnel segment misalignment.

[0100] This application uses a three-dimensional laser scanning rail transit vehicle, combines a three-dimensional laser scanner with a rail transit vehicle, and is equipped with an odometer, a built-in computer, and a power supply component. It uses mobile laser scanning to quickly scan the tunnel and obtain complete information on the tunnel lining surface, greatly improving the efficiency of field measurement and analyzing the tunnel misalignment based on point cloud data.

[0101] The 3D laser scanner scans the shield tunnel in a spiral manner;

[0102] Compile corresponding programs for the connection, control, and output of the 3D laser scanner, which enables more flexible operation of the scanner during the scanning process to avoid affecting the scanning process;

[0103] Using the scanned point cloud data, mileage correction is performed to obtain the time relationship between mileage and railcar position. Projection transformation is performed based on coordinates and reflection intensity values to generate a high-definition orthophoto image of the shield tunnel lining surface. Then, based on the dynamic scheduling of data from the quadtree micro-level model, fast interactive browsing of the shield tunnel lining surface image is achieved.

[0104] The segment boundary line is extracted from the orthophoto image of the tunnel lining surface. The cross-section point clouds on both sides are extracted based on the mileage value of the boundary line. The cross-section point clouds are fitted and denoised, and then projected onto the same plane for comparison to determine the size and amount of misalignment.

[0105] This technology is highly accurate and reliable, meeting the precision requirements of misalignment detection. The system also features a simple, fast, and efficient workflow, significantly reducing staff workload and possessing broad application prospects.

[0106] 1. 3D Laser Scanning Rail Transit Vehicle System Architecture

[0107] The hardware of the 3D laser scanning rail transit vehicle measurement system includes a 3D laser scanner, a rail transit vehicle, a built-in computer, an odometer and a tilt sensor, as well as a power supply. The rail transit vehicle is the main structure, and the other parts are closely integrated by being mounted on the vehicle body structure. It is the link connecting the various parts. The odometer calculates the relationship between the mileage of the scanner and the operating time, and the tilt sensor calculates the posture of the scanner. The built-in computer is the control center and storage center of the entire system, controlling the operation and stop of other components and saving the collected data on the hard disk.

[0108] 2. Generation and processing of shield tunnel lining surface extension images

[0109] The point cloud data obtained by scanning the shield tunnel using mobile scanning measurement is projected through the establishment of a point cloud model to restore the original image information of the shield tunnel lining surface, thereby obtaining the real scene of the tunnel surface. These graphics generated from the point cloud data reflect the real situation of the shield tunnel lining surface, which is intuitive, rich in information and high in precision, and the required content can be extracted from it.

[0110] Based on the generated image information, the misalignment between tunnel segments is analyzed, and finally the misalignment of shield tunnel segments is obtained. Since the generated image data is highly accurate and the data volume is large, in order to facilitate browsing and retrieval of images, software design is carried out based on OSG, and an image processing platform is established to complete the visualization and dynamic scheduling of image graphs.

[0111] (1) Generating stretched images

[0112] The scanning starting mileage is used as the origin, and a three-dimensional coordinate system is established. The three-dimensional coordinates of the point cloud data relative to the scanning center are calculated. The high-precision and high-density three-dimensional coordinate information of the tunnel surface is obtained, and the outline of the real object is restored by building a model, thereby analyzing the actual situation of the measured tunnel surface.

[0113] 1. Calculate the scanning point coordinates

[0114] The mileage value of the scanning point is used as the Y coordinate of the scanning point, and the two-dimensional point cloud is restored to a three-dimensional point cloud. At this time, the three-dimensional point cloud is cylindrical. The scanning center at the start of the scan is the origin, and the radial direction is the X axis, the forward direction of the rail vehicle is the Y axis, and the vertical road direction is the Z axis to establish a coordinate system, such as Figure 1 shown.

[0115] The three-dimensional coordinates of any point P on the tunnel surface are:

[0116] (X P ,Y P ,Z P )=(S·SINβ,y P -y0,S·COSβ) Formula 1

[0117] Where S is the scanning distance, β is the scanning altitude angle, and y P is the mileage value of the scanning point, y O is the mileage value of the coordinate origin.

[0118] 2. Correct coordinates based on design files

[0119] Because tunnels have curved sections, rail vehicles experience tilt and superelevation when navigating them, resulting in uneven scanning of both sides of the track. Odometry alone only measures the distance traveled by the laser scanner vehicle during scanning and the vehicle's mileage at any given moment, but cannot accurately determine the vehicle's center position. This mileage data is corrected based on subway tunnel design documents.

[0120] Design files are generated based on the drawings used during design and construction and are divided into two types: horizontal curves and vertical curves. The horizontal curve design file contains information such as the line type for each mileage segment, the corresponding starting mileage, ending mileage, starting point coordinates, ending point coordinates, starting azimuth, and ending azimuth. The vertical curve design file contains the elevation of the slope change point, the slope of the front and back slopes at the slope change point, the mileage at the slope change point, and the radius of the circular curve near the slope change point. With the design file, the tunnel design line type at any mileage can be determined. Based on the mileage given by the odometer, the corresponding line type can be found and the coordinates of the projection point can be corrected.

