Shield segment slab staggering detection method and system fusing visual and geometric features

Abstract

The invention discloses a shield segment dislocation detection method and system fusing visual and geometric features, and the method comprises the steps: employing a mobile track three-dimensional laser scanning system, and rapidly obtaining the three-dimensional point cloud data of a tunnel segment; projecting the three-dimensional point cloud data of the tunnel segment to a two-dimensional image according to the scanning parameters and the image preset resolution, and mapping the laser point reflection intensity into a pixel gray value; shield segment inter-ring joints and shield segment in-ring joints are extracted respectively, and segment joint positioning is completed; and according to the seam positioning information, extracting local point clouds at the two sides of the seam, respectively carrying out circular model fitting, calculating the height difference of circular models at the two sides at the seam, and obtaining an in-ring slab staggering value. According to the method, the visual information of the point cloud reflection intensity and the geometric information of the spatial position are fused, rapid and accurate shield segment in-ring slab staggering detection is realized, and the efficiency and the accuracy are high.

Description

Technical Field

[0001] The present invention relates to the field of shield tunnel segment misalignment detection for rail transit, and in particular to a shield tunnel segment misalignment detection method and system integrating visual and geometric features. Background Art

[0002] With the acceleration of urbanization and the deepening development of underground space, rail transit has become a core component of modern urban transportation systems. By 2023, 55 cities had opened urban rail transit lines, with a total operating mileage exceeding 10,000 kilometers, of which shield tunnels accounted for over 75%. Shield tunneling, with its strong adaptability to complex geological conditions, high construction efficiency, and minimal surface disturbance, has become the preferred construction technology for major projects such as subway tunnels and cross-river tunnels. However, the quality of shield tunnel segment assembly is directly related to the long-term service performance of the tunnel structure. Statistics show that approximately 68% of operational shield tunnels in China have varying degrees of segment misalignment, of which 12.3% have misalignment exceeding 10mm. Segment misalignment not only causes direct hazards such as stress concentration and the formation of water leakage channels, but can also trigger chain reactions such as bolted connection failure and structural stiffness degradation. Segment misalignment is the most common apparent defect in shield tunnels. Under normal circumstances, when the misalignment exceeds the design allowable value (usually 5-8mm), it will significantly change the load transfer path between segment rings, inducing secondary disasters such as concrete spalling and bolt shear failure. Studying the spatial distribution characteristics of misalignment can reflect the accumulation pattern of assembly errors during construction and provide reverse feedback for shield posture correction. Long-term monitoring data generated through regular inspections can establish a correlation model between misalignment evolution and tunnel convergence deformation, supporting the assessment of the remaining life of the structure. In addition, digital misalignment detection results can also promote the construction of BIM operation and maintenance platforms and assist in the management of smart subway infrastructure. Therefore, segment misalignment detection has significant engineering application value in tunnel structural safety, operation and maintenance decision-making, life prediction, and intelligent construction.

[0003] Currently, mainstream inspection methods in the industry can be categorized as manual, semi-automated, and fixed laser scanning. All of these methods present significant technical bottlenecks. The manual inspection method requires climbing to the segment joints using a rail-mounted ladder vehicle to perform single-point measurements with a steel ruler. One inspection section is selected every ten rings, and a misalignment distribution diagram is drawn. This method suffers from low inspection efficiency (a single section takes >30 minutes), insufficient data density (sampling rate <0.1%), significant subjective error (visual interpretation error reaches ±2mm), and high safety risks (requiring service interruption and deployment of a rail-mounted ladder vehicle). The semi-automated inspection method utilizes an industrial camera and a feature matching algorithm to analyze the displacement between adjacent ring images and convert them into segment misalignment. However, image matching is significantly affected by lighting conditions (uneven illumination within the tunnel leads to a mismatch rate >15%) and data processing relies on manual intervention (requiring manual annotation of joint locations). The fixed laser scanning method uses a frame-mounted 3D laser scanner (such as Faro Focus, Z+F 5016, etc.) to obtain a point cloud around the entire tunnel. After constructing a complete tunnel model by stitching multiple stations together, the misalignment is calculated based on a geometric feature extraction algorithm. However, the method has drawbacks such as low field efficiency (200 measuring stations are required for a 1km tunnel), accumulated point cloud registration errors (stitching error >10mm per kilometer), and insufficient algorithm robustness (the existing ICP algorithm is sensitive to local deformation, with an error rate of 20%).