[0121] 3. Establish the projection plane

[0122] To obtain the information of the shield tunnel lining surface, it is necessary to generate an orthophoto image of the shield tunnel lining surface and project the cylindrical three-dimensional point cloud data onto a two-dimensional plane. First, establish a projection plane and calculate the coordinates of all projection points under the projection plane, such as Figure 2 As shown in the figure, the tunnel is transformed from the original three-dimensional coordinate system to two-dimensional plane coordinates based on the geometric relationship by using the angle values of the three-dimensional laser scanning points and isometric tangent projection;

[0123] A two-dimensional coordinate system is established with the arch position at the start of the tunnel scan as the origin, the forward direction of the railcar as the m-axis, and the vertical forward direction as the n-axis. The coordinates of the cylindrical point cloud data are converted into a rectangular projection plane, as shown in the following example: Figure 3 As shown, the dome position n = 0;

[0124] The coordinates of any point P on the plane are:

[0125] (m P ,n P )=(Y P ,βπr / 180) Equation 2

[0126] Where Y P is the coordinate of point P in the three-dimensional coordinate system, β is the scanning height angle, and r is the radius of the shield tunnel;

[0127] The point cloud data stretched into a two-dimensional rectangular plane is the basis for generating orthophoto images. A cross section obtained by a 3D laser scanner scanning a circle is stretched into a column of points with the same m coordinates. The length in the m direction is the distance traveled in the scan.

[0128] The rectangular projection plane is the minimum enclosing rectangle of all projection points. Through the conversion formula, the position of any point on the tunnel lining surface can be obtained from its mileage value and the scanning point angle value, providing coordinate transformation for the generation of the image.

[0129] 4. Establish image coordinate system

[0130] The image plane coordinate system xoy is established with the projection plane. The coordinate axis direction is consistent with the projection plane coordinate axis. The lower left corner of the projection rectangle is the coordinate origin. The coordinates of any point P on the plane are:

[0131] (x P ,y P )=(m P , n P -n min ) Formula 3

[0132] Where m P 、n P is the coordinate of the scanning point P in the projection plane, n min is the point with the smallest n coordinate in the projection plane;

[0133] After establishing the image plane coordinate system, all scanning points are converted into two-dimensional coordinate points on the image plane. The image plane is used as the base map for generating the reflection intensity image. The content of the image depends on the scanning point information on each coordinate. The scanning sampling interval of the point cloud data is used as the pixel size (step size) of the image to be generated. The image plane is divided into a grid according to the step size, and each grid point is a pixel point. The width M and height N of the image are calculated as shown in Formula 4:

[0134] M=ceil(m max / delta)+1

[0135] N=ceil((n max -n min ) / delta)+1 Formula 4

[0136] In the formula, ceil() means rounding; delta means the sampling interval of the point cloud;

[0137] The image plane coordinate system is divided into grid points according to pixel size, and the projection plane is also divided into corresponding grids. All scanning points fall into the corresponding grids according to their positions. The scanning parameters set by the scanner determine the size of the grid.

[0138] 5. Generate reflection intensity map

[0139] The image plane based on the divided grid points generates a reflection intensity image. The image content is determined by the size and pixel value of the grid. The pixel value of the grid point depends on the number of scanning points falling into the grid and the reflection intensity value of each scanning point.

[0140] The reflection intensity value of the laser point cloud data is the value obtained by the laser pulse emitted by the 3D laser scanner, reflected by the target, and then received by the scanner's receiving component. The magnitude of the reflection intensity depends on the reflective characteristics, color, material, scanning distance, and scanning angle of the target. Different scanning targets will obtain different reflection intensity values, which serve as the basis for generating the image.

[0141] The reflection intensity values of the scanning points recorded by the 3D laser scanner are in the range of [0, 1]. This range is converted to [0, 255] and used as the grayscale value of the corresponding pixel on the image to generate the required reflection intensity image.

[0142] Image generation is achieved based on OPENCV. The CvMat function is used to create an image matrix. The height of the matrix is the number of rows of the image, and the width of the matrix is the number of columns of the image. The number of rows of the image depends on the number of points scanned in the vertical direction, that is, the scanning accuracy of the 3D laser scanner. The number of columns of the image depends on the number of sections scanned in the forward direction of the rail vehicle, that is, the scanning frequency of the 3D laser scanner and the forward speed of the rail vehicle.

[0143] Calculate the width M and height N of the image plane, use the CvMat function of OPENCV to create a matrix with a width of M and a height of N, and then assign values to the MXN size one by one. Starting from the origin of the image plane coordinate system, with the pixel size as the step size, search for the scanning point closest to the grid point row by row and column by column, and use the reflection intensity value of the point as the grayscale value of the grid point, and assign the grayscale value to the matrix member of the corresponding row and column;

[0144] The pixel values of a grid are determined by the reflection intensity of the scan points that fall within the grid. However, a grid often contains no scan points or multiple scan points. In terrestrial 3D laser scanning, due to instrument errors, the reflective properties of the measured object, and environmental factors, the laser receiver may not receive the returned laser light, resulting in missing scan points. Furthermore, if the scanning resolution is too high and the scan point spacing is smaller than the grid size, multiple scan points may appear in the same pixel. Therefore, before generating an image of the tunnel lining surface, it must be processed to determine the pixel value of each grid cell in the image plane.