[0004] Prior art patents 202011623502.1 and 202311672566.4 disclose a high-precision device for measuring lining segment misalignment. This device uses a fixed base at the misalignment site and a guide wheel and track wheel to push an internal measuring ruler along the segment. The extended length of the ruler is then read as the misalignment. This method suffers from low detection efficiency, high installation risk, and the risk of omissions. Furthermore, the quality of detection relies on the skill and experience of the engineers, making it unsuitable for large-scale batch testing.

[0005] Patent 202311676701.2 utilizes a tunnel inspection vehicle to acquire high-precision point cloud data and high-definition image data from the tunnel lining surface. It then employs a deep learning network to automatically identify and extract longitudinal seams in shield tunnels, segmenting the partial point clouds on either side of the seams. This method uses polar coordinate transformation to restore the tunnel outline and calculate the misalignment value. When applied to shield tunnels with varying geological conditions, segment materials, and construction techniques, this method suffers from high data annotation and training costs, insufficient generalization capabilities, and significant interference from data sparsity and inhomogeneity, resulting in unstable detection results.

[0006] In an existing paper, Hong Chengyu et al. from Shenzhen University deployed displacement transmission plates at the locations of segment misalignment and used the resistance change data from flex sensors sliding between the plates to calculate the change in segment misalignment. Patent 202010099364.5 determines the amount of segment misalignment by analyzing the distance measurement curve measured by a millimeter-wave radar ranging sensor. Both of these methods involve contact measurement, and the placement of sensors may be affected by environmental factors such as electromagnetic interference and humidity, which can affect measurement accuracy. This may prevent comprehensive and accurate detection of complex misalignment situations.

[0007] Hao Cui et al. proposed a point cloud segmentation workflow for subway tunnels, identifying bolts as markers for misalignment measurement. They introduced a method for locating and matching bolt areas of interest (AOIs), and calculated misalignment values ​​by fitting a reference surface to the AOIs. However, this method relies heavily on features such as bolts, and damaged or missing bolts can affect detection results.

[0008] German researchers J. Luhmann et al. used cameras to capture images of tunnel segments within a shield tunnel, acquiring multiple images from different angles. Based on photogrammetry principles, they used image matching and feature extraction techniques to recover 3D information about the scene and target, constructing a 3D model of the segment and analyzing its misalignment. This method, however, requires significant computational effort during image analysis and 3D reconstruction, resulting in low detection efficiency. In cases where surface features are unclear or contaminated, this can lead to inaccurate feature extraction, affecting measurement results.

[0009] British scholar TK Gaydecki and his colleagues placed ultrasonic sensors on or within the segments. By analyzing the ultrasonic signal's amplitude, phase, and propagation time, they determined the location and approximate amount of segment misalignment. This method requires high uniformity in the segment material; significant nonuniformity can interfere with the ultrasonic signal and lead to misjudgment. For even small misalignments, the signal characteristics are not significantly altered, making accurate measurement difficult.

[0010] Therefore, studying a non-contact, efficient and accurate method for positioning and detecting misalignment within the shield tunnel segment ring has important theoretical significance and practical research value for meeting the safety transportation needs of the rapidly developing rail transit tunnels. Summary of the Invention

[0011] Therefore, the purpose of the present invention is to provide a shield segment misalignment detection method and system that integrates visual and geometric features to solve the problems of inaccurate detection and inability to perform batch detection in the prior art.

[0012] The rail transit tunnel and lining segment point cloud data used in the following examples are as follows: Figure 2 As shown in the figure, the data is obtained by the mobile track 3D laser scanning system. Figure 3 This section of rail transit shield tunnel is 50m long, with a designed inner radius of 1.75m. A single shield ring is assembled using 6 prefabricated lining segments: 1 capping segment, 2 adjacent segments, and 3 standard segments.

[0013] In order to achieve the above-mentioned object, the present invention provides a shield segment misalignment detection method integrating visual and geometric features, comprising the following steps: S1, uses a mobile track 3D laser scanning system to quickly obtain 3D point cloud data of tunnel segments; S2, based on the scanning parameters and the preset image resolution, projects the 3D point cloud data of the tunnel segment into a 2D image and maps the laser point reflection intensity into pixel grayscale values; S3, extract the inter-ring joints and intra-ring joints of shield segments respectively to complete the segment joint positioning; S4, remove noise from the cross-section point cloud; S5, based on the joint positioning information, extract the local point clouds on both sides of the joint, perform circular model fitting on each side, calculate the height difference between the circular models on both sides at the joint, and obtain the misalignment value within the ring.

[0014] Further preferably, in S2, projecting the three-dimensional point cloud data of the tunnel segment into a two-dimensional image and mapping the laser point reflection intensity into a pixel grayscale value comprises the following steps: S201, constructing a cylindrical projection surface, presetting the image resolution, and calculating the image pixel coordinates using the following formula based on the laser point incident angle; Among them, X Pixel : X coordinate of laser point Pi in the grayscale image; Y Pixel : The Y coordinate of the laser point Pi in the grayscale image; i: The difference between the survey line where the laser point Pi is located and the starting survey line; v: The carrier running speed, in m / s; f: The scanner line frequency, in Hz; H: The preset horizontal resolution of the image; V: The set vertical resolution; θ: 1 / 2 of the angle range value that the cross-section point cloud needs to be mapped to the image; R: The radius of the projection cylinder, which is the designed inner radius of the tunnel; α represents the incident angle of the laser point Pi with the tunnel center as the origin.

[0015] S202, mapping the intensity of the laser reflection point to a grayscale value, and normalizing the grayscale value to the same scale; S203, assigning an average value of the grayscale values ​​of all laser points contained in each pixel as the pixel value, and performing equalization processing on the image.

[0016] Further preferably, in S202, the intensity of the laser reflection point is mapped to a grayscale value according to the following formula (2): (2) Where: Pi g : Gray value of laser point Pi; Pi intensity : Reflection intensity value of laser point Pi; Pi intensity，min : Minimum reflection intensity of laser point Pi; Pi intensity，max : Maximum reflection intensity of laser point Pi.

[0017] Further preferably, in S3, when extracting the inter-ring seams of shield segments, the following steps are included: Based on the two-dimensional image obtained in S2, the Canny operator is used to identify the edge of the joint between the segments; The Hough transform method is used to detect and cluster the adjacent edge points with similar parameters among the identified edges into line segments.

[0018] Further preferably, in S3, when extracting the inner seam of the shield segment ring, the following steps are included: S301, using the inter-ring joint of the shield segment as a reference, visually determine the offset distance d based on the grayscale image and select the cross-section point cloud position; S302: Calculate the center point of the cross section using the robust ellipse fitting method on the selected cross section point cloud, and sort the distances from all laser points to the cross section center point by point number to form a point set; S303: Using the local maximum method to detect peaks in the point cloud, a search window K is set, and a Gaussian function is used as the peak function to detect peaks within the window K and include them in the candidate peak set. The peak height threshold h and the peak direction are set as positive as the limiting conditions, and the true peaks are screened from the candidate peak set. S304, extracting the laser point serial number corresponding to the true peak to complete the segment joint positioning.

[0019] Further preferably, in S303, the Gaussian function is expressed by the following formula (3): (3) Where: y0: wave offset baseline; A: peak area; w: peak width; x c : peak center; x: laser point number; y: laser point distance.

[0020] Further preferably, in S4, noise removal is performed on the cross-section point cloud, including: S401: Calculate the distance set from all laser points in the section to the center of the section, select the median of the distance set as the reference value, set the floating range threshold k, and use the straight-through filtering method to extract the laser points on the inner wall of the tunnel lining to complete the main noise removal; S402: After removing the main noise of the cross-section point cloud, a statistical filtering method is used to remove the remaining sparse outlier noise points to obtain the final tunnel cross-section point cloud.

[0021] Further preferably, in S5, according to the joint positioning information, the local point clouds on both sides of the joint are extracted, the circular model fitting is performed respectively, the height difference of the circular models on both sides at the joint is calculated, and the misalignment value within the ring is obtained; including S501, taking the tunnel inner diameter as the prior information to constrain the circular model fitting radius, the orthogonal distance regression method is iteratively calculated using the following formula (4) to obtain the accurate lining segment circular models C1 and C2 on both sides of the joint. The circular model parameters include the coordinates of the center of the circle (X c , Y c ) Radius R c , as shown in formula (5); (4) (5) Where: : coordinate component x i The weight of : coordinate component y i The weight of : coordinate component x i The residual of : coordinate component y i The residual of : is the fitting parameter vector, f: circle model function, : X-axis coordinate component of the laser point; : Y-axis coordinate component of the laser point; S502: Draw a straight line m between the center of the cross section and the joint, calculate the coordinates of the intersections of m with the circular models C1 and C2, and calculate the difference between the intersection coordinates, which is the misalignment value within the ring.