[0145] This application uses a weighted balance filtering method to solve the grid pixel assignment problem. A weighted balance filtering template is constructed, and the original pixel value is replaced by the average value in the template. If there are multiple scanning points in the grid, the average value of the reflection intensity values of all scanning points is taken as the grayscale value of the pixel. If there are no scanning points in the grid, the average value of several adjacent pixels within a certain range is taken as the grayscale value of the pixel. The image grayscale value at the pixel point (x, y) is:

[0146] g(x, y)=∑(f(x, y)*p) / N Equation 5

[0147] Where N is the total number of scanning points or adjacent pixels;

[0148] Finally, after calculating the pixel value of each grid point on the image plane, traverse and assign the value to the Mat matrix member at the corresponding position, then write the image data into the image file and compress the image size as needed;

[0149] The image matrix that has been assigned is used to directly generate the image, and the image information of the shield tunnel lining surface is obtained. The tunnel mileage value is marked on the top and bottom of the image. The resolution of the image generated in this experiment is 5mm. Figure 4 shown.

[0150] (2) Stretching image processing platform

[0151] Due to the high scanning accuracy of 3D laser scanners, visualization of point cloud data on the shield tunnel lining surface requires high-resolution images, resulting in slow loading times and high memory consumption. Furthermore, subway tunnels are long and narrow, resulting in rectangular images with length significantly greater than width. This makes dragging images inconvenient when the most detailed information isn't always needed. In practice, it's crucial to ensure that when viewing an image, workers can zoom in on a specific area to display the most accurate image content and detailed information. Therefore, processing of the stretched images is necessary to achieve real-time, efficient image scheduling and retrieval.

[0152] Based on the quadtree micro-level model, data is dynamically scheduled to achieve stretched image processing. The micro-level adjusts the subdivision level as needed to reduce the amount of data required for rendering. The pre-generated content of different levels of subdivision is first saved and indexed in advance.

[0153] 1. Establish a micro-level model

[0154] By combining micro-level data control strategies with internal and external memory scheduling, real-time interactive browsing of shield tunnel lining surface images is achieved. This allows the computer to selectively ignore some subdivisions when visualizing the image, depending on the distance of the image from the viewpoint, thereby improving rendering efficiency. The core of the micro-level is the establishment of a micro-level model. By simplifying the original high-resolution image at different subdivision levels and arranging them at different levels according to the resolution, a top-down model is formed.

[0155] Generate an image of the shield tunnel lining surface at its original resolution. Due to the long and narrow shape of the original image, the image is subdivided into 256 segments by length. This serves as the bottom, most subdivided level in the micro-level model. Then, every four images are merged, resampled, or compressed to create an image of a lower resolution, which serves as the next subdivision level. This process is repeated to generate images at different depths, forming different subdivision levels of the micro-level model. In this micro-level model, the resolution of the upper image is twice that of the lower image. Higher-resolution images provide more detailed subdivision accuracy.

[0156] 2. Create an index

[0157] The entire image is divided into images of different subdivision levels. The subdivision levels of low-level images are coarser, and the subdivision levels of high-level images are richer. The number of images in each layer is 4 to the power of n. Then, the image is processed using the above-mentioned quadtree encoding. The entire tunnel image is used as the root node. The root node is continuously divided from top to bottom. Each non-leaf node has four child nodes. That is, the image of the previous layer is divided into four equal parts as the image of the next layer. The image area of the parent node is the same as the image area composed of the four child nodes.

[0158] The image structure index established using the quadtree is as follows Figure 5 As shown, id is the node code, which is the index number that uniquely identifies the node. The structure also records the parent node and four child nodes of the node, and length is the number of layers where the node is located.

[0159] 3. Fast rendering of stretched images

[0160] Implementing dynamic image scheduling based on micro-level based on OSG. First, define the image reading method. According to the format of the image file, construct an osgImage image of the same size and traverse and assign values to it according to the pixel storage format. Then, according to the size and subdivision level of the image, define the coordinates of its outer bounding rectangle and create a texture for it.

[0161] After completing the above steps, use OSG to create a micro-level model. Based on the quadtree index, traverse all leaf nodes and load images of different resolutions into different levels of the micro-level model to build a top-down micro-level model. Then, set the center position and display distance range of each image in the micro-level model, and OSG automatically loads the image.

[0162] After the above settings are completed, when the computer performs real-time rendering, the distance is calculated based on the current viewpoint position, and the images of the corresponding subdivision level in the micro-level model are automatically loaded. The high-resolution images are displayed when the distance is closer to the viewpoint, and the low-resolution images are displayed when the distance is farther away from the viewpoint. Only the image part within the window is displayed, and the image outside the window is automatically unloaded;

[0163] Use micro-level to generate interactive browsing of shield tunnel lining surface images, data such as Figure 6 、 Figure 7 When the distance from the viewpoint is far, the micro-level model automatically uses a coarse image. When the image is zoomed in, it loads a fine image. At the same time, the micro-level model preloads images near the window to improve rendering efficiency when zooming, and automatically unloads the parts outside the window to reduce unnecessary rendering subdivisions.