[0022] The present invention also provides a shield segment misalignment detection system that integrates visual and geometric features, which is used to implement the steps of the shield segment misalignment detection method that integrates visual and geometric features, including: a rail trolley that integrates a cross-sectional laser scanner and multiple types of data collectors, an image conversion module, a joint positioning module, a filtering module, and a fitting calculation module; The track trolley uses a mobile track 3D laser scanning system to quickly acquire 3D point cloud data of the tunnel segments; The image conversion module projects the three-dimensional point cloud data of the tunnel segment into a two-dimensional image according to the scanning parameters and the preset image resolution, and maps the laser point reflection intensity into pixel grayscale values; The joint positioning module extracts the inter-ring joints and intra-ring joints of shield segments respectively to complete the segment joint positioning; The filtering module removes noise from the cross-section point cloud; The fitting calculation module extracts local point clouds on both sides of the seam according to the seam positioning information, performs circular model fitting on each side, calculates the height difference between the circular models on both sides at the seam, and obtains the misalignment value within the ring.

[0023] The shield segment misalignment detection method and system disclosed in this application integrates visual and geometric features. It has at least the following beneficial effects: 1. This method comprehensively utilizes the reflectivity and spatial position information of point clouds, integrating visual and geometric features to quickly and accurately obtain information on the misalignment within the shield tunnel lining segments. Using this method, a 1km tunnel can be inspected with automated data processing in just 15 minutes. 2. This invention uses a peak detection algorithm with a Gaussian peak function sliding window to rapidly locate the joints within the shield segment ring, avoiding interference from pipelines and other equipment attached to the inner wall of the tunnel lining, and improving the efficiency and accuracy of the misalignment detection algorithm. 3. This invention uses an algorithm that combines median-pass filtering with statistical filtering. Through a "main denoising + fine denoising" strategy, it achieves rapid noise removal from tunnel cross-section point clouds, improving the accuracy of misalignment detection. 4. This invention uses the tunnel inner diameter as prior information to constrain the lining segment circular model, suppressing the problems of result divergence and excessive number of iterations encountered when fitting local information to the overall model, thereby improving the accuracy and efficiency of the detection algorithm.

[0024] 5. Compared with existing detection methods that only use image visual information or three-dimensional spatial information, this invention combines the advantages of image visual information and geometric features, is more efficient and accurate, and is suitable for practical engineering applications. It helps to improve the technical level of mobile three-dimensional laser scanning technology in the field of rail transit tunnel segment misalignment detection, and serves operation and maintenance management departments in carrying out repair and maintenance work. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The figure is a flow chart of the shield segment misalignment detection method that integrates visual and geometric features according to the present invention.

[0026] Figure 2 Point cloud data for rail transit tunnels and lining segments.

[0027] Figure 3 This is a diagram of the composition of the mobile track 3D laser scanning system.

[0028] Figure 4 This is the point cloud depression feature map at the joint of the shield segment.

[0029] Figure 5This is the point cloud peak feature map at the joint of the shield segment.

[0030] Figure 6 This is the denoising effect diagram of the median method straight-through filtering.

[0031] Figure 7 Schematic diagram of sparse outlier noise points.

[0032] Figure 8 Schematic diagram of calculation of misalignment within a circular model fitting ring.

[0033] Figure 9 This is a statistical chart comparing the accuracy of the in-ring misalignment detection results with the manual method. DETAILED DESCRIPTION

[0034] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] like Figure 1 As shown, an embodiment of the present invention provides a shield segment misalignment detection method that integrates visual and geometric features, including the following steps: S1. Use a mobile track 3D laser scanning system to quickly acquire 3D point cloud data of the tunnel segments. In this embodiment, the mobile track 3D laser scanning system has the following features.

[0036] The scanning system integrates cross-sectional laser scanner, inertial navigation unit, mileage encoder and other data acquisition sensors on the PLC carrier vehicle, which constitutes the main structure of the vehicle-mounted mobile scanning system. Figure 3 As shown in the figure, the surveyor uses a handheld device to send instructions. The electric drive system propels the machine forward at a constant speed along the track while the scanner emits a spiral laser line perpendicular to the tunnel, generating a uniform and dense 3D point cloud. Multi-source data fusion and solution yields a 3D laser point cloud with precise mileage and geometry.