[0164] 3. Analysis of Segment Misalignment in Shield Tunnel

[0165] First, the boundary lines of adjacent segments are extracted from the generated shield tunnel surface lining image. After the boundary lines are extracted from the image, the mileage value data of the boundary lines are obtained. Then, the corresponding cross-sectional point cloud is extracted from the 3D point cloud data based on the mileage value data. The point clouds on both sides of the segment are projected onto the same plane, and the difference between the adjacent segment sections is compared. The places with obvious differences are where the misalignment occurs. Finally, the misalignment of the entire shield tunnel is calculated, and a misalignment analysis chart is output.

[0166] (1) Extracting the segment boundary line

[0167] The location of the segment boundary line is obtained, and the segment point clouds on both sides of the boundary line are compared. A mapping relationship is established between the two coordinate spaces, and the detected features form a peak at a point in the other coordinate system, converting the problem of detecting the shape into a problem of calculating the peak value. The three curves intersect at one point on the polar coordinate plane, representing a straight line passing through the three points (x0, y0), (x1, y1), and (x2, y2) at the same time. The θ-ρ polar coordinate parameter value of this point is the polar diameter and polar angle of this line. For all points on a straight line, the curves drawn on the θ-ρ plane of the polar coordinate system intersect at one point.

[0168] The straight line is detected by counting the number of curves that intersect at a point on the plane θ-ρ. By converting each point on the image into polar coordinates and tracing the intersection points between the corresponding curves, the number of curves that intersect at a point is obtained. The greater the number, the more points there are on the straight line. At the same time, a critical value is pre-set to define the detection of a straight line. Only when the number of curves that intersect at a point exceeds the critical value, a straight line is detected. The parameter value of this intersection (θ, ρ θ ) are the polar diameter and polar angle of the line.

[0169] 1. High-frequency filtering of edge images

[0170] Due to the complexity of the information in the image, before extracting the characteristic shape of the image, the image is first subjected to edge high-frequency filtering: first, the image is smoothed using Gaussian filtering, second, the intensity gradient of the image is found, third, non-maximum suppression is performed, fourth, double critical value detection is used, and fifth, hysteresis boundary tracking is performed;

[0171] The pixels of the confirmed strong edge are regarded as the true edge points, while the weak edge points still need to be judged and tracked. If there is a strong edge point in the 8-connected area of the weak edge point, the weak edge point is determined to be the true edge.

[0172] The generated shield tunnel lining surface extension image is subjected to edge image high-frequency filtering, and the data of a part of the image is as follows Figure 8 shown.

[0173] 2. Segment boundary line detection

[0174] All pixels on the image are converted into polar coordinates. A straight line passing through each point is plotted as a sine curve on the θ-ρ plane. The number of curves intersecting at a point is counted. Points with a number exceeding a critical value are considered to be a straight line. Finally, all detected straight lines are plotted on the original image.

[0175] Write a line transformation function to detect straight lines on the image. The input value of the function is the edge binary image. Determine the distance accuracy when searching for a straight line. The default setting is 1 pixel, that is, all pixels on the image are calculated. Then determine the angle accuracy when searching for a straight line. The default setting is 1°, that is, each point is calculated 360 times when detecting a straight line. If the straight line detection is not accurate enough, it can be solved by improving the angle accuracy. Finally, the pixel coordinates of each point are used to calculate the corresponding ρ value. For each pair of parameter values (θ, ρ θ ) to perform calculations;

[0176] The critical value parameters are pre-set in the online transformation function, and only the parameter values (θ, ρ θ) is greater than the critical value, it can pass the detection and be judged as a straight line. At the same time, the minimum length of the detected line segment is also set. If the detected line segment length is lower than this value, it will not be displayed.

[0177] The function was modified to only extract lines with a polar angle of approximately 0°. Furthermore, since there are many structures on the tunnel wall, horizontal interference is reduced and the tunnel vault is intercepted for line extraction. This not only improves the accuracy of line detection but also increases work efficiency.

[0178] The generated binary image is extracted using cumulative probability transformation. Two closely spaced lines are considered to be the same line. Finally, the position of each line is output and converted into the corresponding mileage value.

[0179] (2) Extracting cross-section information

[0180] 1. Section point extraction

[0181] By using the mileage of the segment boundary line, the cross-section information at the corresponding position is extracted. The core of the cross-section point extraction process is to directly extract a series of points with specific mileage values from the scan file.

[0182] The steps for extracting cross-section point clouds based on a specific mileage are as follows:

[0183] Step 1: Based on the given mileage value S, obtain the time t when the laser scanning railcar arrives at the mileage S;

[0184] Step 2: Extract a series of points whose scanning time is closest to t in the scan file, which are the cross-section points to be extracted;

[0185] In actual operation, the segment sections on both sides of the extracted boundary line are compared. In order to prevent the influence of inaccurate extraction of the segment boundary line, a 3cm thick cross-section point cloud is extracted from both ends of the given mileage value and then projected onto the same plane for comparison. Figure 9 shown.

[0186] 2. Cross-section point fitting and denoising

[0187] Because shield tunnels are equipped with pipelines, lighting, evacuation platforms, and other equipment and facilities, and because the tunnel contains a lot of dust and debris, many scan points are not on the tunnel wall, affecting the scan quality and resulting in a large number of noise points in the cross-section information. Therefore, it is necessary to remove these noise and debris points from the cross-section information before comparing the segments.