[0037] The cross-sectional laser scanner operates in line scanning mode. The scanning prism rotates only in the vertical plane at a rate of 100 lines per second. Each line contains 10,900 measurement points, forming a complete cross section. The measurement point spacing at a range of 5 meters is 3 mm, with a measurement point accuracy of 0.2 mm.

[0038] S2: Based on the scanning parameters and the preset image resolution, the 3D point cloud data of the tunnel segment is projected onto a 2D image, and the laser point reflection intensity is mapped to pixel grayscale values. The scanning parameters are the line survey frequency f and the vehicle travel speed v. In this embodiment, the line survey frequency f is 100 Hz, and the vehicle travel speed v is 0.56 m / s.

[0039] In S2, projecting the three-dimensional point cloud data of the tunnel segment into a two-dimensional image and mapping the laser point reflection intensity into pixel grayscale values ​​includes the following steps: S201, constructing a cylindrical projection surface, presetting the image resolution, and calculating the image pixel coordinates using the following formula based on the laser point incident angle; Among them, X Pixel : X coordinate of laser point Pi in the grayscale image; Y Pixel : The Y coordinate of the laser point Pi in the grayscale image; i: The difference between the survey line where the laser point Pi is located and the starting survey line; v: The carrier running speed, in m / s; f: The scanner line frequency, in Hz; H: The preset horizontal resolution of the image; V: The set vertical resolution; θ: 1 / 2 of the angle range value that the cross-section point cloud needs to be mapped to the image; R: The radius of the projection cylinder, which is the designed inner radius of the tunnel; α represents the incident angle of the laser point Pi with the tunnel center as the origin.

[0040] In this example, a Z+F 9012 laser scanner was used in line scan mode for continuous vertical measurement of the tunnel's inner wall. Within the same measurement line section, the zenith angle of incidence was set to 0, while the nadir angles were set to -PI and PI. The scanner's back was facing forward, representing the positive boresight direction. The incident angle for the left laser point was negative, while that for the right was positive. The preset horizontal and vertical resolutions for the image were 12 mm and 9 mm, respectively.

[0041] S202, mapping the intensity of the laser reflection point to a grayscale value, and normalizing the grayscale value to the same scale; further preferably, in S202, mapping the intensity of the laser reflection point to a grayscale value according to the following formula (2): (2) Where: Pi g : Gray value of laser point Pi; Pi intensity : Reflection intensity value of laser point Pi; Pi intensity，min：激光点Pi的反射强度最小值；Piintensity，max : Maximum reflection intensity of laser point Pi.

[0042] In this embodiment, the laser point reflection intensity value obtained by the Z+F 9012 laser scanner is in the range of 0 to 65535. After being processed by formula (2), the reflection intensity value is converted into a grayscale value in the range of 0 to 255.

[0043] S203: The average grayscale value of all laser points within each pixel is assigned as the pixel value, and the image is equalized. In this embodiment, each pixel contains approximately 24 laser points. After taking the average grayscale value of all laser points as the pixel value, a histogram equalization algorithm is used to enhance contrast. The pixels where the lining segment joints are located appear black, making them clearly distinguishable from the gray lining segment pixels.

[0044] S3, respectively extracting the inter-ring joints and intra-ring joints of the shield segments to complete the segment joint positioning; including using the Canny operator to identify the edges of the inter-ring joints of the segments based on the two-dimensional image obtained in S2; In the two-dimensional grayscale image generated by this embodiment, the inter-ring joints of the shield are distributed vertically, which is obviously different from the horizontally distributed pipelines. The "Canny operator + Hough transform" method can be used to obtain the inter-ring joint pixels. The joint pixel number can be converted to the scanning line number corresponding to the joint to achieve scanning line positioning.

[0045] The Hough transform method is used to detect and cluster the adjacent edge points with similar parameters among the identified edges into line segments.

[0046] In S3, extracting the inner joint of the shield segment ring includes the following steps: S301, using the inter-ring joint of the shield segment as a reference, visually determine the offset distance d based on the grayscale image and select the cross-section point cloud position; S302: Calculate the center point of the cross section using the robust ellipse fitting method on the selected cross section point cloud, and sort the distances from all laser points to the cross section center point by point number to form a point set; Segment misalignment must be detected ring by ring, necessitating the extraction of a cross-sectional point cloud within each shield ring for calculation. Preferably, in this embodiment, the inter-ring seams are selected as the offset reference, and the cross-sectional positions are uniformly offset 10 cm in the forward direction to avoid the adverse effects of structures such as handholes on the lining segments on the inspection results. In this embodiment, the extracted cross-sections are calculated using a robust ellipse fitting method to determine the cross-sectional center point. The distances from all laser points within the cross-section to the center point are calculated and stored in the data set Vec in order of point number.

[0047] S303: Using the local maximum method to detect peaks in the point cloud, a search window K is set, and a Gaussian function is used as the peak function to detect peaks within the window K and include them in the candidate peak set. The peak height threshold h and the peak direction are set as positive as the limiting conditions, and the true peaks are screened from the candidate peak set. Further preferably, in S303, the Gaussian function is expressed by the following formula (3): (3) Where: y0: wave offset baseline; A: peak area; w: peak width; x c : peak center; x: laser point number; y: laser point distance.

[0048] S304, extracting the laser point serial number corresponding to the true peak to complete the segment joint positioning.

[0049] like Figure 4 、 Figure 5 As shown, the geometric characteristics of the lining segment joints are similar to those of a Gaussian function, so a Gaussian function is used as the peak function. Preferably, in this embodiment, when the search range K is set to 200 points and the height percentage threshold h is set to 2%, a stable and accurate detection effect is achieved.

[0050] S4, remove noise from the cross-section point cloud; Further preferably, in S4, noise removal is performed on the cross-section point cloud, including: S401, calculate the distance set from all laser points in the section to the center of the section, select the median of the distance set as the reference value, set the floating range threshold k, use the straight-through filtering method to extract the laser points on the inner wall of the tunnel lining, and complete the main noise removal; a single denoising method will increase the possibility of over-denoising and under-denoising, and the effect is not ideal, so a combination of multiple methods is used for step-by-step denoising. The proportion of noise points in the entire section is much lower than that of the inner wall points of the tunnel lining, so in the set Vec, the median must be located on the inner wall of the tunnel lining. Preferably, in this embodiment, taking the median as the reference value and setting the floating range of 3cm can effectively extract the inner wall points of the tunnel lining, thereby completing the main noise removal, such as Figure 6 shown.

[0051] S402, after removing the main noise of the cross-section point cloud, the statistical filtering method is used to remove the remaining sparse outlier noise points to obtain the final tunnel cross-section point cloud. The main part of the noise points in the cross section has been removed, and only a small part of the sparse outlier noise points remains in the cross section, such as Figure 7 In this embodiment, a statistical filtering method is used to filter out sparse outlier noise points to obtain "clean" tunnel lining inner wall points.

[0052] S5, based on the joint positioning information, extract the local point clouds on both sides of the joint, perform circular model fitting on each side, calculate the height difference between the circular models on both sides at the joint, and obtain the misalignment value within the ring.

[0053] In S5, according to the joint positioning information, the local point clouds on both sides of the joint are extracted, and the circular models are fitted respectively. The height difference of the circular models on both sides at the joint is calculated to obtain the misalignment value within the ring; including S501, taking the tunnel inner diameter as the prior information to constrain the circular model fitting radius, the orthogonal distance regression method is iteratively calculated using the following formula (4) to obtain the accurate lining segment circular models C1 and C2 on both sides of the joint. The circular model parameters include the coordinates of the center of the circle (X c , Y c ) Radius R c , as shown in formula (5); (4) (5) Where: : coordinate component x i The weight of : coordinate component y i The weight of : coordinate component x i The residual of : coordinate component y i The residual of : is the fitting parameter vector, f: circle model function, : X-axis coordinate component of the laser point; : Y-axis coordinate component of the laser point; S502: Draw a straight line m between the center of the cross section and the joint, calculate the coordinates of the intersections of m with the circular models C1 and C2, and calculate the difference between the intersection coordinates, which is the misalignment value within the ring.

[0054] In S501, the block circle model is fitted: the small areas on both sides of the inner ring joint are basically "clean" inner wall points of the pipe segment and have high continuity, which can effectively reduce the model fitting residual. Preferably, in this embodiment, the area within 30 cm on both sides of the inner ring joint is selected for circle model fitting. In order to overcome the problem of excessive model deviation caused by overfitting when a small number of local points are fitted to the whole, this embodiment uses the tunnel radius of 1.75m as the prior information to constrain the orthogonal distance regression formula (4), and sets the data-model correlation coefficient to exceed 0.95 as the judgment condition for stopping the algorithm iteration.