[0188] Eliminate noise points and debris points in cross-section information:

[0189] Step 1: The extracted cross-section information N contains normal points, noise points, and debris points, and the model M of the normal points is obtained in advance. The cross-section model in the shield tunnel is a standard circle;

[0190] Step 2: Randomly extract the minimum subset n that can generate the model M (circular section) from the cross-section point cloud, and fit a circular section m with the subset n;

[0191] Step 3: Set a critical value e, and determine whether all points in N can meet the critical value e for the circular cross section m. If the point can meet the critical value e, then add it to n;

[0192] Step 4: When all points in N have completed the third step, the points in n can be identified as normal points describing the model M, and the remaining points are noise points and debris points;

[0193] Randomly select a few points in the cross-section point cloud to fit a circle, then calculate the distance from each scanning point on the cross section to the fitted circle, and retain the points with a distance less than 0.05 meters as normal points on the cross section. Repeat the above steps. If the number of normal points accounts for 60% of the total, the fitted circle is considered reasonable. Figure 10 shown.

[0194] (3) Segment misalignment data and analysis

[0195] 1. Error calculation data

[0196] After fitting and denoising the cross-sectional point clouds on both sides of the segment dividing line, the next step is to calculate the misalignment of the cross-sectional points. Since the intercepted cross-sectional information is of a certain thickness on both sides of the dividing line, the two parts of the point cloud must first be projected onto the same plane for comparison. The misalignment value is the distance between the two cross-sectional points. The point cloud information is projected onto the same vertical plane of the tunnel. The scanned data of the point cloud is extended with the forward direction as the third-dimensional coordinate. The coordinates of the two parts of the point cloud in this direction are set to 0, and then projected onto the same vertical plane of the tunnel.

[0197] After projecting the two cross-section point clouds onto the same plane, they are represented by different colors. After visualization, the places with obvious differences are the places where the misalignment occurs. One of the cross sections is used as the reference surface and the other as the comparison surface. The center position of the reference surface is calculated. Then, the angle of all cross-section points relative to the center of the circle is obtained by calculating the inverse tangent value. The points with the same angle on the two cross sections are extracted, and the distance between them is the misalignment value.

[0198] Since the scanning frequency of the scanner is very high and there are many points on the cross-section circle, the amount of calculation for each position is very large. Therefore, in this application, only 1° is used as the step size to calculate the misalignment value on the cross-section circle. The results are as follows: Figure 11 shown.

[0199] 2. Analysis of the mismatch

[0200] Select several segment rings scanned in the interval, calculate the misalignment values between the segment rings, calculate the size, number and proportion of the misalignment values, and generate an analysis table of the misalignment situation of the tunnel segments.

Claims

1. A laser scanning method for analyzing the misalignment of segments in subway tunnels of rail transit vehicles, characterized in that: The laser scanning rail transit vehicle combines a 3D laser scanner with a rail transit vehicle and is equipped with an odometer, a built-in computer, and a power supply. It uses a mobile laser to quickly scan the tunnel, obtain complete information on the tunnel lining surface, and analyze the tunnel misalignment based on the point cloud data. The 3D laser scanner scans the shield tunnel in a spiral manner; Corresponding programs are compiled for the connection, control, and output of the 3D laser scanner, allowing for flexible operation of the scanner during the scanning process. Using the scanned point cloud data, mileage correction is performed to determine the temporal relationship between mileage and railcar position. Projection transformation is performed based on coordinates and reflection intensity values to generate a high-definition orthophoto of the shield tunnel lining surface. Dynamic data scheduling based on the quadtree micro-level model enables fast interactive browsing of the shield tunnel lining surface image. Segment boundary lines are extracted from the orthophoto of the tunnel lining surface, and cross-sectional point clouds on both sides are extracted based on the mileage values of the boundary lines. The cross-sectional point clouds are fitted and denoised, then projected onto the same plane for comparison to determine the size and amount of misalignment. a. Generating and processing stretched images of the shield tunnel lining surface: First, a shield tunnel lining surface image is generated. Using the mileage data recorded in the odometer for the corresponding time period, the scanned point cloud data of the tunnel lining surface is stretched from two-dimensional coordinates to three-dimensional absolute coordinates. A geometric transformation is then performed to obtain a projection plane. This is then used to generate a shield tunnel lining surface image through gridding. Using a dynamic data scheduling method based on a quadtree micro-level model, the micro-level model is divided into different subdivision levels. By establishing a quadtree index, images of the appropriate subdivision level are dynamically scheduled based on the distance from the viewpoint to the image, achieving stretched image processing. b. Analysis of shield tunnel segment misalignment: First, based on line transformation and image edge detection, the segment boundary line in the image is extracted. Then, the segment cross-section information on both sides of the boundary line is extracted based on the mileage value, and fitting and denoising are performed. Finally, the cross-section points are projected onto a plane for comparison to obtain the segment misalignment. The size and amount of misalignment are calculated to generate a misalignment analysis table to obtain the shield tunnel misalignment situation.

2. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 1 is characterized in that: Generate stretched image: Scan the starting mileage as the origin, establish a 3D coordinate system, calculate the 3D coordinates of the point cloud data relative to the scanning center, and use the high-precision, high-density 3D coordinate information of the tunnel surface to restore the outline of the actual object by building a model, thereby analyzing the actual situation of the measured tunnel surface; 1) Calculate the coordinates of the scanning point: Use the mileage value of the scanning point as the Y coordinate of the scanning point, and restore the 2D point cloud to a 3D point cloud. The 3D point cloud at this time is cylindrical. The scanning center at the start of the scan is used as the origin, and a coordinate system is established with the radial direction as the X axis, the forward direction of the railcar as the Y axis, and the direction perpendicular to the road surface as the Z axis. The three-dimensional coordinates of any point P on the tunnel surface are: (X P , Y P , Z P ) = (S·SINβ, y P - y0, S·COSβ) Equation 1 Where S is the scanning distance, β is the scanning altitude angle, and y P is the mileage value of the scanning point, y O is the mileage value of the coordinate origin; 2) Correct coordinates based on design files: Design files refer to files generated based on the drawings used during design and construction, and are divided into horizontal curves and vertical curves. The horizontal curve design file contains information such as the line type of each mileage segment, the starting mileage, ending mileage, starting point coordinates, ending point coordinates, starting azimuth, and ending azimuth of the corresponding line type. The vertical curve design file contains the elevation of the slope change point, the slope of the front and back slopes of the slope change point, the mileage of the slope change point, and the radius of the circular curve near the slope change point. Based on the design file, obtain the tunnel design line type at any mileage, find the corresponding line type based on the mileage given by the odometer, and correct the coordinates of the projection point.

3. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 2 is characterized by: 3) Establishing a projection plane: Generate an orthophoto of the shield tunnel lining surface by projecting the cylindrical 3D point cloud data onto a 2D plane. First, establish the projection plane, calculate the coordinates of all projected points on the projection plane, and then, using the angle values of the 3D laser scanning points and conformal tangent projection, convert the tunnel's original 3D coordinate system into 2D coordinates based on geometric relationships. A two-dimensional coordinate system is established with the arch position at the start of the tunnel scan as the origin, the railcar forward direction as the m-axis, and the vertical forward direction as the n-axis. The coordinates of the cylindrical point cloud data are converted into a rectangular projection plane, with the arch position n = 0; The coordinates of any point P on the plane are: (m P , n P ) = (Y P , βπr / 180) Equation 2 Where Y P is the coordinate of point P in the three-dimensional coordinate system, β is the scanning height angle, and r is the radius of the shield tunnel; The point cloud data stretched into a two-dimensional rectangular plane is the basis for generating orthophoto images. A cross section obtained by a 3D laser scanner scanning a circle is stretched into a column of points with the same m coordinates. The length in the m direction is the distance traveled in the scan. The rectangular projection plane is the minimum enclosing rectangle of all projection points. Through the conversion formula, the position of any point on the tunnel lining surface can be obtained from its mileage value and the scanning point angle value, providing coordinate conversion for the generation of the image map. 4) Establish the image coordinate system: Use the projection plane to establish the image plane coordinate system xoy. The coordinate axis direction is consistent with the projection plane coordinate axis. The lower left corner of the projection rectangle is the coordinate origin. The coordinates of any point P on the plane are: (x P , y P ) = (m P , n P -n min ) Equation 3 Where m P 、n P is the coordinate of the scanning point P in the projection plane, n min is the point with the smallest n coordinate in the projection plane; After establishing the image plane coordinate system, all scanning points are converted into two-dimensional coordinate points on the image plane. The image plane is used as the base map for generating the reflection intensity image. The content of the image depends on the scanning point information on each coordinate. The scanning sampling interval of the point cloud data is used as the pixel size (step size) of the image to be generated. The image plane is divided into a grid according to the step size, and each grid point is a pixel point. The width M and height N of the image are calculated as shown in Formula 4: M=ceil(m max / delta)+1 N = ceil((n max - n min ) / delta) + 1 Equation 4 In the formula, ceil() means rounding; delta means the sampling interval of the point cloud; The image plane coordinate system is divided into grid points according to pixel size, and the projection plane is also divided into corresponding grids. All scanning points fall into the corresponding grids according to their positions. The scanning parameters set by the scanner determine the size of the grid.

4. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 2 is characterized by: 5) Generate reflection intensity map: Generate a reflection intensity image based on the image plane divided into grid points. The image content is determined by the size and pixel value of the grid. The pixel value of the grid point depends on the number of scan points falling into the grid and the reflection intensity value of each scan point. The reflection intensity value of the laser point cloud data is the value obtained by the laser pulse emitted by the 3D laser scanner, reflected by the target, and then received by the scanner's receiving component. The magnitude of the reflection intensity depends on the reflective characteristics, color, material, scanning distance, and scanning angle of the target. Different scanning targets will obtain different reflection intensity values, which serve as the basis for generating the image. The reflection intensity values of the scanning points recorded by the 3D laser scanner are in the range of [0, 1]. This range is converted to [0, 255] and used as the grayscale value of the corresponding pixel on the image to generate the required reflection intensity image. Image generation is achieved based on OPENCV. The CvMat function is used to create an image matrix. The height of the matrix is the number of rows of the image, and the width of the matrix is the number of columns of the image. The number of rows of the image depends on the number of points scanned in the vertical direction, that is, the scanning accuracy of the 3D laser scanner. The number of columns of the image depends on the number of sections scanned in the forward direction of the rail vehicle, that is, the scanning frequency of the 3D laser scanner and the forward speed of the rail vehicle. Calculate the width M and height N of the image plane, use the CvMat function of OPENCV to create a matrix with a width of M and a height of N, and then assign values to the MXN size one by one. Starting from the origin of the image plane coordinate system, with the pixel size as the step size, search for the scanning point closest to the grid point row by row and column by column, and use the reflection intensity value of the point as the grayscale value of the grid point, and assign the grayscale value to the matrix member of the corresponding row and column; This application uses a weighted balance filtering method to solve the grid pixel assignment problem. A weighted balance filtering template is constructed, and the original pixel value is replaced by the average value in the template. If there are multiple scanning points in the grid, the average value of the reflection intensity values of all scanning points is taken as the grayscale value of the pixel. If there are no scanning points in the grid, the average value of several adjacent pixels within a certain range is taken as the grayscale value of the pixel. The image grayscale value at the pixel point (x, y) is: g(x,y)=∑(f(x,y)*p) / N Equation 5 Where N is the total number of scanning points or adjacent pixels; Finally, after calculating the pixel value of each grid point on the image plane, traverse and assign the value to the Mat matrix member at the corresponding position, then write the image data into the image file and compress the image size as needed; The image matrix that has been assigned is used to directly generate an image, obtain the image information of the shield tunnel lining surface, and mark the tunnel mileage value above and below the image.

5. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 1 is characterized in that: Stretched image processing platform: Dynamically schedules data based on a quadtree micro-level model to process stretched images. The micro-level adjusts the subdivision level as needed to reduce the amount of data required for rendering. Pre-generated content at different levels of subdivision is saved and indexed in advance. 1) Establishing a micro-level model: By combining micro-level data control strategies with internal and external memory scheduling, real-time interactive browsing of shield tunnel lining surface images is achieved. This allows the computer to selectively ignore some subdivisions during image visualization, depending on the image's distance from the viewpoint. The core of the micro-level model is the establishment of a micro-level model. This is achieved by simplifying the original high-resolution image at different subdivision levels and arranging them at different levels according to the resolution, forming a top-down model. Generate an image of the shield tunnel lining surface extension at its original resolution. Due to the long and narrow shape of the original image, the image is subdivided into 256 parts by length. This is used as the bottommost and most subdivided level in the micro-level model. Then, every four images are merged, resampled, or compressed to obtain an image of a lower resolution. This is used as the next subdivision level. This step is repeated to generate images at different depth levels, thus obtaining different subdivision levels of the micro-level model. 2) Indexing: The entire image is divided into images of different subdivision levels. The subdivision levels of low-level images are coarser, and the subdivision levels of high-level images are richer. The number of images in each layer is 4 to the power of n. Then, the image is processed using the aforementioned quadtree encoding. The entire tunnel image is used as the root node. The root node is continuously divided from top to bottom. Each non-leaf node has four child nodes. In other words, the image of the previous layer is divided into four equal parts to form the image of the next layer. The image area of the parent node is the same as the image area of the four child nodes. The image structure index is established using a quadtree. id is the node code, which serves as the index number that uniquely identifies the node. The structure also records the parent node and four child nodes of the node. Length is the number of layers the node is located at. 3) Fast rendering of stretched images: Implementing dynamic micro-level image scheduling based on OSG. First, define the image reading method. Based on the image file format, construct osgImage images of the same size and traverse and assign values to them according to the pixel storage format. Then, based on the image size and subdivision level, define the coordinates of its outer bounding rectangle and create a texture for it. After completing the above steps, use OSG to create a micro-level model. According to the quadtree index, traverse all leaf nodes, load images of different resolutions into different levels of the micro-level model, and build a top-down micro-level model. Then set the center position and display distance range of each image in the micro-level model, and OSG automatically loads the image; After the above settings are completed, when the computer performs real-time rendering, the distance is calculated based on the current viewpoint position, and the images of the corresponding subdivision level in the micro-level model are automatically loaded. The high-resolution images are displayed when they are closer to the viewpoint, and the low-resolution images are displayed when they are farther away from the viewpoint. Only the image part within the window is displayed, and the images outside the window are automatically unloaded. Micro-level models are used to interactively browse the generated shield tunnel lining surface images. When the distance from the viewpoint is far, the micro-level model automatically uses a coarse image. When the image is zoomed in, a fine image is loaded. At the same time, the micro-level model preloads images near the window to improve rendering efficiency when zooming, and automatically unloads the parts outside the window to reduce unnecessary rendering subdivisions.

6. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 1 is characterized in that: Extract the segment boundary line: Get the location of the segment boundary line, compare the segment point clouds on both sides of the boundary line, establish a mapping relationship between the two coordinate spaces, and form a peak at a point in the other coordinate system. This converts the shape detection problem into a peak calculation problem. The three curves intersect at one point on the polar coordinate plane, indicating that a straight line passes through the three points (x0, y0), (x1, y1), and (x2, y2) at the same time. The θ-ρ polar coordinate parameter value of this point is the polar diameter and polar angle of this line. For all points on a straight line, the curves drawn on the θ-ρ plane of the polar coordinate system intersect at one point. The straight line is detected by counting the number of curves that intersect at a point on the plane θ-ρ. By converting each point on the image into polar coordinates and tracing the intersection points between the corresponding curves, the number of curves that intersect at a point is obtained. The greater the number, the more points on the straight line. At the same time, a critical value is pre-set to define the detection of a straight line. Only when the number of curves that intersect at a point exceeds the critical value, a straight line is detected. The parameter value of this intersection (θ, ρ θ ) are the polar diameter and polar angle of the line.

7. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 6 is characterized in that: Edge image high-frequency filtering: Before extracting the characteristic shape of the image, the image is first subjected to edge high-frequency filtering: first, using Gaussian filtering to smooth the image, second, finding the intensity gradient of the image, third, non-maximum suppression, fourth, using double critical value for detection, and fifth, hysteresis boundary tracking; The pixels of the confirmed strong edge are regarded as the true edge points, while the weak edge points still need to be judged and tracked. If there is a strong edge point in the 8-connected area of the weak edge point, the weak edge point is determined to be the true edge.

8. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 6 is characterized in that: Segment boundary line detection: All pixels on the image are converted into a polar coordinate system. A straight line passing through each point is plotted as a sine curve on the θ-ρ plane. The number of curves intersecting at a point is counted. Points with a number exceeding a critical value are considered to be a straight line. Finally, all detected straight lines are plotted on the original image. Write a line transformation function to detect straight lines on the image. The input value of the function is the edge binary image. The distance accuracy of the line search is determined. The default setting is 1 pixel, that is, all pixels on the image are calculated. Then the angle accuracy of the line search is determined. The default setting is 1°, that is, each point is calculated 360 times during the line detection. When the line detection is not accurate enough, the angle accuracy is improved to solve it. Finally, the pixel coordinates of each point are used to calculate the corresponding ρ value. For each pair of parameter values (θ, ρ θ ) to perform calculations; The critical value parameters are pre-set in the online transformation function, and only the parameter values (θ, ρ θ ) is greater than the critical value, it can pass the detection and be judged as a straight line. At the same time, the minimum length of the detected line segment is also set. If the detected line segment length is lower than this value, it will not be displayed. The function is modified to only extract lines with polar angles around 0° to reduce horizontal interference. The arch part of the tunnel is intercepted for line extraction. Two lines that are very close to each other will be considered as the same line. Finally, the position of each line is output and converted into the corresponding mileage value.

9. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 1 is characterized in that: Extract cross-section information: 1) Section point extraction: By using the obtained segment boundary line mileage, the section information at the corresponding location is extracted. The core of the section point extraction process is to directly extract a series of points with specific mileage values from the scan file; The steps for extracting cross-section point clouds based on a specific mileage are as follows: Step 1: Based on the given mileage value S, obtain the time t when the laser scanning railcar arrives at the mileage S; Step 2: Extract a series of points whose scanning time is closest to t in the scan file, which are the cross-section points to be extracted; In actual operation, the segment sections on both sides of the extracted dividing line are compared. A 3cm thick cross-section point cloud is extracted from both ends of a given mileage value and then projected onto the same plane for comparison. 2) Cross-section point fitting denoising: Remove noise points and debris points in the cross-section information: Step 1: The extracted cross-section information N contains normal points, noise points, and debris points, and the model M of the normal points is obtained in advance. The cross-section model in the shield tunnel is a standard circle; Step 2: Randomly extract the minimum subset n that can generate the model M (circular section) from the cross-section point cloud, and fit a circular section m with the subset n; Step 3: Set a critical value e, and determine whether all points in N can meet the critical value e for the circular cross section m. If the point can meet the critical value e, then add it to n; Step 4: When all points in N have completed the third step, the points in n can be identified as normal points describing the model M, and the remaining points are noise points and debris points; Randomly select several points in the cross-section point cloud to fit a circle, then calculate the distance from each scanning point on the cross section to the fitted circle, retain the points with a distance less than 0.05 meters as normal points on the cross section, repeat the above steps, if the number of normal points accounts for 60% of the total, then the fitted circle is considered reasonable.

10. The laser scanning rail transit vehicle subway tunnel segment misalignment analysis method according to claim 1, characterized in that: Segment misalignment data and analysis: After fitting and denoising the cross-sectional point clouds on both sides of the segment dividing line, the next step is to calculate the misalignment of the cross-sectional points. Since the intercepted cross-sectional information is of a certain thickness on both sides of the dividing line, the two parts of the point cloud must first be projected onto the same plane for comparison. The misalignment value is the distance between the two cross-sectional points. The point cloud information is projected onto the same vertical plane of the tunnel. The scanned data of the point cloud is extended with the forward direction as the third-dimensional coordinate. The coordinates of the two parts of the point cloud in this direction are set to 0, and then projected onto the same vertical plane of the tunnel. After projecting the two cross-section point clouds onto the same plane, they are represented by different colors. After visualization, the places with obvious differences are the places where the misalignment occurs. One of the cross sections is used as the reference surface and the other as the comparison surface. The center position of the reference surface is calculated. Then, the angle of all cross-section points relative to the center of the circle is obtained by calculating the inverse tangent value. The points with the same angle on the two cross sections are extracted, and the distance between them is the misalignment value. The misalignment value on the cross-section circumference is calculated with a step size of only 1°. Several segment rings scanned in the interval are selected, and the misalignment values between the segment rings are calculated. The size, number and proportion of the misalignment values are calculated to generate an analysis table of the misalignment situation of the tunnel segments.

Images