[0055] In S502, the misalignment value is calculated: a straight line is constructed between the laser point at the segment joint and the center point of the cross section, and the straight line intersects with the circular models on both sides. The distance between the two intersection points is calculated to obtain the misalignment value at the joint position, such as Figure 8 As shown. Calculate the misalignment values ​​at each segment joint within the shield ring, completing the circumferential misalignment test for the entire inspection area. Determine the misalignment values ​​according to relevant specifications and promptly report any misalignment values ​​exceeding the limit to the operation and maintenance department for processing.

[0056] By using the method of the present invention, it only takes 15 minutes to automatically process the data for inspecting a tunnel with a length of 1 km, thus greatly improving the inspection efficiency.

[0057] Experimental precision analysis: The traditional manual inspection method of measuring the misalignment of lining segments with a steel ruler is the main method for detecting misalignment in rail transit tunnels. The measurement accuracy of this method is 2mm and it has been widely used and recognized. The detection results of the method of the present invention are compared with the results of the traditional manual steel ruler detection. The maximum error is 1.17mm and the error in the detection is 0.34mm. The comparison results are as follows: Figure 9 The detection accuracy of the method of the present invention is comparable to that of traditional manual methods and meets the requirements of the current "Code for Construction and Acceptance of Shield Tunneling" (GB 50446-2017). The present invention also provides a shield segment misalignment detection system that integrates visual and geometric features, which is used to implement the steps of the shield segment misalignment detection method that integrates visual and geometric features, including: a rail trolley that integrates a cross-sectional laser scanner and multiple types of data collectors, an image conversion module, a joint positioning module, a filtering module, and a fitting calculation module; The track trolley uses a mobile track 3D laser scanning system to quickly acquire 3D point cloud data of the tunnel segments; The image conversion module projects the three-dimensional point cloud data of the tunnel segment into a two-dimensional image according to the scanning parameters and the preset image resolution, and maps the laser point reflection intensity into pixel grayscale values; The joint positioning module extracts the inter-ring joints and intra-ring joints of shield segments respectively to complete the segment joint positioning; The filtering module removes noise from the cross-section point cloud; The fitting calculation module extracts local point clouds on both sides of the seam according to the seam positioning information, performs circular model fitting on each side, calculates the height difference between the circular models on both sides at the seam, and obtains the misalignment value within the ring.

[0058] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A shield segment misalignment detection method integrating visual and geometric features, characterized in that: The following steps are involved: S1, uses a mobile track 3D laser scanning system to quickly obtain 3D point cloud data of tunnel segments; S2, based on the scanning parameters and the preset image resolution, projects the 3D point cloud data of the tunnel segment into a 2D image and maps the laser point reflection intensity into pixel grayscale values; S3, extract the inter-ring joints and intra-ring joints of shield segments respectively to complete the segment joint positioning; S4, remove noise from the cross-section point cloud; S5, based on the joint positioning information, extract the local point clouds on both sides of the joint, perform circular model fitting on each side, calculate the height difference between the circular models on both sides at the joint, and obtain the misalignment value within the ring.

2. The shield segment misalignment detection method integrating visual and geometric features according to claim 1 is characterized in that: In S2, projecting the three-dimensional point cloud data of the tunnel segment into a two-dimensional image and mapping the laser point reflection intensity into pixel grayscale values ​​includes the following steps: S201, constructing a cylindrical projection surface, presetting the image resolution, and calculating the image pixel coordinates using the following formula based on the laser point incident angle; Among them, X Pixel : X coordinate of laser point Pi in the grayscale image; Y Pixel : Y coordinate of laser point Pi in the grayscale image; i: the difference between the survey line where laser point Pi is located and the starting survey line; v: carrier running speed, unit: m / s; f: scanner line measurement frequency, unit: Hz; H: preset image horizontal resolution; V: set vertical resolution; θ: 1 / 2 of the angle range value that the cross-section point cloud needs to be mapped to the image; R: projection cylinder radius, which is the tunnel design inner radius value; α represents the incident angle of laser point Pi with the tunnel center as the origin; S202, mapping the intensity of the laser reflection point to a grayscale value, and normalizing the grayscale value to the same scale; S203, assigning an average value of the grayscale values ​​of all laser points contained in each pixel as the pixel value, and performing equalization processing on the image.

3. The shield segment misalignment detection method integrating vision and geometric features according to claim 1 is characterized in that: In S202, the intensity of the laser reflection point is mapped to a grayscale value according to the following formula (2): （2） Where: Pi g : Gray value of laser point Pi; Pi intensity : Reflection intensity value of laser point Pi; Pi intensity，min : Minimum reflection intensity of laser point Pi; Pi intensity，max : Maximum reflection intensity of laser point Pi.

4. The shield segment misalignment detection method integrating vision and geometric features according to claim 1 is characterized in that: In S3, when extracting the inter-ring joints of shield segments, it includes: Based on the two-dimensional image obtained in S2, the Canny operator is used to identify the edge of the joint between the pipe segments; The Hough transform method is used to detect and cluster the adjacent edge points with similar parameters among the identified edges into line segments.

5. The shield segment misalignment detection method integrating vision and geometric features according to claim 1 is characterized in that: In S3, extracting the inner joint of the shield segment ring includes the following steps: S301, using the inter-ring joint of the shield segment as a reference, visually determine the offset distance d based on the grayscale image and select the cross-section point cloud position; S302: Calculate the center point of the cross section using the robust ellipse fitting method on the selected cross section point cloud, and sort the distances from all laser points to the cross section center point by point number to form a point set; S303: Using the local maximum method to detect peaks in the point cloud, a search window K is set, and a Gaussian function is used as the peak function to detect peaks within the window K and include them in the candidate peak set. The peak height threshold h and the peak direction are set as positive as the limiting conditions, and the true peaks are screened from the candidate peak set. S304: extract the laser point serial number corresponding to the true peak to complete the segment joint positioning.

6. The shield segment misalignment detection method integrating vision and geometric features according to claim 5 is characterized in that: In S303, the Gaussian function is expressed by the following formula (3): （3） Where: y0: wave offset baseline; A: peak area; w: peak width; x c : peak center; x: laser point number; y: laser point distance.

7. The shield segment misalignment detection method integrating vision and geometric features according to claim 1 is characterized in that: In S4, noise removal is performed on the cross-section point cloud, including: S401: Calculate the distance set from all laser points in the section to the center of the section, select the median of the distance set as the reference value, set the floating range threshold k, and use the straight-through filtering method to extract the laser points on the inner wall of the tunnel lining to complete the main noise removal; S402: After removing the main noise of the cross-section point cloud, a statistical filtering method is used to remove the remaining sparse outlier noise points to obtain the final tunnel cross-section point cloud.

8. The shield segment misalignment detection method integrating vision and geometric features according to claim 1 is characterized in that: In S5, according to the joint positioning information, the local point clouds on both sides of the joint are extracted, and the circular models are fitted respectively. The height difference of the circular models on both sides at the joint is calculated to obtain the misalignment value within the ring; including S501, taking the tunnel inner diameter as the prior information to constrain the circular model fitting radius, the orthogonal distance regression method is iteratively calculated using the following formula (4) to obtain the accurate lining segment circular models C1 and C2 on both sides of the joint. The circular model parameters include the coordinates of the center of the circle (X c , Y c ) Radius R c , as shown in formula (5); （4） （5） Where: : coordinate component x i The weight of : coordinate component y i The weight of : coordinate component x i The residual of : coordinate component y i The residual of : is the fitting parameter vector, f: circle model function, : X-axis coordinate component of the laser point; : Y-axis coordinate component of the laser point; S502: Draw a straight line m between the center of the cross section and the joint, calculate the coordinates of the intersections of m with the circular models C1 and C2, and calculate the difference between the intersection coordinates, which is the misalignment value within the ring.

9. A shield segment misalignment detection system integrating visual and geometric features, characterized in that: The steps for implementing the shield segment misalignment detection method integrating visual and geometric features as described in any one of claims 1 to 8 include: a rail trolley integrating a cross-sectional laser scanner and multiple types of data collectors, an image conversion module, a joint positioning module, a filtering module, and a fitting calculation module; The track trolley uses a mobile track 3D laser scanning system to quickly acquire 3D point cloud data of the tunnel segments; The image conversion module projects the three-dimensional point cloud data of the tunnel segment into a two-dimensional image according to the scanning parameters and the preset image resolution, and maps the laser point reflection intensity into pixel grayscale values; The joint positioning module extracts the inter-ring joints and intra-ring joints of shield segments respectively to complete the segment joint positioning; The filtering module removes noise from the cross-section point cloud; The fitting calculation module extracts local point clouds on both sides of the seam according to the seam positioning information, performs circular model fitting on each side, calculates the height difference between the circular models on both sides at the seam, and obtains the misalignment value within the ring.

Images